…and this is a 
There he goes again, making up nonsense and making ridiculous claims that have no relationship to reality. Ray Kurzweil must be able to spin out a good line of bafflegab, because he seems to have the tech media convinced that he's a genius, when he's actually just another Deepak Chopra for the computer science cognoscenti.
His latest claim is that we'll be able to reverse engineer the human brain within a decade. By reverse engineer, he means that we'll be able to write software that simulates all the functions of the human brain. He's not just speculating optimistically, though: he's building his case on such awfully bad logic that I'm surprised anyone still pays attention to that kook.
Sejnowski says he agrees with Kurzweil's assessment that about a million lines of code may be enough to simulate the human brain.
Here's how that math works, Kurzweil explains: The design of the brain is in the genome. The human genome has three billion base pairs or six billion bits, which is about 800 million bytes before compression, he says. Eliminating redundancies and applying loss-less compression, that information can be compressed into about 50 million bytes, according to Kurzweil.
About half of that is the brain, which comes down to 25 million bytes, or a million lines of code.
I'm very disappointed in Terence Sejnowski for going along with that nonsense.
See that sentence I put in red up there? That's his fundamental premise, and it is utterly false. Kurzweil knows nothing about how the brain works. It's design is not encoded in the genome: what's in the genome is a collection of molecular tools wrapped up in bits of conditional logic, the regulatory part of the genome, that makes cells responsive to interactions with a complex environment. The brain unfolds during development, by means of essential cell:cell interactions, of which we understand only a tiny fraction. The end result is a brain that is much, much more than simply the sum of the nucleotides that encode a few thousand proteins. He has to simulate all of development from his codebase in order to generate a brain simulator, and he isn't even aware of the magnitude of that problem.
We cannot derive the brain from the protein sequences underlying it; the sequences are insufficient, as well, because the nature of their expression is dependent on the environment and the history of a few hundred billion cells, each plugging along interdependently. We haven't even solved the sequence-to-protein-folding problem, which is an essential first step to executing Kurzweil's clueless algorithm. And we have absolutely no way to calculate in principle all the possible interactions and functions of a single protein with the tens of thousands of other proteins in the cell!
Let me give you a few specific examples of just how wrong Kurzweil's calculations are. Here are a few proteins that I plucked at random from the NIH database; all play a role in the human brain.
First up is RHEB (Ras Homolog Enriched in Brain). It's a small protein, only 184 amino acids, which Kurzweil pretends can be reduced to about 12 bytes of code in his simulation. Here's the short description.
MTOR (FRAP1; 601231) integrates protein translation with cellular nutrient status and growth signals through its participation in 2 biochemically and functionally distinct protein complexes, MTORC1 and MTORC2. MTORC1 is sensitive to rapamycin and signals downstream to activate protein translation, whereas MTORC2 is resistant to rapamycin and signals upstream to activate AKT (see 164730). The GTPase RHEB is a proximal activator of MTORC1 and translation initiation. It has the opposite effect on MTORC2, producing inhibition of the upstream AKT pathway (Mavrakis et al., 2008).
Got that? You can't understand RHEB until you understand how it interacts with three other proteins, and how it fits into a complex regulatory pathway. Is that trivially deducible from the structure of the protein? No. It had to be worked out operationally, by doing experiments to modulate one protein and measure what happened to others. If you read deeper into the description, you discover that the overall effect of RHEB is to modulate cell proliferation in a tightly controlled quantitative way. You aren't going to be able to simulate a whole brain until you know precisely and in complete detail exactly how this one protein works.
And it's not just the one. It's all of the proteins. Here's another: FABP7 (Fatty Acid Binding Protein 7). This one is only 132 amino acids long, so Kurzweil would compress it to 8 bytes. What does it do?
Anthony et al. (2005) identified a Cbf1 (147183)-binding site in the promoter of the mouse Blbp gene. They found that this binding site was essential for all Blbp transcription in radial glial cells during central nervous system (CNS) development. Blbp expression was also significantly reduced in the forebrains of mice lacking the Notch1 (190198) and Notch3 (600276) receptors. Anthony et al. (2005) concluded that Blbp is a CNS-specific Notch target gene and suggested that Blbp mediates some aspects of Notch signaling in radial glial cells during development.
Again, what we know of its function is experimentally determined, not calculated from the sequence. It would be wonderful to be able to take a sequence, plug it into a computer, and have it spit back a quantitative assessment of all of its interactions with other proteins, but we can't do that, and even if we could, it wouldn't answer all the questions we'd have about its function, because we'd also need to know the state of all of the proteins in the cell, and the state of all of the proteins in adjacent cells, and the state of global and local signaling proteins in the environment. It's an insanely complicated situation, and Kurzweil thinks he can reduce it to a triviality.
To simplify it so a computer science guy can get it, Kurzweil has everything completely wrong. The genome is not the program; it's the data. The program is the ontogeny of the organism, which is an emergent property of interactions between the regulatory components of the genome and the environment, which uses that data to build species-specific properties of the organism. He doesn't even comprehend the nature of the problem, and here he is pontificating on magic solutions completely free of facts and reason.
I'll make a prediction, too. We will not be able to plug a single unknown protein sequence into a computer and have it derive a complete description of all of its functions by 2020. Conceivably, we could replace this step with a complete, experimentally derived quantitative summary of all of the functions and interactions of every protein involved in brain development and function, but I guarantee you that won't happen either. And that's just the first step in building a simulation of the human brain derived from genomic data. It gets harder from there.
I'll make one more prediction. The media will not end their infatuation with this pseudo-scientific dingbat, Kurzweil, no matter how uninformed and ridiculous his claims get.
(via Mo Constandi)
I've noticed an odd thing. Criticizing Ray Kurzweil brings out swarms of defenders, very few of whom demonstrate much ability to engage in critical thinking.
If you are complaining that I've claimed it will be impossible to build a computer with all the capabilities of the human brain, or that I'm arguing for dualism, look again. The brain is a computer of sorts, and I'm in the camp that says there is no problem in principle with replicating it artificially.
What I am saying is this:
Reverse engineering the human brain has complexities that are hugely underestimated by Kurzweil, because he demonstrates little understanding of how the brain works.
His timeline is absurd. I'm a developmental neuroscientist; I have a very good idea of the immensity of what we don't understand about how the brain works. No one with any knowledge of the field is claiming that we'll understand how the brain works within 10 years. And if we don't understand all but a fraction of the functionality of the brain, that makes reverse engineering extremely difficult.
Kurzweil makes extravagant claims from an obviously extremely impoverished understanding of biology. His claim that "The design of the brain is in the genome"? That's completely wrong. That makes him a walking talking demo of the Dunning-Kruger effect.
Most of the functions of the genome, which Kurzweil himself uses as the starting point for his analysis, are not understood. I don't expect a brain simulator to slavishly imitate every protein, but you will need to understand how the molecules work if you're going to reverse engineer the whole.
If you're an acolyte of Kurzweil, you've been bamboozled. He's a kook.
By the way, this story was picked up by Slashdot and Gizmodo.
Life Science
Comments
Posted by: Rorschach
|
August 17, 2010 8:48 AM
To be fair, he once wrote lucidly on Intelligent Machines.
Posted by: Buster
|
August 17, 2010 8:51 AM
Altogether now...
Bomb, bomb, bomb...
Bomb, bomb Iran!
http://www.slate.com/id/2264064/
Iiiiiiiiiiiiit's Hitchy!
Posted by: KG
|
August 17, 2010 8:52 AM
Like you, PZ, I'm disappointed in (and amazed at) Terence Sejnowski - if he's being correctly cited. Kurzweil is, apparently, bright enough within his areas of expertise - optical character recognition, text-to-speech and speech-recognition software. Outside that, he's a kind of super-kook, who thinks that not only will he live long enough to "upload" his mind to computer hardware, but that he'll be able to resurrect his dead father using DNA recovered from the latter's grave plus records of his life. IOW, he believes in magic.
Posted by: Sastra
|
August 17, 2010 8:53 AM
Many people would call (and have called) Kurzwell's argument here an example of scientism -- the overreaching doctrine that science can understand and solve any problem, and will eventually lead to Utopia. And yet, though this term fits this situation in some ways, the criticism of the theory is itself science-based. Kurzwell's problem isn't that he's asking too much of science: his problem is that he isn't doing science correctly. He's making scientific mistakes.
When critics of science use examples such as this one to show why "scientism" is a bad thing, there's a tendency to confuse problems doing science, with problems with science itself.
Posted by: KG
|
August 17, 2010 8:57 AM
Buster,
Piss off. I loathe Hitchens' support for American imperialism (and his misogyny), but he is not the subject of this thread.
Posted by: KG
|
August 17, 2010 8:59 AM
I wonder if the True Believers in Its Imminence the Singularity will turn up to defend their prophet?
Posted by: crosbie.fitch
|
August 17, 2010 9:03 AM
There's always the expectation Kurzweil may have that scientific/technological progress happens in leaps and bounds.
If we extrapolate the future as a gradual improvement based on our contemporary experience of it, future progress appears glacial, but with unknown leaps that we do not expect, progress can be surprisingly fast.
It may be wise to expect the unexpected, albeit accompanied by a small risk of disappointment.
Posted by: james.boehmer
|
August 17, 2010 9:09 AM
His science may be wrong, but I don't think the conclusion is. It may not be 20 years, but we will eventually understand how the brain works, as well as be able to create organisms in a controlled manner. Simply because we don't understand the process now doesn't imply that it's voodoo. It just means we think Kurzweil's getting into sci-fi territory.
Posted by: John Morales
|
August 17, 2010 9:09 AM
As I understand PZ's post, expressed in computerese: the genome is not just self-modifying "code", but a self-modifying "kernel" that iteratively constructs self-modifying "hardware" to run said "software" via stochastic processes which are dependent on environmental feedback.
Or something.
</bafflegab>
Posted by: Duckbilled Platypus
|
August 17, 2010 9:12 AM
Here:
Corrected the bolding for you. Forget about biology, this is just plain stupid numerology applied to software development. Any decent programmer would bash his head on a keyboard when reading that.Posted by: Buster
|
August 17, 2010 9:16 AM
Buster,
Piss off. I loathe Hitchens' support for American imperialism (and his misogyny), but he is not the subject of this thread.
I wonder if we could have a thread about it, because...
Catholics are always asked to refudiate fellow Catholics for XYZ even if those Catholics have their disagreements already. But atheists just seem to go around saying, oh well I really like that guy except when he tries to tell the world to plunge itself in to a new war.
Atheists! Refudiate this madman!!!
Posted by: Duckbilled Platypus
|
August 17, 2010 9:16 AM
Bash *their* head (apologies for my unintended genderalisation).
Posted by: hje
|
August 17, 2010 9:21 AM
Excellent post. Of course all skeptics will be left behind in the Rapture of the Nerds ; )
Posted by: DaveH
|
August 17, 2010 9:21 AM
Even shorter analogy: I know all the materials (plastics, metal alloys, etc.) that go into a computer, therefore I can build a computer!
On a side note, anyone have any good stratagies for dealing with Kurzweil fanboys that you can't avoid talking to, like a brother who constantly sends you emails, Skype, etc. about Kurzweil and his "genius"?
Posted by: KG
|
August 17, 2010 9:22 AM
Buster,
You're an idiot troll:
1) There are recent threads on Hitchens.
2) Catholics all belong to an organization (it's known as the "Catholic Church", in case you didn't know) which they are (these days) free to leave. They can therefore reasonably be held accountable to some degree for what that organization does. While there are atheist organizations, most atheists don't belong to any of them, and even fewer belong to any organization Hitchens belongs to. See the difference?
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 9:23 AM
Sure.
What the fuck does "refudiate" mean? I hope it doesn't mean "fellate."
Posted by: commanista
|
August 17, 2010 9:24 AM
I realized how full of himself Kurzweil was a few years ago, when I saw him suggest that the human eye can only perceive about 10MP, when the real figure is actually closer to about 1200MP.
(Take the lowest end of tests on the maximum dpi people can discern (600), then do the math for the arc of human vision)
He's also suggested at times that all the perceptions and subsequent memories of an entire lifetime would come to no more than 100gb. Which is bollocks, just based on the visual information we receive each and every second.
Posted by: Sastra
|
August 17, 2010 9:26 AM
Buster #11 wrote:
I think this criticism is a refreshing change of pace from the criticism that atheists are always fighting amongst themselves instead of coming together on the things they agree with. Take your pick.
But yes, this is the wrong thread to drag Hitchens in.
Posted by: Derek
|
August 17, 2010 9:26 AM
CogSci graduate student here...
PZ is spot on. Kurzweil is a raving loon in his simplified comparisons of the information and encoding of biological systems to artifacts. Read just about any of his books, and you'll see ridiculous charts comparing things like the brains of slugs to ENIAC. To make those kinds of comparisons, you have to have a reasonable understanding of both things you are comparing, and when it comes to understanding how brains work, we are still relatively clueless.
Kurzweil may be right in a broken clock sort of way about us reverse engineering the brain, though it's highly unlikely within that time frame, and it's definitely not going to be via plunging down to the level of genes, and it's most certainly not going to be in faithfully replicating the encoding in software.
I'm pretty certain the analogy will be much more like artificial flight, only far more complex. We have to continue to study the brain to uncover the fundamental principles upon which it works. In flight, these are base concepts like lift and thrust. Once we understand the core functions, we can abstract away all the unnecessary biological messiness. Jets don't fly like birds. In some ways they're better (speed) and in other ways they're worse (maneuverability), but they both essentially fly based on the same principles. Eventually we'll build machines that perform similar algorithms as the neocortex, though they'll likely be qualitatively different as well. We won't do it until we figure out the base principles, and we most certainly won't do it by staring at protein sequences.
Posted by: John Morales
|
August 17, 2010 9:28 AM
[meta]
You're trying hard to derail the thread, in true troll fashion.
The topic: Ray Kurzweil does not understand the brain.
What the hell do Hitchens and Iran have to do with it?
Posted by: ian.monroe
|
August 17, 2010 9:28 AM
As a member of the "CS set", the paragraph you quote doesn't even make sense from a CS perspective. I've never heard of someone predicting the size of a code base like this. Software "engineering" (I put it in quotes, since we'd all be screwed if bridges were built on the principles of software engineering) has a hard enough time figuring out whether a given project will take 3 months or 30 months.
Maybe Kurzweil should go hit the keyboards and remember what software development is actually like.
Posted by: Darrell E
|
August 17, 2010 9:30 AM
Posted by: Sastra | August 17, 2010 8:53 AM
The scientism accusation is of the same type as the it takes just as much faith to believe in science or materialism as it does to believe in god stuff accusation. The person making such claims is projecting their own fallacies, and at the same time seems to be aware at some level that they are, and that that is a bad thing.
Ray Kurzweil should probably switch to writing soft science fiction stories. He seems to have a good imagination but he doesn't seem to have the self discipline to balance his strong desires that is necessary when attempting to seriously predict technological trends. Especially when you are dealing with things outside your own area of expertise, in which case extra effort is required to make sure you understand what you are talking aobut.
Posted by: HappyHax0r
|
August 17, 2010 9:30 AM
Wow, just... wow... I have no words.
Kurzweil used to have some decent ideas and thoughts, heck even some "fairly grounded" predictions, but this is just fucking absurd.
Posted by: Gingerbaker
|
August 17, 2010 9:30 AM
Well, you may be pretty nifty with all this science-y stuff, but let's see you invent an electric piano!
Posted by: Alex
|
August 17, 2010 9:37 AM
Minor remark:
I don't think that's correct. If you represent every amino acid by 5 bits, it takes 83 bytes for the whole string. I don't think you can get that down to only 8 bytes. In fact, the minimum number of bits necessary in order to be able to represent every string of length 132 with an alphabet of 20 characters is 571 (or 72 bytes).
Posted by: Duckbilled Platypus
|
August 17, 2010 9:38 AM
hje at #17: I actually would hope that a Rapture would occur and leave us behind, for then, truly, the Geek Shall Inherit the Earth.
Posted by: Antiochus Epiphanes
|
August 17, 2010 9:39 AM
Kurzweil doesn't understand the brain. Or the genome. Or computers.
Where is my air-car?
Posted by: Stephen Wells
|
August 17, 2010 9:39 AM
@25: You're right, but Kurzweil would compress it to 8 bytes anyway.
Therein lies the problem...
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 9:40 AM
To come to Kurzweil's defense here, you don't store all the visual information you perceive. Your own perception is extremely lossey. That's part of the reason eye-witness testimony is secondary to physical evidence.
100GB does seem a little low, but what do I know?
The problem with Kurzweil is that he's passed his "use-by" date. When he's correct, he is trivially correct (when he talks about emergence, for instance, which we've known about since before Conway's game of life). When he is wrong, he is terribly wrong.
I think the problem is he expects all emergence to be reducible to simple rules. This is fundamentally untrue. Systems of complex rules also result in emergence. You don't know when deconstructing a system whether the fundamental rules are simple or complex.
Here, he expects the DNA to provide the fundamental rules. That isn't the case. It's complex chemistry, with all the strange effects and interactions complex chemistry provides. This includes potential quantum effects, such as quantum tunneling in photosynthesis.
When you look at it, we are merely one tiny part of a long, continuous chemical reaction that started over 4 billion years ago.
DNA is a catalyst in this chemical reaction. Kurzweil sees it as the cause of the reaction, but it isn't. You can't deconstruct life to a set of rules encoded in DNA.
Posted by: startlingmoniker
|
August 17, 2010 9:40 AM
I actually think the guy is right on. I know he's totally burned in the skeptic circles, but I don't see too many good arguments against his ideas other than:
1) They're really very wild.
2) They're very difficult.
Frankly, this skeptic community concept that the brain somehow isn't going to be understood in this manner almost seems like you're ready to put it back in supernatural territory. Take the f**king thing apart, replicate the pieces in software, replicate the chemicals in software. It's only a complex piece of meat, we can do this!
As for the brain-genome-pattern statement, I have no idea. I'll admit, though-- when I read Dawkin's latest book, and he was discussing origami and how it relates to our structure, it DID get me thinking about how we could reverse-engineer this sort of thing. If there's only so many ways the structure could proceed from a given point, isn't this basically the same as what Kurzweil is saying?
Posted by: andrew h
|
August 17, 2010 9:40 AM
kurzweil's futurism usually seems to be something of the type "we know a little bit about something resembling X; therefore in 10 years we will be able to simulate X on a computer!"
i remember having a good argument years ago with an engineer friend who had read a kurzweil book; something about how, since we had developed some relatively functional speech/character recognition software, we were just steps away from simulating the mind. people who make these kinds of arguments are necessarily people who know just about nothing about psychology, psychophysics, and neuroscience, and about just how freaking complicated the mind/brain really is.
my prediction (i am a postdoc psychologist working in vision science): we will understand the brain and be able to simulate its functions in about 300 years.
Posted by: Steven Dunlap
|
August 17, 2010 9:42 AM
Ray Kurzweil has been driving me nuts for years. He has had a regular column in Library Journal for some years and spouted all sorts of nonsense that demonstrated a a marginal, threadbare understanding of information theory and human behavior. He belongs to a camp that believes in technological solutions to everything. He has called himself a "futurist" which has all the credibility of a fortune-teller (along with the same degree of accuracy).
Posted by: Yoav
|
August 17, 2010 9:44 AM
It's the same type of stupidity you get from the idiots who claim we don't need to experiment on animals since it can all be done by computer simulations.
Posted by: pveljko
|
August 17, 2010 9:45 AM
I freely admit that I, poor CS peon that I am, know very little about the functioning of the brain. But I do know a thing or two about how programming works and what Kurzweil is trying to sell here is the CS equivalent of homeopathy. Even if his starting premise is correct, the method of getting lines of code (which is such an objective measure of code complexity) he used is, essentially, voodoo.
If the genome is the code, then the computer used to run it is the universe. Or, rather, the brain-breakingly complex chemical environment and development pathways that actually build the brain. Those aren't the equivalent of a computer. Any code that purports to turn genomic data into a brain would have to emulate this environment in exhaustive detail. This may be arbitrarily difficult.
I agree with Kurzweil insofar as I believe that one day, strong AI will be possible. We know that a thinking machine can be built (evolved, in fact) because we are one. The difference is that I think such developments are far off. Very far off. Certainly more that fifty years, which puts it well beyond the prediction horizon. No magic technology will appear to save anyone currently living from death. And that's fine, really. With luck, we may have a few years extra as compared to our ancestors.
Posted by: Yoav
|
August 17, 2010 9:49 AM
@#34.
You ignore the option that the time frame for development of a thinking machine will be significantly shortened by copying the chips of a killer robot sent from the future to kill the leader of the human resistance.
Posted by: KG
|
August 17, 2010 9:50 AM
startlingmoniker,
FFS, PZ's post is an argument against Kurzweil's idiocies that does not fit either of your two categories.
Don't. Be. Silly. No-one has said this is literally impossible - but it won't be done in ten years or anything like it. For that matter, it's a foolish misunderstanding to think that reverse-engineering the brain would give you a true AI: the human brain evolved to develop and operate in a body, in a social system: it doesn't just boot itself up like a computer.
That much is quite obvious.
Posted by: zenstoic
|
August 17, 2010 9:51 AM
While I agree that Kurtzweil's assessment of the complexity of the brain is very wrong, even if he's off by orders of magnitude his time estimate for simulating a brain is not that wrong. Assuming Moore's Law holds, and it may not when we strike the silicon wall at the atomic limit of how small silicon based processors can go, then if he's underestimated the difficulty by half, it delays simulating the brain only by two years or so.
Assuming we don't really understand the brain, and simply brute force the physics of the development of an entire human, and that this task is 1000 times more difficult than Kurtzweil estimates, it delays doing it by only 20 years or so.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 9:53 AM
And don't forget:
3) They rely on breakthroughs in understanding which we simply cannot predict
And, as was pointed out in the original post, this prediction is based on a false premise, that the brain can be trivially derived from DNA.
Hell, we still don't have an elegant solution to the three-body problem, which is far less complex than modeling protein folding (just one problem that requires solving).
Posted by: zenstoic
|
August 17, 2010 9:55 AM
While I agree that Kurtzweil's assessment of the complexity of the brain is very wrong, even if he's off by orders of magnitude his time estimate for simulating a brain is not that wrong. Assuming Moore's Law holds, and it may not when we strike the silicon wall at the atomic limit of how small silicon based processors can go, then if he's underestimated the difficulty by half, it delays simulating the brain only by two years or so.
Assuming we don't really understand the brain, and simply brute force the physics of the development of an entire human, and that this task is 1000 times more difficult than Kurtzweil estimates, it delays doing it by only 20 years or so.
Posted by: Pinkydead
|
August 17, 2010 9:57 AM
Are you sure this isn't misquoted or something?
The start of the article from Gizmodo makes a lot of sense - sensible statements about petaflops and singularities are what you would expect from someone with a good grasp of AI.
But suddenly it starts into genomes? That's alcohol talking. (Ironically, a brain state that can be easily simulated currently with any PC).
Posted by: Stephen Wells
|
August 17, 2010 9:58 AM
@30: you say we should "replicate the pieces in software, replicate the chemicals in software."
I'll get on with writing a completely accurate ab-initio quantum molecular dynamics program that handles around 10^26 atoms and runs in faster than real-time, you get on with obtaining the location of every atom in a human body (got to get the initial conditions right), and we should have that done by teatime, right?
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 9:58 AM
So, we're going to have quantum mechanics licked within 20 years? Are you sure about that?
When will we have a verified quantum model that will allow us to model the human brain based on DNA code?
Posted by: te24hours
|
August 17, 2010 10:05 AM
@28: Well yes. At least, one of myriad problems...
Posted by: austin.rochford
|
August 17, 2010 10:06 AM
Here you're assuming that the only thing holding such a simulation back is processing power; I would argue that the situation is much more subtle than that.
Posted by: llewelly
|
August 17, 2010 10:13 AM
crosbie.fitch | August 17, 2010 9:03 AM
Kurzweil does not believe technological progress happens in leaps and bounds. He believes technological progress zooms down a wormhole and comes out in a different universe.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 10:19 AM
As anyone who has actually worked on a software project should know, the complexity, feasibility, and effort of an implementation is almost a complete unknown until specification and requirements documents are written. Until then, you really don't have any fucking clue what you are talking about.
As we don't have a specifications or requirements document, anyone who comes in and says, "Oh, it'll take x years," is completely full of shit.
If I have learned anything in my years, it's that software is always more complex and more difficult, and will take far longer, than anyone initially estimates.
In this case, we don't even have an idea how long a requirements document would take. Until we are able to produce one, anyone proposing a feasibility timeline is pulling it out of Kurzweil's ass, the same place he got his estimate.
Posted by: swelke
|
August 17, 2010 10:20 AM
You're such a biologist, PZ...
"The genome is not the program; it's the data. The program is the ontogeny of the organism, which is an emergent property of interactions between the regulatory components of the genome and the environment, which uses that data to build species-specific properties of the organism. "
I agree with you... almost. But the program doesn't have to be the ontogeny of the organism. As you point out, we're nowhere near being able to build that yet. Besides that, if we follow your model then every time a person is discovered to have a unique genotype of any protein, we'd have to go back to the labs to figure out what the heck it does.
A more robust (though also more computationally intensive) model would be to use the genome as the data and the laws of the universe as the program. If you do a complete atom-for-atom, quantum state-for-quantum state simulation of the human brain in question then the ontogeny of the organism will, in fact, be an emergent property of the program. You'd have to run the program all the way from conception to the age you want to examine. Since we can't even simulate a single cell that way yet, this is too far out there to be able to even guess how long it will take. Ten years is certainly too soon though.
Posted by: fractal
|
August 17, 2010 10:21 AM
"He has to simulate all of development from his codebase in order to generate a brain simulator"
No he doesn't. He just needs to make a blank slate, and unleash his electronic brain on the world. Once you have a brain - it can learn.
Just as Mandelbrot discovered you need only the simplest of rules to make infinite complexity... a million lines of code could very well create a model for the brain. You're not teaching this brain English - you're giving it the ability to learn it.
Posted by: llewelly
|
August 17, 2010 10:21 AM
nigelTheBold | August 17, 2010 9:23 AM:
To "refudiate" an argument means to refute with the totality of knowledge which can come only from reading all the newspapers.
Posted by: moioci
|
August 17, 2010 10:21 AM
If he were to argue from function and not ontogeny, that over the next decade we could better understand and model neuronal interactions so as to engineer a strikingly better simulation, I doubt that anyone would argue with him, except regarding lines of code.
Posted by: Naked Bunny with a Whip
|
August 17, 2010 10:22 AM
The Catholic Church is a hierarchical organization with a specific set of doctrines and well-defined management; membership and participation are entirely voluntary. Hardly equivalent to simply being an atheist.
That said, I almost never see Hitchens' name come up without a fair number of people bringing up his atrocious cheerleading of Middle Eastern butchery. Now wander off to some pro-Catholic forums and see if you get a similar result every time the Pope is mentioned.
(I'm not even sure why I responded to someone using words coined by right-wing morons.)
Anyway, back on topic. It's pretty bad when even a CS guy like myself can see the gaping holes in Kurzweil's thesis. I do think we'll eventually develop artificial minds, but they will either be based on interlinked, predefined abstractions, or they will be grown "organically" like the mind seeds in Greg Egan's Diaspora.
(These begin as a DNA-like chunk of data which is then run through a complex virtual machine "womb" which interprets the mind seed in numerous iterations, each iteration building on the prior one. It's still a functional abstraction, however, not mimicking actual chemistry.)
It sure as shit won't be happening in ten years, though.
P.S. Who the hell still measures software complexity in "lines of code", anyway?
Posted by: Katharine
|
August 17, 2010 10:22 AM
Kurzweil is a businessman, not a biologist.
This should tell you something - the man has had no training in this area at all.
Why anyone trusts Kurzweil on this I don't know.
Posted by: Cuttlefish, OM, CR
|
August 17, 2010 10:22 AM
Ah, thanks for this one, PZ!
One of the things that always bothered me about (many, but not all) cognitive scientists and neurobiologists (including a good many who are working on the neurobiology of consciousness, as Sejnowski is) is their adherence to a mechanist philosophy rather than a contextualist one. The questions and answers are very different under each model, and it matters. For a mechanist, all too often the search is for the proximal causes of a particular conscious event, with the assumption being that a neural state must necessarily give rise to a particular conscious state.
No.
It makes sense that an evolutionary biologist would get it right--natural selection is an ideal example of a functional contextualist explanation (another is radical behaviorism, which is too often dismissed by those who have ignorantly confused it with methodological behaviorism). Proximal causes are all well and good, but the ultimate causes are what are important, and those are always found in an organism's (or a species's) history of interaction with the environment. The exact same (mechanically speaking) actions are vastly different things in different environments, and are vastly different things depending on different histories with that (or other) environments.
One cannot understand consciousness in slices. Consciousness does not exist in single frames, but in continuous interaction with our environment (including, of course, one another). The number of possible variables one must code for in this interpretation makes Kurzweil's mere million lines of code child's play in comparison. If he wants to reverse engineer a brain that way, good luck to him--at best, he'll get something that, if hooked up to a reverse-engineered body, in a reverse-engineered environment, and left to interact for a few years, might develop rudimentary communication skills.
Oh... was this supposed to be in verse?
Similar topics, among others, were discussed in verse (mostly other people's verse, but some of mine) here: http://digitalcuttlefish.blogspot.com/2009/02/daniel-dennetts-darwin-day-delivery.html
Posted by: winterblood
|
August 17, 2010 10:26 AM
As a professional programmer, I'd just like to observe that there are well over a million lines of code in LEGO Harry Potter.
Posted by: ihatelivelyids
|
August 17, 2010 10:27 AM
So basically, Kurzweil only talks about storing the data structures (proteins) but NOT about the business logic (interactions between proteins). So it'll be like having an empty database model without an actual application running it.
Sounds a bit useless...
Posted by: tjhewitt
|
August 17, 2010 10:27 AM
Where is the program stored?
Posted by: irenedelse
|
August 17, 2010 10:27 AM
Regarding Kurzweil: could it be an instance of the Peter Principle?
Posted by: Sven DiMilo
|
August 17, 2010 10:28 AM
The brain does not understand Ray Kurzweil.
The brain doesn't understand the Cuttlefish either, but thinks it writes prose good.
Posted by: Darrell E
|
August 17, 2010 10:28 AM
Posted by: zenstoic | August 17, 2010 9:55 AM
You are leaving out the most important consideration. Computing hardware capabilities are not the limiting factor here. Understanding how the human brain functions well enough to simulate is the issue. What the OP is saying is that the difficulty of doing that is such that Kurzweils prediction is very likely way off. And the reasoning he uses illustrates that he does not understand the problem.
It doesn't matter how many calcs per second your computer can do if you don't know how to construct the simulation.
While we are far far away from being able to do what you propose here, we know enough to understand that this task would be orders of magnitude more complex, and larger, than what you are suggesting. See comment #41 above for more.
Posted by: Daniel B
|
August 17, 2010 10:30 AM
Nice to see you attempt a translation. At a theoretical level the program and the data are interchangeable so what you've said isn't exactly a surprise to any 'computer science guy'. However the addition information encoded in the environment is certainly relevant. It would be interesting to hear you speculate about the order of magnitude of this additional data.
To my naive 'computer science guy' mind it would appear that it could at most double the amount information content. Surely all the external goings-on of the external world aren't necessary to 'run the program' of life. I would have though that gravity, some details of the the biology of the mother and perhaps some information about nutrients would suffice. Details such as playing Mozart to the unborn child may be have an influence but are they entirely necessary? Reports of successful interspecies surrogacy do seem to support this believe.
Posted by: tommorris
|
August 17, 2010 10:31 AM
Oh yes. Another opportunity to post the Mitch Kapor quote that I love so much about Kurzweil and the Singularity kooks:
"It's intelligent design for the IQ 140 people. This proposition that we're heading to this point at which everything is going to be just unimaginably different - it's fundamentally, in my view, driven by a religious impulse. And all of the frantic arm-waving can't obscure that fact for me, no matter what numbers he marshals in favor of it. He's very good at having a lot of curves that point up to the right."
Posted by: KG
|
August 17, 2010 10:35 AM
zenstoic,
Ah. I hadn't realised that when you have the hardware, all the software writes itself.
Jesus wept. Just read what Stephen Wells wrote.
Posted by: Harry Tuttle
|
August 17, 2010 10:37 AM
The Catholic Church is a hierarchical organization with a specific set of doctrines and well-defined management; membership and participation are entirely voluntary. Hardly equivalent to simply being an atheist.
But, apropos of nothing, a 1:1 correlation to being a scientist.
Change some terms and PZ's post right here would read like a papal condemnation of a heterodoxy.
Posted by: Cos
|
August 17, 2010 10:39 AM
I've finally gotten around to reading Carl Sagan's "The Demon-Haunted World", and it was the exposure of this kind of pseudo-scientific nonsense that he was lamenting. With all of the interesting things that real neuroscientists are understanding about the brain, this drivel is what the poorly educated public will latch on to. If only Kurzweil's immortality could have been transferred to Sagan - we miss you, Carl.
In my experience in bioinformatics, I've found that it is *much* easier for biologists to learn enough computer science to do something useful than the other way around.
Posted by: KG
|
August 17, 2010 10:42 AM
Daniel B.Naive indeed. How do you think your brain got to its current state, Daniel? How did you learn language? Did gravity and nutrients teach you? How did you learn to interpret what you see? To walk, swim, ride a bicycle? To understand social situations? (Well, maybe you didn't need this one.) There is vastly more information in the environments you encountered than in your genome.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 10:43 AM
Uhm, could you expand this sentence fragment into a full thought? You lost me here.
Right. Except PZ's objections are based on, you know, actual knowledge of how reality operates.
Posted by: KG
|
August 17, 2010 10:48 AM
Harry Tuttle,
Don't be so fucking stupid. PZ does not claim the authority to define doctrine; the Pope does. See the difference?
Posted by: paul.joseph.richardson
|
August 17, 2010 10:50 AM
How to Simulate the Metropolitan Brain
I think it's important here, to distinguish between the human mind and the human brain.
If you want to simulate the mind, there are two approaches. Start with the easy stuff (conscious processing of near monkeys) or build a foundation of fundamental mechanisms (memory, schemas, etc.).
But if you want to build a simulator for the 'brain', you are indeed going to be faced with something far more complex.
First, the biggest mistake people make (in my opinion) is that they frequently think of the brain as an isolated object, suspended in a vacuum and only occasionally penetrated by data entering or leaving.
I would suspect that a more successful approach to simulation of a fully developed human brain, would be to approach the 'brain' as an urban road system, centered in the downtown section of a major metropolis. In such a system, there are at least 2 major KINDS of evolutionary change.
The metropolis of the brain, is barely more than a GEOGRAPHIC BOUNDARY symbolized by a thin organic material penetrated by the many blood vessels and nerves that feed in or out of it.
When you consider the metropolis of the brain in this fashion, you can see that one kind of evolution relates to changes in traffic over time (function, design, capacity, density, etc). But a FAR MORE DRAMATIC kind of evolution, for our metropolis of the brain, is that of the environment. The environmental impact on change over time, is hard to overstate.
So, where to start?
If you begin with the 'big bang', you are going to have a large number of unnecessary task ahead of you. And for that matter, who is expert enough in evolutionary theory to correctly gauge biological momentum? That is to say, most paths are close enough early on, that we can approximate them to be equal. We do this in nuclear physics all the time, and can still predict thermonuclear reactions within a reasonable error level in energy plants.
Now, if the evolutionary direction is not easily changed, due to some sort of momentum (I think this a very small probability), then these small differences early on, can lead to very large future differentiation. Thusly, I would say that we may justifiably hold one hand over our eyes, and arbitrarily 'pick' a starting point in the storyline of the metropolis of the brain, early on, but not too much so.
Humans are apparently better than computers, for 'averaging' out patterns from the chaos of what may seem stochastic, but needs only someone to update probability guesses constantly (as in Bayesian logic), as environmental pressures evolve. The recent 'Nature' article on FoldIT (http://fold.it/portal/) seems to indicate that anyhow.
If I didn't know better, I'd be tempted to say that we should start our story of the Metropolitan brain AFTER the epoch of protein folding, and just pretend they are all folded already for us.
But there is a level of complexity in this problem not yet emphasized (not enough, anyhow), which is that the protein folding itself, is an ongoing and integral part of how the metro-brain develops. There may be several different conformational shapes in a folding protein, each of which is timed and controls or exposes active areas for interactions or signaling.
Posted by: phantomreader42
|
August 17, 2010 10:53 AM
John Morales @ #20:
Buster's purpose in this thread is to prove Kurzweil right, by regurgitating preprogrammed bullshit without regard to logic or context, thereby demonstrating that the actions of something supposedly controlled by a human brain could in fact be simulated by a handful of lines of code.
Posted by: jasonnyberg
|
August 17, 2010 10:54 AM
Arguing about the environment genes are expressed within (in the context of Kurtzweil's argument) is just about as superfluous as criticizing him for not accounting for the architecture of the computer system that would run the analogous "million lines of code" he talks about.
You don't need to explain the details of gene expression to explain the process of evolution, do you? It's an argument directed at a layman, at least one who understands that genes are the medium in which evolutionary adaptations are written.
All Kurtzweil is saying is that if evolution can produce a brain essentially by accident (a process that took billions of years to accomplish), that we with
A) our intelligence directed at solving the problem,
B) ever improving technology supporting our attempt to solve the problem, and
C) the existence of many, many actual examples of brains of varying degrees of complexity to study,
we ought to be able to engineer a reasonable, functional facsimile of a brain, and we might just be on the verge of doing so.
And it might well be done in a million lines of code. Of course, the "intelligence" part will come from what happens when that code is expressed... i.e. what happens when it runs, in the environment it runs in, how it is exposed to training, etc.
(From what I've seen, the pieces are already in place. We're going to see some amazing things in the next few years that people will dismiss as being "too easy" to be "real" intelligent behavior... And they'll be wrong, mistaking lack-of-scale for lack-of-quality.)
Posted by: Steve LaBonne
|
August 17, 2010 10:56 AM
#68 makes a good point. I personally know people who work on protein-folding prediction, and the difficulty of that far "simpler" problem really ought to sober up anybody who imagines that we'll be simulating a brain in the foreseeable future.
Posted by: viggen
|
August 17, 2010 10:56 AM
First of all, I agree strongly with everything said here: Kurzweil is a wank. I'm not too fond of his whole singularity prediction either.
Now, one amendment I would make to what you've said. One thing we are doing right now is using sequence homology to make predictions about the functions of some unknown proteins based upon empirically determined knowledge about the functions of proteins with homologous sequences. We are inferring functionality in humans based upon things explored in other animals or in hybrid systems. This is obviously not what Kurzweil is promising, but it is a way to focus our research, and it may knock decades off of being able to do what Kurzweil suggests.
Also, I would circumvent Kurzweil for the wank he is and invoke Turing. Why try to simulate the brain by simulating it's genetic functions? If you build a "mind" by whatever means whose behavior humans can't distinguish from a person, how have you not achieved the same end? By this method, I would say we're a lot closer to the goal because we know a lot more about the necessary tools.
Posted by: KG
|
August 17, 2010 10:58 AM
- fractalThe Singulatarians really are as stupid and stubborn as Jehovah's Witnesses. First, brain development - even before birth - is not fully specified in the genome: it depends on how that genome interacts with its environment - the zygote, and the mother's physiology. How many more fucking times do you need to be told this? Second, a brain can't just "go forth and learn". It needs a body, and a physical environment. For a human brain (and to a lesser extent most other mammalian brains), it also needs a social environment - it needs to be taught.
Posted by: davidnicol
|
August 17, 2010 10:59 AM
Sure, just a million lines of code.
The VM that that million lines of code runs on, however, is left as an exercise for the reader.
Posted by: danielm
|
August 17, 2010 11:03 AM
I'll state I'm an advocate of the nerd rapture straight off, but I don't think he's the prophet for a new millenium. I'm also quite convinced that getting an opinionated person like Kurzweil to talk on subjects he doesn't underestand is a recipe for hilarity. Getting quoted by a reporter who also doesn't know what he's talking about is doubly so. Quite likely somebody like Kurzweil would dumb-down the response until he sounds like an idiot even if he does know what he's talking about, so keep that in mind.
Having said that, if you just had to give a number for the amount of coding, you could do worse than what Kurzweil did. Is it correct? NFI - and I certainly hope he doesn't think it is either.
On the other hand, you don't need to simulate a growing brain to simulate a working brain. You don't need to simulate chemistry if you can build a brain-like network with comparable geometry, and you definitely don't need a brain hardwired to deal with all the plumbing that humans (or any other animal) tags along with them unless you're going the whole hog and simulating an entire nervous system to go along with your simulated brain.
So, to address some of the "concerns", whatever gets built over the next few years as we work up from where we are now (just scratching the surface with things the size of rat and cat brains) it won't be created by aping DNA, protein folding and hormones.
Whatever gets run on one of these simulated brains, it might be human-like in ability and it might even "remember" being human if we can manage to upload memories from a real, live human, but it probably will NOT be a brain simulating every physical aspect of a meat brain, it will probably be something built to look like a brain that runs like a brain and does what a brain does, just not the same way.
And if you want to talk about "resurrecting" dead people inside a computer, all you need for a convincing simulation is enough information to know what said dead person would or would not do - I say "all", but it's a shitload of data, responses, knowledge, images, thoughts and feelings. It will not be "bringing person X back to life", it will be "creating something similar to person X" and it may even call itself "X" but it won't be X.
Posted by: Mumon
|
August 17, 2010 11:03 AM
The genome is not the program; it's the data.
Actually, you're wrong too. The genome is both the program and the data.
But I agree with you, for entirely different reasons, as an Ph.D. E.E.
The problem with Kurzweil is the "brain" as a system is neither controllable nor observable based on what we know because the "brain" is not and cannot be isolated from everything else, unlike Young Frankenstein's Ab-normal.
And because it's not isolated, it changes.
Posted by: paul.joseph.richardson
|
August 17, 2010 11:07 AM
PS. I suppose I should have prefaced my comment by disclosing that I think the human brain simulation must be built by modeling the phylogentic (as in cladistic) development of the human brain, as opposed to exclusive ontogenetic development. Hence, we pick a point in the story, early enough to define trajectories.
Posted by: phantomreader42
|
August 17, 2010 11:08 AM
Harry Tuttle @ #63:
So, you're saying all science is controlled by a hierarchical organization led by an elderly Nazi pedophile in a dress who claims to be infallible? And you're putting PZ at the top of this imaginary conspiracy you hallucinate, as Lord of All Science?
You don't have the slightest idea what you're talking about. Are you trying to support Kurzweil's claims that the brain can be easily simulated by demonstrating that YOUR brain is unusually simple?
Posted by: KG
|
August 17, 2010 11:08 AM
jasonnyberg,
True.
False.
Posted by: Ed S
|
August 17, 2010 11:11 AM
I think the idea of modeling the brain based on duplicating the biology is dumb. Certainly, as PZ points out, the biological processes in each brain cell are extremely complicated, but on the other hand, most of those are related to keeping the cell alive, making it grow, replicate, die, etc. A computer program can ignore all that. The only thing essential to replicate the brain is to create a machine with an appropriate number of processors running programs that replicate what the each cell in the brain does in interaction with it's neighbors. On an individual cell basis, the interactions may be fairly straightforward. The complexity lies in the number of cells all working at once and the nature of the connections between those cells.
The number of processors can be less than the number of cells in the brain if those cells which are exclusively dedicated to keeping the body working, or engaged in other non-essential (for our purpose) activities are eliminated, leaving only those involved in actual conscious thought.
That's still an intimidating number of independent processors (cells), but it seems like something that could be done eventually. There is also the challenge that cells make new connections in the brain all of the time, so a static arrangement will not be sufficient. The networking of the processors would have to be virtual in some way, allowing for the new connections to arise as the machine "learns". The will probably be the most serious obstacle.
Posted by: Dan Kaminsky
|
August 17, 2010 11:12 AM
===
Here's how that math works, Kurzweil explains: The design of the brain is in the genome. The human genome has three billion base pairs or six billion bits, which is about 800 million bytes before compression, he says. Eliminating redundancies and applying loss-less compression, that information can be compressed into about 50 million bytes, according to Kurzweil.
About half of that is the brain, which comes down to 25 million bytes, or a million lines of code.
===
Hmm. Would this work for software itself?
"The World of Warcraft DVD is 5 gigabytes. Eliminating unnecessary quests and applying lossless compression brings us down to 500 megabytes. About 20% of that is core gameplay, which comes down to 100 megabytes, or two million lines of code."
So, the compressibility of some arbitrary subset of a working codebase links directly to an estimate of how many lines of code it would take to reimplement that codebase in a compilable language?
Hurm.
Now throw in that we're actually not reimplementing on the same architecture, or even the same computational model, and I think we have the gold standard for "worst project estimation effort ever".
Posted by: jidashdee
|
August 17, 2010 11:13 AM
To simplify it so a computer science guy can get it...
Hey! That wasn't called for! As a math and comp sci grad I must protest on behalf of my colleagues who do actually get it. Won't you be catching some flak for that from a certain daughter-unit?
That said, yes, Kurzweil is an idiot. The first clue is that he's talking about compression right off the bat. Compression is the least of the problems at hand.
Even if we could extrapolate from DNA to the physical structure of the brain assuming a given set of parameters for growth - which will not be possible for many, many years to come - that still doesn't really tell us very much about how thought works. In fact, screw thought. We don't even understand memory to the point where you could code it.
Even if we had a brain scanner that could read the exact chemical encoding of our memories we'd still need to understand thought fully to be able to tell how much of our memories are actually manufactured as we're in the process of recalling them, not to mention how we make decisions or perceive our environments in the first place.
I hate the term "emergent property" but it really does apply in the case of human thought, perception, and memory. The systems locked inside the average human headbone are quite likely to be as complex as the earth's climate - something we still don't have a good handle on.
In conclusion: Kurzweil is either an idiot, a flimflam man, or both, but please don't tar all of us with the same brush.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 11:15 AM
I think comparing evo-devo with computers and programs is a false equivalency. They both process data, but do so in such fundamentally different ways, there's no way to really compare them.
DNA is one element in a biochemical process. There is nothing to compare in a computer -- not the hardware, nor the software, nor even the data processed by the software.
If we ever develop AI, it sure as fuck won't be by simulating the brain. We'll develop it based on our understanding of intelligence and consciousness. The brain may be our guide to unraveling some of these processes. But our AI won't come from modeling the brain directly.
There. Now I'm a futurist.
Posted by: shawn.hartsock
|
August 17, 2010 11:16 AM
Lines of code and complexity are not directly correlated. It *might* be possible to simulate the entire universe (let alone the human brain) in one million lines of code... but they might be very very long lines or very very complex lines written in an obscenely complex language created for the specific domain. The creation of the programming language itself might be orders of magnitude more complex than creating a human brain in the resulting language.
Or perhaps, it is all astoundingly simple relying on very simple cellular automata ... but it will require profound intelligence to understand something so simple. In such a simple language of infinite complexity a single phrase might encompass the entire universe.
And we still won't have a clue what that protein does.
Posted by: Thom Blake
|
August 17, 2010 11:17 AM
Kurzweil has been pretty good at making money via his predictions so far. It's easy to say "We couldn't possibly predict..." based on "my experience" or whatever. But he developed his prediction methodology as a way to ensure his inventions will hit the market when they're ready to take off, and it works. Certainly there's reason to be skeptical, but I don't see any reason to resort to name-calling just because he keeps making predictions. I do hope the myriad people saying nasty things about Kurzweil now are willing to apologize profusely for their rudeness in the (unlikely) event that he ends up being correct on this one. Or, you know, stop being jerks even if he's wrong.
Posted by: John Frum
|
August 17, 2010 11:17 AM
I always make sure each l
ine of code is 25 bytes l
ong
Posted by: morgan in austin
|
August 17, 2010 11:17 AM
Futurists, SMBC
http://www.smbc-comics.com/index.php?db=comics&id=1968#comic
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 11:22 AM
Dude, whatever his prediction methodology was back when he was actually inventing things, his new prediction methodology is based entirely on a fear of death.
And in this case, his prediction is based on a complete misunderstanding of basic biology. It isn't a "we can't possibly predict" situation. It's a, "For a smart guy, he's being a fucking moron" situation.
Posted by: donsalva.wordpress.com
|
August 17, 2010 11:29 AM
I concur!
Posted by: KG
|
August 17, 2010 11:29 AM
The current state of the art in "brain simulation" does not produce anything that actually does what a real brain does: enables an animal to make sense of its environment, decide what to do, communicate with others, etc. See here. A quote:
But the Blue Brain Project is focused on creating a physiological simulation for biomedical applications. By its own admission, it is not (yet) an artificial intelligence project. However, from an AI perspective, large-scale simulations of neural processes can be used as a virtual laboratory to study the neural architectures that generate natural intelligence and cognition.
It's good, useful science. It is not even an attempt to do what Kurzweil apparently thinks it is - to make an artificial intelligence equivalent in computational power to a mammalian brain.
Posted by: inklesspen.com
|
August 17, 2010 11:32 AM
Kurzweil is almost certainly right that we will someday have the ability to upload an existing human mind into a computer and run it there.
But he's almost certainly wrong about how easy it is, so it probably won't be someday soon.
Posted by: Flex
|
August 17, 2010 11:34 AM
Derek @19 has the right idea,
When consciousness in AI actually occurs we may not even recognize it as such because it will be structurally different than the consciousness that we are familiar with.
At least with man-made, powered flight there are a few signals which allow us to identify the common factors which link human-created powered flight with the powered flight which evolved within nature. The principles of lift are similar whether we are talking about birds, bats, or airplanes. But birds and airplane don't look/feel the same and it's foolish to assume that any consciousness which does emerge from our technology will look/feel like ours.
As for startlingmoniker @30, there are plenty of places where extended arguments against Kurzweil's idea are made, a comment thread is not really a good place to look. The first place to look would be where Kurzweil himself should have looked, at the plenty of on-line material discussing AI research and the nature of consciousness.
The short answer is that we don't know if duplicating all the physical structures of the brain would result in a conscious creature. We don't even know if we have identified all the things floating around in the brain. As an example, gaseous neurotransmitters were unknown 30 years ago.
Posted by: danielm
|
August 17, 2010 11:34 AM
nigel@83
If we ever develop AI, it sure as fuck won't be by simulating the brain.
but...why? what is it about the brain which is fundamentally un-simulatable? More to the point, where is this knowledge coming from that lets you know that the brain cannot, in any way, shape or form, be simulated?
It's complicated, yes, it's going to take some massive increases in power and/or some fundamental changes in how some of our hardware works...but why should it be impossible?
there are people claiming all sorts of bullshit like waving around the magic word "quantum" in an embarrassing, ignorant way...kind of like a drunken uncle at a party that nobody can ignore because he's so loud and obnoxious, and doesn't know he's drunk.
Maybe that's it, maybe the brains all quantum and stuff. Of course you can't show your work, because it's quantum.
Never mind they've been simulating neural nets for yonks, never mind they've been simulating slug, rat and cat brains (almost at least) and are already aiming higher, oh no, it's not-doable because fuck you, that's why not.
Posted by: KG
|
August 17, 2010 11:37 AM
The penultimate paragrpah of #90 should be within the blockquote.
Posted by: paul.joseph.richardson
|
August 17, 2010 11:42 AM
Ed S (comment #80),
You are assuming here that the 'conscious mind' operates by the use of neuronal electrochemical signals and synaptic connections.
There are many 'exotic' theories of consciousness (eg., Stuart Hameroff), which have not yet been completely ruled out. They are mostly dismissed by the conservative scientific community, and yet, we really just don't know VERY MUCH AT ALL about how cognition, mental analysis, imagination, memory, etc., all relate to the physical structures of the brain, other than a very basic sort of "this area is involved in speech (and x, and y, and z, etc.)".
Instead of searching abstracts for terms related to cognitive functions, search for keywords related to anatomical objects, structures, and chemicals. Then it will become painfully obvious that (for whatever reason, call it space limitations, or call it the need for pubs), not every background section and not every introduction section of peer reviewed neurocognitive journals, includes mention of the many OTHER FUNCTIONS implicated for the biological components they are attributing their cognitive functions to in their article.
Remember, IMAGE IS EVERYTHIGN. If you want to be hailed as a great scientific genius, you only talk about what you think you've figured out, and sort of ignore the vast complexity of how your 'discovery' is still obsfucated by other studies with contradictory or duplicitous conclusions.
If some thing A, means not just B and C, but means many many other things, then exactly how well do you understand A?
It's a very popular past-time to define your research agenda as a quest for understanding of the biological basis of cognitive functions. It's not so popular to define your research agenda as a quest for understanding of the cognitive basis for biological structures.
Posted by: Ewan R
|
August 17, 2010 11:44 AM
My guess on this would be that evolved structures aren't necessarily the best way to do things - the brain evolved to do what it did, probably using a torturously circuitous route to get there with all manner of seemingly random crap thrown in the way that had to be "evolved around" rather than redesigned from scratch - think recurrent laryngeal nerves, the path sperm have to take, the crossing of the air and food lines, Pelvis size vs head size, photorespiration and all manner of perfectly obvious shitty design flaws in nature - it'd be hard to imagine that at the molecular and non-obvious scale these same motifs of retardedness don't creep in on a near epic scale.
Posted by: Buster
|
August 17, 2010 11:44 AM
I think being an atheist must be like waking up on Christmas Day and running to the bedknob to find a sock with nothing at all in it.
Posted by: jasonnyberg
|
August 17, 2010 11:45 AM
#79, have you seen Jeff Hawkin's latest work on unraveling and simulating the functionality of the neocortex? He's got a very plausible model (HTM) for it's behavior, with actual corroborative implementations.
#84, it might be possible to simulate the entire universe with
Posted by: KG
|
August 17, 2010 11:46 AM
- danielmNope - not in the sense necessary for AI. See the link I gave @90.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 11:47 AM
Nowhere have I said we won't be able to simulate the brain in the future. We surely can't do it now, nor are we even close -- we just don't understand all the fundamental processes that occur within the brain. We understand a lot, certainly; but nowhere near enough to simulate it.
Neural nets do not simulate the brain. They are merely information processors that were inspired by the way neurons work; but they are not neurons, nor do they really simulate neurons.
As for the claims of simulating slug, rat, and cat brains, those are over-puffed PR pieces. They are not actually simulating the brains of a cat yet. They are more closely doing what I suggested will happen: they are mimicking the information processing that takes place in the brain.
There's a huge distinction between mimicking information processing, and actually modeling the brain.
We may be able one day to actually simulate the brain. There's no way to predict when. But I can assure you this: it will be far more efficient to build AI using information processing fundamentals, than it will be to create AI by simulating the brain.
Posted by: Brownian, Most Vicious & Petty of Pharyngulites
|
August 17, 2010 11:47 AM
Think harder.
Posted by: sqlrob
|
August 17, 2010 11:48 AM
Sorry, that does not follow. Let's look at another system for an example. How's weather prediction working out?
Posted by: jidashdee
|
August 17, 2010 11:48 AM
I think being a theist must be like waking up on Christmas Day and running to the bedknob (sic) to find a sock with nothing at all in it but imagining it to contain a very powerful and judgmental friend.
Posted by: KG
|
August 17, 2010 11:49 AM
Buster@97,
You're wrong. And a fuckwitted troll. Piss off.
Posted by: dnebdal.myopenid.com
|
August 17, 2010 11:51 AM
@KG
That depends on what brains you compare to, certainly? It wouldn't surprise me too much if we make something that is as "smart" [1] as the lowest end of mammalian brains in not too long.
Scaling up from there to hard AI isn't something I expect in my lifetime, though. (And yes, I'm a young computer scientist.)
[1] Just defining "smart" in a way that makes this comparable is hard, of course. Ability to adapt to new situations?
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 11:51 AM
I think being a theist must be like perpetually pretending to be a Klingon from Star Trek and realizing deep down that it's all make-believe.
Posted by: segfaultvicta
|
August 17, 2010 11:52 AM
Sometime Kurzweil fan here, and futurist, and transhumanist, getting that out of the way: I appreciate posts like this, because I appreciate when views that I hold, or views related to views that I hold, or those held by people I respect, are challenged, and Kurzweil is dead wrong on the specifics here - I've read these claims myself, they've made me cringe, both as a computer scientist (btw, PZ, what's up with the "dumbing this down for the CS majors" attitude? We can't all be biologists!) and as someone who tries his best to keep himself educated about biology and genetics and their proper applicability. The entire approach to brain resimulation coming from a standpoint of emulation on a level lower than that of an individual cell strikes me as more or less silly.
However, I think it's important to seperate what Kurzweil says about biology from what he says about things he's got a working knowledge of; to seperate that from what, say, researchers working on neuron simulation and cognitive science who happen to be pursuing the sorts of goals futurists talk about are saying, and to separate -that- from what somewhat more grounded futurists are saying. I think there's also a fair amount of strawmanning and lumping various dissimilar views together going on in this thread, and a lot of more or less random hatorade being directed at the -ideas- of transhumanism, even those which aren't particularly wild-eyed Utopian fantasies. By the same token, though, and as I mentioned above, it's good to be criticised, especially when a public and well-known voice associated with a set of ideas says something - and has a habit of saying things, especially concerning biology - so flagrantly, blatantly, disappointingly silly as what was quoted in the article, and it's good to see this sort of silliness pointed out. Thank you.
Posted by: KG
|
August 17, 2010 11:55 AM
Clicking on Buster's nym takes you to a page about George W. Bush. Very appropriate - two fuckwitted godbots.
Posted by: segfaultvicta
|
August 17, 2010 11:56 AM
Also, the subtitle of Kurzweil's movie - "A True Story About The Future" - really bothers me on definitional / honesty grounds. I don't care what your personal degree of certainty is about a prediction, you can't use the word 'true' like that! Nitpicking over an eyecatching subtitle, I know, but it bothers me.
Posted by: daveau
|
August 17, 2010 11:56 AM
I'd say: Oh boy! A sock! If I celebrated christmas.
Go away dimwit.
Posted by: Brownian, Most Vicious & Petty of Pharyngulites
|
August 17, 2010 12:00 PM
Yeah, but I like this meme. Here's my submission:
I think being a theist must be like waking up every day and running head-first into the bedknob.
Posted by: jasonnyberg
|
August 17, 2010 12:02 PM
We can simulate weather extremely well. The limitations of using simulations to predict weather also happen to be well understood. (Chaotic systems are by definition "unpredictable" as errors in initial state are amplified non-linearly.)
(Simulating a neocortex is a different problem than using a simulation of a neocortex to predict how another neocortex will behave.)
Posted by: Ewan R
|
August 17, 2010 12:05 PM
I think being a theist must be like waking up on Christmas Day and running to the bedknob to find a sock full of presents and goodness and believing with all your heart that Santa Claus put it there because you were on the good list.
Posted by: sqlrob
|
August 17, 2010 12:09 PM
Simulating it is a different problem than prediction.
Reverse engineering is not.
Posted by: KG
|
August 17, 2010 12:09 PM
- denebdalNo: that's not the point I'm making. The current generations of "brain simulations" aim at reproducing some of the dynamic features that can be measured in real brains. They do not perform any of the cognitive tasks - seeing, running, learning, decision-making - that real brains (even those of slugs) do. AI programs that do attempt some of these tasks work in completely different ways from animal brains - even those that are referred to as "neural networks". Kurzweil does not appear to know the difference between these two branches of computational work - which is surprising, given his background, but I think can be explained by his desperate need to prove to himself that he, Ray Kurzweil, is not going to die.
Posted by: First Approximation, L'esprit de l'escalier
|
August 17, 2010 12:15 PM
I think being a theist is like hitting the snooze button and going back to dreaming about things that aren't real.
Posted by: KG
|
August 17, 2010 12:16 PM
I think being a theist must be like waking up on Christmas Day and realising the bedknob is more intelligent than you are.
Posted by: raven
|
August 17, 2010 12:22 PM
Being a theist is like being a slug.
They probably play some minor role in the ecology but no one cares unless they are attacking the brocoli. And they are vaguely aware that pretending that mythology is real is stupid but are not capable of doing anything more intelligent.
Posted by: First Approximation, L'esprit de l'escalier
|
August 17, 2010 12:23 PM
I think being a singularitist is like waking up on Christmas Day to find a toy car and extrapolating from that that you'll get a real Boeing 747 by Easter.
Posted by: SheepdogB
|
August 17, 2010 12:23 PM
I concluded, after a brief flirtation myself with the whole Singularity/Noosphere rapture-of-the-geeks thing, that many techno-utopians like Kurzweil were succumbing to apophenia.
There seems to be a rampant over-simplification that happens in their minds similar to the "holism vs. reductionism" argument-which I reject as an expression of a tendency toward binary thinking in humans. It's impossible, I believe, to gain any understanding of a system from one level only. Kurzweil's lack of understanding of and apparent dismissal of the necessity of understanding of how the brain functions at the biomolecular level is equally as incomplete as the idea that historical environmental influences are inconsequential in understanding how the brain evolves.
"It can scarcely be denied that the supreme goal of all theory is to make the irreducible basic elements as simple and as few as possible without having to surrender the adequate representation of a single datum of experience."-Albert Einstein, at the Herbert Spencer Lecture at Oxford. This is the original of the often used but possibly misattributed quote-" Everything should be made as simple as possible, but no simpler."
Posted by: hanspeter
|
August 17, 2010 12:30 PM
Saying that the design of the human brain is anywhere else than in our DNA sounds somewhat creationistic. Isn't that implying than there is brain-design in proteins, cells and molecules?
Posted by: daveau
|
August 17, 2010 12:33 PM
I think that being a theist is like never waking up at all.
Posted by: grudgedk
|
August 17, 2010 12:33 PM
Well you're wrong too then. It's not just a matter of magnitude, but let's crunch the numbers:The most complex software we can write today is an operating system. The "modern" operating system (Linux, Windows 7, Apple OS X) already contains 80 times more lines of code than Kurzweil's estimate. The "modern" operating system performs actions that are completely trivial to the human brain.
The modern computer is designed inherently differently than the human mind, and we only understand computers.
Even our most significant breakthroughs in AI is not fooling anyone. Sure CleverBot is better than ELIZA, but it took 50 years, and you think we're gonna have BladeRunner in the next decade? 10 years in software is the difference between Windows XP and Windows 7 by the wealthiest software giant in the world. There is no way this is going to happen.
First of all, the complexity of "brute forcing" the task is probably more like 1000 orders of magnitude more complex than Kurtzweil estimates.This of course completely ignores that fact that you can't brute force, what you don't understand, because you have no way of knowing which of the uncountable number of solutions is the correct one.
Let's say I give you something as simple as a 256-bit hash. With current technology, and without knowing the algorithm used to derive the hash, it will be impossible to brute force. Then remember that Mr. Scientist here is talking about several million bits.
Posted by: HertfordshireChris
|
August 17, 2010 12:36 PM
Let's face it. Anyone who thinks they can design a program that accurately mirrors the way the brain handles intelligence is someone who is a believer in intelligent design (even if they reject Intelligent Design). After all the core philosophy of the stored program computer is that there is an intelligent designer to write the program.
When computers started they were basically automatic calculating systems optimised to handle highly repetitive numeric functions. There was a mad rush to exploit the technology and no time for anyone to do any related “blue sky” research into the design of electronic processing systems that could work well with people. We all know how we have become trapped by the QWERTY keyboard – and the run-away success of the stored program computer may have had an even more powerful, effect on the human mind. In some ways it is analogous to a successful religion that actually delivers the goods. If you have sufficient “high priests” (= programmers and systems analysts) to instruct the otherwise incomprehensible black box they will work miracles for you. Everyone knows that what you have to do is program the computer – as we teach it (brain wash?) the idea at school – and most current university researchers will have been through the mill.
There is the story of the Englishman who got lost in Ireland and asked a local yokel the way to Ballymony. The yokel thought for a bit and then said, “If I wanted to get to Ballymony I wouldn't start from here.”
If you go back to first principles and ask the “blue sky” research question – “what do we require of an electronic information processing system that works well with people” you would take one look at the way the stored program computer works and say “If I wanted an information processing system that worked well with people I wouldn't start from here.” Computers were designed to handle tasks which people find difficult because they are so far removed from the environment in which human intelligence evolved.
Forty years ago I found myself working on the architecture of an alternative architecture – which at the conceptual hardware level handled sets and partitions of sets, rather than integers and arithmetic operations. The underlying idea started a thousand miles from conventional brain research or artificial intelligence studies – I was designing the human interface of a sales accounting system involving 250,000 customers and 5,000 product lines!. It is not appropriate to go into great detail here but the “intelligence” of the system resided in the way the items of information were linked together. There was NO application program, but instead a general purpose “decision making unit” prowled the knowledge base (along the links) assessing whether a small number of “highlighted” items (equivalent to the short term memory) were true, and changing them as appropriate. It should be noted that the system relied very heavily on recursion, with one simple “universal” mechanism being used again and again.
You probably won't have heard of this because the philosophy was incompatible with the philosophy underlying the stored program computer “religion” and all requests for funding these clearly heretical ideas were turned down, and I abandoned the research as the result of shere exhaustion complicated by a breakdown after a family suicide.
Recently I decided to go online to see what everyone had been doing in recent years and I suspect my model would map well onto neural nets. For instance the recursion of my model could be extended to handle information processing at various levels, and to represent further ways of linking. The “decision making routine” could be mapped onto parallel working on a network of cells which also store the information – becoming simpler in the process. Because my approach need no pre-programming it is possible that the model could provide a mechanism for each child to “evolve” his own intelligent knowledge base. It may well be that if you take the approach I used one could model the decision making process underlying intelligence with very small number of options per cell. If you are being pedantic this would be multiplied by the number of cells in the brain – and there may be variations between cells of different types
OK, I may be wrong - but don't dismiss my ideas because you are convinced that computer programmed “intelligent design” models are the way to explore how the brain and “intelligence” work. Even if you think you think “neural nets” I get the impression that most researchers have been brain-washed by the “success” of the stored program computer. May I simply suggest that thinking “program” to explain how the brain works is attempting to climb the “Mount Improbable” of intelligence – because you are looking for an Intelligent Designer who does not exist. There might be a far simpler, and evolutionarily more acceptable way round the back by starting from the assumption that a humans as a species, and the growing child, start from a state of ignorance about the laws governing their environment.
Posted by: https://www.google.com/accounts/o8/id?id=AItOawkYaU2mqAkf9EtKy1VPdBJjRlL0gv9TBac
|
August 17, 2010 12:38 PM
A number of other commenters have already pointed out one major flaw with this line of reasoning, but I'd just like to point out that the other major flaw is that Moore's "Law" doesn't hold, and we already hit the practical quantum limits of conventional process shrinks several years ago.
Posted by: m3kw
|
August 17, 2010 12:40 PM
Is very easy to say why it is not possible from the little info you know, the info you were given was not detailed at all, and then you attack it using many assumptions that you don't even state.
Posted by: Matt "Nora" Penfold
|
August 17, 2010 12:41 PM
Try looking at it this way.
DNA does not contain the instructions on how to make a brain. It contains instructions of how to make the infrastructure that will make a brain. That infrastructure will then take data from DNA and the environment it finds itself in and use both when making a brain.
Posted by: jfl5926
|
August 17, 2010 12:43 PM
Speaking of kooks ... there's 10 minutes of my life I'll never get back reading this piece of blog tripe.
Posted by: pinkboi
|
August 17, 2010 12:44 PM
Well, enough programmers have already stated why his comments are mistaken (even though computer science is the one field Kurzweil does understand) so I needn't rehash what others said. I will add, though, that a lot of times, we programmers think we're smarter than we are. You can't be a programmer if you're dumb, but when you work on the computer, you get a lot of reward for just a little bit of work compared to other sciences, even if you're doing low-level programming, but especially if you're doing high-level programming (any old fool can write a Python script). I'm studying biotech just so that I don't fall in that trap and actually understand how real science works, computer science being applied mathematics, not true science.
I will say that I'm glad I'm not the only person who's not a big fan of Kurzweil. Of course my grudge against him has a lot to do with his work in wavetable synthesis. Ick.
Posted by: Brownian, Most Vicious & Petty of Pharyngulites
|
August 17, 2010 12:45 PM
$50 says you've never done much with any given ten minutes of your life, so...
Posted by: Matt "Nora" Penfold
|
August 17, 2010 12:46 PM
He could have a wank, and still have 9mins 30secs left.
Posted by: drdrang
|
August 17, 2010 12:56 PM
Here's Edsger Dijkstra on computers and intelligence:
Posted by: Brownian, Most Vicious & Petty of Pharyngulites
|
August 17, 2010 12:56 PM
Plenty of time for him to show up at a blog he considers a waste of time and waste even more time commenting his displeasure.
Posted by: KG
|
August 17, 2010 12:57 PM
hanspeterNo: it's thinking that there is a "design of the human brain" that is creationistic. The adult (or even infant) brain is the result of a developmental process, necessarily involving complex interactions between DNA, complex cellular machinery, and a complex external environment. The developmental process itself is the result of billions of years of natural selection (basically, trial and error). No design, and nothing much like design, anywhere.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 1:11 PM
Hmmm. Looks like this made slashdot. This could be bad...
Posted by: Paul
|
August 17, 2010 1:23 PM
I think being a Kurzweilist is like waking up on Christmas to celebrate the birth of robot Jesus and pray that he'll keep us from ever having to halt existence.
Posted by: nelc
|
August 17, 2010 1:23 PM
Comminista @17: although the discussion's moved on, I'd like to point out that you might be assuming a constant resolution over the whole visual field. Considering rods & cones together and being a trifle inexact, the resolution is about 1 arcmin at the fovea, falling to around 5 arcmin at around 25° radius, and then shallowly falling further to around 10 arcmin at 60°.
In other words, that 600 dpi res at the centre of our vision falls rapidly to a weak 120 dpi at about the width of a newspaper page at a comfortable reading distance, then to a feeble 60 dpi at the edge of our vision. This is one reason why we have to track our eyes along each line of text, rather than mentally photographing each page as a whole.
I haven't got my math head on today, so I can't work out how many retinal cells that is, but fortunately someone's already worked it out to about 7 million. So, probably equivalent to 7 million pixels for one eye, doubled if we want to count each eye as a separate imager, or maybe 10 if we integrate the separate images. I think Kurzwiel's in the right ballpark.
Posted by: nelc
|
August 17, 2010 1:26 PM
Me @137: In the right ballpark as regards eye resolution, that is. I don't think his ten-year prediction on copying brains holds much water.
Posted by: Skeptic Tim
|
August 17, 2010 1:29 PM
It appears to me that ideas like the possible existence of artificial "minds" - (perhaps I should use the word "souls" here; these two words seem to be synonymous.) - or of the possibility of "uploading a mind to a computer" both require that we accept an unstated hypotheses; - i.e there exists an entity called the "mind" (or soul) that is independent of its host and can, therefore, be transferred to an arbitrary host like a computer, or to another biological entity, and still be identified as the original mind (or soul). - I find it very difficult to accept this hypothesis, and any of its corollaries, until someone can present good evidence of the validity of the implied hypothesis!
Posted by: mvanbebber
|
August 17, 2010 1:30 PM
PZ begins the article with a straw man argument of Kurzweil's position. It isn't nearly so one sided or simple. First, Kurzweil says that there are 3-4 methods of reverse-engineering the brain, and another method by which we could achieve human-level AI without reverse engineering it (machine learning.) We could also reverse engineer the brain from brain-scans, which exponentially increase in both spatial and temporal resolution. Henry Markram and IBM's Blue Brain project in Switzerland has taken the approach of modeling the entire cortical column, from dissection of real brains, which is accurate to the level of every single cell and every single dendritic connection between all of them. Interestingly, the leaders of the Swiss team also claim that they will have completely modeled the human brain in 10 years. All of these methods are proceeding at an exponential rate, as Kurzweil has predicted, and as appears to be a law of the universe.
The main crux of PZ's argument is "these protein interactions are too complex, therefore it will be impossible to model the brain in 10 years." Kurzweil talks extensively about the "level of emergence," levels of detail needed to simulate a process, and points out that with other emergent properties, you don't need to simulate the details of the lower level processes. For example, you can perfectly describe the physical interactions of two objects colliding with each other without describing: 1. The molecular bonds of each object 2. The atomic bonds 3. The subatomic bonds and 4. The quantum indeterminacy of each atom that makes up each molecule that makes up each object. You just need to know the emergent properties, WAY above the level of detail that PZ is talking about.
Using PZ's argument, you would have to know the protein interactions of each cell in the eye in order to reverse engineer it. But the eye-brain vision system HAS been reverse engineered with nearly human-level accuracy already, using properties of emergence, and realizing that it is the optics of the eye that is the only important detail, not the underlying molecules that make it up.
To claim that Kurzweil is pseudo-scientific is patently absurd and he knows it. Talk about the pot calling the kettle black! Take a look at both men's resumes and see who is more of a scientist, it will be pretty obvious.
Posted by: Buster
|
August 17, 2010 1:32 PM
Clicking on Buster's nym takes you to a page about George W. Bush. Very appropriate - two fuckwitted godbots.
Say what you like about George W. Bush. Some people thought he wasn't that smart but George W. Bush was an actor not a thinker!
Liberals hate him because he didn't live an intellectual life of subtleties and nuance. He lived a black and white life knowing good from evil.
But being a liberal is like waking up on Christmas Day (or maybe Multicultural Day) and running to the bedknob to find someone stuck a nuke in his sock and he can't figure out who would be so nasty to him/her.
There are bad people out there.
Do take care now!
Posted by: Matthew C. Tedder
|
August 17, 2010 1:34 PM
The criticism seems valid but the tone was not recommendable. It's better to guide people with energy on a subject to reality. None of us are there 100%, in any case. Many things we think we know in biology gets overturned on a fairly regular basis. The energy and desire to explore is commendable, even if he's way off the mark.
I myself have developed a more biologically valid neural simulation that translates environmental efferences into something similar to spatial/temporal/abstract clay. In other words, the neural substrate richly classifies experiences (into abstract objects and their inter-relationships) and reports on which experiences are being recognized currently, at what levels of confidence.
I parametrically simulate all of: both short and long term potentiation from digital signal frequencies, axonial growth and destruction, formation of excitatory or inhibitory receptors, as appropriate (based on source neurotransmitter type or if it is a correlate verses anti-correlate interconnection). I have rich quantities of peer-reviewed literature to support all of this.
I am now trying to use this model (as first principles) to functionally reconstruct the basal ganglia + some artificial cerebral substrate.
One might wildly but properly argue, given the right working definitions, that the result of this might be a conscious, sentient mind.
Posted by: Steve LaBonne
|
August 17, 2010 1:36 PM
And I say there are fairies in the back garden. Argument by assertion is fun, isn't it?
Posted by: dnebdal.myopenid.com
|
August 17, 2010 1:38 PM
@KG :
I know the approaches are quite different - I didn't quite think through the implications of the word "facsimile", and was thinking more along the lines of "seem comparable from the outside if treated as black boxes". Apologies for that sidetrack.
Posted by: daveau
|
August 17, 2010 1:44 PM
@141-
k'bai.
Posted by: Gregory Greenwood
|
August 17, 2010 1:45 PM
I can certainly see the attraction of attempting to reverse engineer the brain, if only to grant us greater insight into how our consciuousness operates, and as a means of putting the whole mind/body dualism thing to rest once and for all.
I personally do believe that it will one day be possible to create an AI advanced enough that it breaches the 'sentience' barrier and becomes self-aware. I have no idea of time frame (though ten years seems extremely optimistic), and I think it very unlikely that any such 'super-AI' would bear any great resemblence to a human consciousness. It would most likely be very alien indeed, and would have to learn things like social mores and conceptualisations of morality over time, if it ever embraces any aspect of human culture at all.
It may also be possible to create an AI that exactly mimics a human consciousness, perhaps by copying an existing brain, perhpas by creating a system that mimics all the complexities of a human mind and then develops its own identity, but I imagine that would prove even more complex, and a period of socilaisation would still be required, as it is for, say, a child.
Of course, either kind of artificial sapient lifeform would throw up all kinds of ethical dilemmas as to its rights and status, and no doubt a whole lot of reactionary 'Skynet syndrome', where people panic that 'teh clevur roobots will murdah us all!'
Obviously, this stuff is great for sci-fi authors, with characters like Data being an ideal means of discussing the social exclusion of disenfranchised minorities in a futuristic setting. However, my enthusiasm for sci-fi does not cause me to lose my head over this. I think it will happen, but I am in no position to say if it will be in 10, 50 or 100 years time.
Posted by: mvanbebber
|
August 17, 2010 1:46 PM
@Steve LaBonne:
"And I say there are fairies in the back garden. Argument by assertion is fun, isn't it?"
I don't like doing peoples' research for them and spoon-feeding luddites, so please read Kurzweil's book and check out the white papers about the Blue Brain Project, etc. Sufficed to say, these assertions, in the form of predictions, have been worked out in painstaking detail and are not blind guesses. That you don't know about it is irrelevant.
Making baseless assertions is fun, isn't it?
Posted by: lasse.j.hansen
|
August 17, 2010 1:47 PM
Funny, lots of similar articles were written back in 1990, who were the fools back then?
People tend to forget about the accumulative nature of both research and technology, and while 2020 may be a bit too big of a PR stunt, I don't think we should underestimate the exponential nature of our technology.
YOUR understanding is insignificant and miniscule compared to the accumulated work of the human species, and naysayers have been put in their place more than once by Mr. Kurzweil.
Posted by: matthew.avant
|
August 17, 2010 1:48 PM
Heavier-than-air flying machines are impossible.
- Lord Kelvin (1824-1907), ca. 1895, British mathematician and physicist
Posted by: Steve LaBonne
|
August 17, 2010 1:49 PM
That you and Kurzweil don't know the abject stupidity of claiming to know how to "reverse engineer the bran", because you understand basically nothing about what that would entail, is not MY problem. Really, try talking to some neurobiologists and developmental biologists, not just computer types.
Posted by: Wise Bass
|
August 17, 2010 1:49 PM
Great - you've just made a map of the brain. Now how do you duplicate and simulate the interactions in that map? Your electronic substrate is going to be very different from the biological substrate of the mind you're duplicating, and you have to work that in somehow.
That's the thing - we're nowhere near the point where we even understand what the "basic principles" would be for emergence.
Posted by: Midwifetoad
|
August 17, 2010 1:52 PM
I'd be impressed if Kurzweil could produce a program that could read written and printed text and transcribe it to an ASCII text file, overcoming problems like smudges, spots, graphics, pencil marks and such, at least as well as an average human.
That's one of the businesses he's in. When he masters his own domain, I'll pay attention.
:)
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 1:53 PM
I thought that was Reagan?
I thought Bush was a failed businessman, not an actor.
If he knew the difference between good and evil, why were his actions evil? He has no excuse.
And being a neocon is like waking up on Christmas Day to find a GI Joe in their sock, and being frightened that it's really a terrorist.
Posted by: Paul
|
August 17, 2010 1:55 PM
Why do so many Kurzweilites not seem to grasp that there is more to technology than multiplying transistor values? It doesn't matter if your technology scales exponentially if you're discussing modeling a phenomenon that is nowhere near fully understood. Even if your transistor density keeps doubling (which has already halted, but let's pretend it hasn't), that does you no good in "simulating the brain" if you don't properly understand the required environment and actors that lead to the emergent phenomena you want to model.
But no, Moore's law! Robot Jesus is coming in 10 years! TECHNOLOGY!
I didn't get along too well with the other Computer Science students...
Posted by: Flex
|
August 17, 2010 1:55 PM
mvanbebber @140 wrote
Which is only a small part of the problem of simulating the mind. Modeling a static tree doesn't automatically tell you how its leaves will move in the wind.
Which isn't to say that it won't happen. I believe that it very likely will. Tedder at #142 seems to be on the right track, from my very much a layperson's view. However, I'm not yet convinced that we have found all the interactions possible between cells in the brain, which severely limits not only our understanding of mind, but also limits our ability to decide that certain processes of our brain are not important in order to simulate a mind. Many processes of the mind probably are unnecessary for consciousness, but which processes are they?
Posted by: dgerard
|
August 17, 2010 1:57 PM
Dawkins these people piss me off. Sceptics are adept at making really quite fetching mincemeat sculptures of religion, alternative medicine and the new age, but we need some serious attention paid to the transhumanist/singularist/cryonicist belief cluster. Because these are smart people, they are likely our friends, they share a lot of our notions and they are proving that the main use apes with delusions of grandeur like ourselves put intelligence to is being stupid with far greater efficiency.
Obligatory RW plug: Cryonics. I was actually neutral-to-positive on the subject until a friend started looking seriously into spending $120k on freezing his head and I started looking seriously into what he was getting into. And goddamn, it's woo all the way down. Woo by people who are ridiculously smarter than you or me and use it to be dumb. How do you fight that sort of woo? Piece by piece, of course. So I have to learn the bollocks on its own terms to take it down (at which point you see goalpost-moving, reversal of burden of proof, etc., all the things apes with delusions of grandeur do so well). And it's just AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
PZ, could you please dismantle cryonics at your leisure? Thank you.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 1:58 PM
Lord Kelvin was often an idiot. Birds are heavier than air; yet they fly all the time.
But which heavier-than-air approaches worked out? The ones that attempted to mimic the flight of birds, with flapping wings, or those that used the science of aerodynamics?
So, I'm gonna have to go with "analogy fail," here.
Posted by: DiscoveredJoys
|
August 17, 2010 1:59 PM
Simulating the function of a brain could be done in several ways (in principle). Simulating each cell in the brain (and associated sensory apparatus, blood chemistry, extended phenotype stuff too) is the hard way of doing things.
The easier way would be to simulate the functions, not the biology. Unfortunately we are still a long way from understanding the functions of the brain well enough to know when we can simulate them realistically.
And there again, why should we want to? We've always wanted robots in the home, but we don't make generalist anthropomorphic devices, we make washing machines, and dishwashers, and computers.
Similarly you could argue that we already have 'devices' that simulate some aspects of brain function. We call them search engines - input a question and it is parsed and the most likely answer is returned. Some people argue that these are better metaphors for the brain than computers themselves.
I, for one, welcome my Google Overlords.
Posted by: davem
|
August 17, 2010 2:03 PM
Kurzweill is obviously wrong, but boy, are there a lot of pessimists and naysayers on this blog today. My prediction? That 2 weeks after we discover how that brains in general work, we'll be running simulations of them. We can start with portions of ant brains - no need to go for the most complicated one first.
They won't need to simulate 10 to the power of a zillion neurons - once we work out how the neuron interaction works, we can simulate what it does in lines of code.
The artificial brain's neurons will be the data not the program. We don't need to model the way the proteins work with each other - we just need the output signals, given the input.
The first human brain simulator won't be HAL of 2001 fame; it will probably run 1000 times slower, until we work out new hardware to run it on. My completely uneducated guess is that when we find out how brains work, we'll say "wow, so complex, yet so simple". The program will have to say 'Do lots of simple things simultaneously to achieve a complex answer'. No need for a million lines of code.
Posted by: rhys.rhaven
|
August 17, 2010 2:06 PM
Concerning "To simplify it so a computer science guy can get it"
I just wanted to say, as a computer science guy, fuck you neighbor. (Via FYN-FYTN Protocol)
PZ, come drinking in Chicago!
Posted by: Steve LaBonne
|
August 17, 2010 2:07 PM
I don't think you have clue one about how much complexity is subsumed in that apparently simple phrase, "how the brain works". It's not like there's some simple gimmick waiting to be discovered.
Posted by: BrianX
|
August 17, 2010 2:07 PM
Kurzweil (along with Eric Raymond, Stephen Wolfram, Bill Joy, and likely Edward Teller) strikes me as being a prime example of the Mensa mentality -- there's no question that the guy is smart, and he's been remarkably influential in the digital audio and computer reading worlds (there's a reason so much of Stevie Wonder's 80s work was done with Kurzweil gear). But a lot of such people (and I could easily have been one of them) tend to assume that being demonstrably more intelligent than most means they're not likely to be wrong about ideas they get into their heads. When someone like this gets into something like the Singularity, they become so convinced of it that they're willing to use themselves as the authority in an argument from authority, and since they're seen as Smart People (tm), they get taken more seriously than the raving nut on the corner.
(I think, by the way, this explains why so many computer geeks are Libertarians -- hammer-and-nail syndrome leads them to think that society itself is a computer-like construct and don't stop to analyze the analogy before jumping in whole-hog. Witness Eric Raymond's "error cascade" claim about how climate scientists could be wrong without them all being corrupt -- he fundamentally misjudges the nature of science and evidence because he only ever takes his nose out of the computer world to go to the firing range.)
Posted by: LuchinG
|
August 17, 2010 2:08 PM
So, is a folding bike a better metaphore?
See this:
http://www.gizmag.com/airnimal-chameleon-hard-case-folding-bike-trailer/11494/picture/76939/
You can´t explain how the chain ring works without how it´s shape depends on all the other parts. It's eficiency and performance is limited by the folding funtion. Every part in the bike influences the shape of the chain ring.
Posted by: Flex
|
August 17, 2010 2:10 PM
DiscoveredJoys @158 wrote,
I'm waiting for an Electric Monk, with or without the horse.
Posted by: First Approximation, L'esprit de l'escalier
|
August 17, 2010 2:11 PM
Yes, because someone was once wrong about airplanes a century ago we should accept every crazy idea we hear today.
Posted by: mvanbebber
|
August 17, 2010 2:12 PM
@Steve LaBonne:
"That you and Kurzweil don't know the abject stupidity of claiming to know how to "reverse engineer the bran", because you understand basically nothing about what that would entail, is not MY problem. Really, try talking to some neurobiologists and developmental biologists, not just computer types."
To claim that a blind naturalistic process can create brains, but not teams of hundreds of thousands of motivated, directed scientists, using the powers of exponentially growing experimental and manipulative power, is the epitome of luddism. Unfortunately, it looks like we have some closet dualists here.
Posted by: Brownian, Most Vicious & Petty of Pharyngulites
|
August 17, 2010 2:12 PM
The white being cocaine?
You're clearly one of the stupidest people I've ever encountered here, Buster.
You're one of them.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 2:13 PM
I'm not sure you're seeing pessimism. Most of the responses are saying things very similar to what you're saying.
The original topic was on the ridiculousness of deriving from DNA a model of the brain. This was one of Kurweil's notions. It just doesn't make sense, and indicates Ray's a little bit out of his depth, as brilliant as he may (or may not) be.
But here's your key sentence:
It's the bolded part that is of interest. Since we're still a ways off from understanding how the brain works in general, we're a ways off from simulating it.
We don't even know how much we don't know about the brain. Coming up with a timeline (say, 10 years) is kind of jumping the gun.
I know I'm not pessimistic about real AI, nor am I pessimistic about creating a model of the brain; but expecting that model of the brain to be the basis of our AI is just a little bit silly. Just like ornithopters are not the best way to fly, using a natural model such as the brain for our AI research is perhaps not the correct path.
At least, that's the way I see it.
Posted by: Greylander
|
August 17, 2010 2:14 PM
PZ, this rant really shows the limits of your knowledge and comprehension of matters computational. You think the computer scientist are just ignorant of all the complexity of biological process which you as an expert are aware of. What you fail to grasp is that computer scientists (the good ones, anyway) are experts at understanding what complexity really is, and is not, and which complexities matter when it comes to solving problems or constructing simulations computationally. Your skepticism is highly understandable and fairly typical for a biologist from my experience, but it stems from ignorance and/or lack of comprehension.
I am disappointed in those claiming a CS background who are chiming in to agree with you.
Regardless of how the information in the genome is extracted, regardless of how complex phenotype may be, with all of it’s myriad proteins and their interrelationships, there are only about 50 million bits of efficiently coded functional information in the genome.
“What does this not mean?”
This does not mean that we can just look at the genome as a “blueprint”. The word “design” is being used loosely/informally in the article. To “interpret” what is coded in the literal genome or in the hypothetical efficiently coded version of the genome, you have to process that data with the appropriate machinery.
This is analogous to a self-extracting, compressed data file: you have to have the right hardware to execute the program to expand itself from that .7 GB compressed movie to 100GB of raw video (ballpark: 1000x1000pixel x3bytes/pixel x 30 frames/second x 2 hours... give or take). The size of binary program required to reconstuct that 100GB stream of video is about .7 GB (most of which is just compressed data with a much smaller portion that actually executes the decompression routine, but in principle there would be no distinction between data and exectuatble within a maximally compressed self-extracting file). Looking at the compressed file you will not see the movie’s plot, the relationships among characters, or what the main character looks like. But the information is there for all that to be reconstructed.
“So what does 50 million bits of efficiently coded information in the genome mean?”
This literally means that a self-extracting file, an executable binary program of about 50 million bits, with no additional data, would be able to reconstruct the sequence of a single human genome. Assuming that the genome contains all necessary information (more on this a bit later), then a functional description of the processes of the human body (from conception forward) require only about 50 million bits -- this is a back-of-the-napkin estimate of the size of the required binary program on a machine which utilizes a reasonably compact binary code. A binary program of this size, plus appropriate machinery to run that program (cytoplasm of the fertilized egg), plus a stream of input data (the environment -- uterus and the world beyond) gets you a simulation of a human being.
“Aha!” you say, “what about that ‘appropriate machinery’?”
In a biological organism, the relevant machinery is the cytoplasm of the fertilized egg. This machinery is functionally described by the equations of physics and chemistry (requiring a few kilobytes at most) and also requires data describing the structure of the cell. There is a great deal of redundancy here, and good reason to assume a great deal of the information content is already in the genome. For example, the RHEB protein. Most, if not all of the function of this protein is determined by the laws of physics and chemistry plus the amino acid sequence which is already code in the genome. There may be some required relevant information in the specific location/orientation of this protein, relative to other proteins and molecules, in the cytoplasm of a newly fertilized egg -- but I think you will agree that the cytoplasm can withstand a great deal of wiggling and jiggling and all those proteins and other molecules will just will wiggle and jiggle themselves back into a healthy functional state just following the laws of physics. So there is not much required information in the exact details of the positions and orientations of all the molecules in the cell.
It is important here that I clarify how I am using the term functional in this context. By functional information I am referring to the information required to describe the rules, laws, behavior of a system as distinct from data required to describe the detailed state of that system at any given time. The amount of data required to describe the position, orientation, and momentum of all the detailed structures of just a single cell is going to be ten the the power of twenty-something-ish, and for an entire human body, ten to the power of something-more. The value of “something”, “-ish”, and “more”, are dependent on the actual level of detail of your simulation (structural, molecular, atomic, sub-atomic scale?).
The salient point is that the binary program required to run this simulation is going to be in the neighborhood of 100 million bits (we’re dealing with ballpark figures here, might as well go for the nearest power of ten). It is safe to say that the a detailed simulation of the cognitive processes in the human brain, modelled somewhere between the molecular and cellular level, will take a significant fraction of the same program. 25 million bits is not unreasonable. (with these kinds of ballpark figures, there’s not much difference between 25 million and 100 million anyway [if you are uncomfortable with my flippant attitude as to the difference between 25 million and 100 million... you need to develop your mathematical intuition regarding estimates and logarithmic scales]).
To get the “million lines of code” we are presumably talking about human readable code. Here Kurzweil is just applying a rough rule of thumb. Of course any “exact figure” will depend greatly on the programming language used, and the skill of the programmers involved. The point is not whether it will take a million or a hundred thousand or ten million lines of code. The point is that it is well within the boundaries of what is doable within a matter of a few years.
To model the information processing capabilities of the brain, it is highly unlikely that it will be necessary to simulate all the underlying molecular machinery, most of which is just about keeping each individual cell alive and functioning with no direct bearing on the information processing being done by the brain as a whole. We can probably move the simulation up to the level of neurons, synapses and keeping track of activation levels and the concentrations of various neurotransmitters inside each cell and in the gap at each synapse. At this level, the program required to do run the simulation will be quite a bit smaller -- maybe by a factor ten (-ish).
“How much data will be needed by that program to represent the information processing state of the brain? “
Well there are around 100 billion (10^11) neurons with something like 10^15 synapses. Suppose the internal state of each synapse requires something like 10K... hell let’s say 100Kb (10^5 bits). There are far fewer neurons than synapses so that, unless the internal sate of each neuron (aside from it’s synapses) requires a billion bits, the estimate of data for synapses is all that is significant here. So we are talking about a pretty high-ball estimate of 10^20 bits or about 10^19 bytes. This is a high-ball estimate, as I think that 10KB per synapse is probably a ridiculously high estimate as to the amount of data required to describe the informational state of a synapse.
“How fast would we need to process the simulation data to simulate a brain in real time?”
Figure we need to update the entire state of the brain 100 times per second. 1000 times per second if you really want to be a hard-ass about it. Figure that a “FloPS” refers to a basic math operation on a “word” of data (typically 8 bytes nowadays). This suggests something like 10^21 FloPS required.
“Wow, those numbers are a lot bigger than the 10^16 bytes and 10^17 FloPS mentioned in the article.”
The preceding estimate is for the entire brain. They are also high-balled. My estimate of 10KB per synapse is pretty wack. Bring that down to to 1KB. Lets bring the update rate to a more reasonable 100Hz. Finally Kurzweil is looking at just the cerebral cortex, not the entire brain, for purposes of this estimate. That’ll save us about another factor of ten. So reduce my numbers to... 10^16 bytes of data and... 10^18 FloPS. Kurzweils numbers are on the optimistic end of perfectly reasonable estimates.
The hardware and software capacity to simulate the cognitive functions of the human brain could easily be less than a decade away.
“I’m just not convinced. You say you are high-balling your numbers, but I think you are still waaaaaay underestimating.”
Fair enough. There is room for a great deal of discussion here. Artificial intelligence enthusiasts used to try to estimate the complexity of the brain by treating each neuron as a single bit processor, ignoring all the data stored in the strengths of connections (size/types of synapses/amounts of neurotransmitters) and also ignoring the possibility that the relevant internal state of a neuron may be much more complex. Even the “on/off” of “excited/not excited” neurons which seems like a 1-bit thing ignores the possibility that the timing between on/off may contain relevant analog data equivalent to several bits of “floating point number” with each transition.
Could I be making a similar blunder? Could it be that the truly interesting information storage and processing occurs at the molecular level, requiring our simulation to model the detailed interactions of individual molecules? Sure. Maybe. If so we would have to Bump my estimates up by a factor of a million or so. Hell, maybe the really interesting stuff about human intelligence will actually require modelling quantum processes at the sub-atomic level. That would probably bump up my estimates by a factor of a trillion or quadrillion or even a zillion. But I think on the whole, present evidence strongly suggests that the cellular/synaptic level of detail is about right.
“OK, suppose I accept your estimates. Even if there are just 50 million bits of functional information required to describe the processes of the human body, we still don’t really understand how to interpret those 50 million bits until we actually see how all those proteins interact in all their complicated ways. We don’t know the rules by which the cellular machinery interprets those 50 million bits, and it is going to take a lot longer than a decade to reverse engineer it all.”
This is a stronger argument than trying to quibble about the order-of-magnitude estimates on the computational complexities involved. I’m not going to get into it in detail in this post, which is already long enough. But I will suggest that you look into the state-of-the-art and prospective rapid advances of “lab on a chip” technologies. Combined with advancing computational capacity (including out ability to simulat ever bigger and more complex molecular interactions) and other automation of bio-chemistry lab work, our capacity to catelog and “reverse-engineer” the molecular machinery of life is going to rapidly advance over the next decade.
Having said all this, Kurzweil is, I think, overly optimistic. I give it three decade at the outside for the first detailed real-time simulation of the cerebral cortex. Call me a pessimist.
“Kurzweil is a nutjob and so are you.”
Yeah. Whatever. You can have this argument with your laptop in 2050.
Posted by: Steve LaBonne
|
August 17, 2010 2:15 PM
Not when the basic knowledge such a team would need is nowhere near existing, and the problem itself is so complex and multifaceted, as well as poorly understood, that a brute force "big science" approach is little help in accelerating progress. How come cancer hasn't been "cured" yet?
Add science itself to the list of things you and Kurzweil simply don't understand.
Posted by: Luke
|
August 17, 2010 2:17 PM
Ray may be stupid about the ten year timescale, but he at least has made cryonics arrangements. It doesn't matter if it takes 300 years to crack the brain, if his essential neural information is preserved.
But yeah he sounds like a kook with that analogy to computer code. DNA runs on the architecture of the universe, which is not trivial to simulate.
Posted by: BrianX
|
August 17, 2010 2:18 PM
mvanvebber:
The blind naturalistic process also required something on the order of three billion years to get that far, and we still have only a very rough idea how the process even began. Surely it would be unsurprising, even optimistic, to expect it to take a couple hundred years for humans to get that far?
Anyway, I always thought the best representation I've seen so far of an AI-enhanced future was "My Life As A Teenage Robot" -- despite increasing technology and connectivity, people still being people and living more or less the same. (Even if some of your classmates are made of plastic and silicon.)
Posted by: Steve LaBonne
|
August 17, 2010 2:19 PM
The Kurzweilites are as dumb and tiresome in their own way as Randroids. (And I'm sure they're overlapping groups.)
Posted by: Sven DiMilo
|
August 17, 2010 2:19 PM
Too small. The first brain to be simulated from the neuron-up level is quite likely to be this one; there's a pretty good start made already. It's still going to be a long time.
Posted by: hibob
|
August 17, 2010 2:20 PM
@Matt Penfold:
That definitely describes the problem Kurzweil will have with his prediction. But that one clause "The design of the brain is in the genome" is still entirely true. Any design used by people (blueprint, CAD diagram, whatever) only encodes for a fraction of the information that's "in" the final product, the rest of the information is assumed. To make a model of the object that yields useful predictions, you need a lot more information than is contained in the design. Computer programs depend on a trivially simple system (OS and firmware), so the "rest of the information" is relatively small: Kurzweil jumped to the same conclusion about the genome and the brain.
Blueprints people use for physical objects include pictorial maps that describe the entire structure, but leave out what wood, steel, aluminum, IC chips, etc, actually are and how they function. If you include that huge amount of engineering information, you can go from a design to a useful (predictive) computer model, but it's not trivial. Genomes (via the interactions of the encoded proteins) include relationship maps that guide (encode) the construction/self assembly of a particular nervous system. The "rest of the information" needed to go from a genome to a useful model of a brain is an understanding of biology at a level we're just beginning to scratch at.
So I'm happy with saying the design for the brain is in the genome, but just as dubious about someone talking about an ab initio model of the brain starting from the genome.
Posted by: BrianX
|
August 17, 2010 2:21 PM
Luke:
Considering things like ice crystal growth and information-death are by no means solved, cryonics is snake oil, plain and simple. That the people who support it can't admit this to themselves does not make it any less so.
Posted by: Steve LaBonne
|
August 17, 2010 2:22 PM
You shouldn't be. I like Dawkins's analogy that the genome is a set of recipes, not a blueprint. Those are very different things.
Posted by: BrianX
|
August 17, 2010 2:24 PM
Greylander:
"I am disappointed in those claiming a CS background who are chiming in to agree with you."
Some of us did pay attention in other classes and realize that a Turing machine is not a very complete map of reality.
Posted by: First Approximation, L'esprit de l'escalier
|
August 17, 2010 2:25 PM
Sheesh, it's not that it can't be done (it quite likely can). It's that it won't be done on the timescale or the biologically ignorant way predicated by Kurzweil. There are a lot of hurdles to overcome and Kurweil answers just seems to be
Magic man done itfuture technology will magically do it (and very soon).Posted by: poke
|
August 17, 2010 2:26 PM
Kurzweil's assertion made me laugh out loud. Beside his total misunderstanding of genetics he clearly knows nothing about the development of the brain. It's highly dependent on experience (especially prenatal experience when it comes to the basic structures). You'd have to simulate the body and its environment to get your simulation to be anything like a normally functioning brain. It would surely be easier, although still utterly implausible, to "reverse engineer" a mature brain.
Posted by: mvanbebber
|
August 17, 2010 2:27 PM
BrianX:
I disagree because there are so many examples of scientific problems being grossly overestimated because people fail to take into account that the tools used to experiment with and manipulate the natural world increase exponentially. I remember listening to NPR in the early 90's and the topic was the genome. Two separate experts (one doctor and one biologist) estimated it would take between 100 and 1,000 years to complete the sequencing of the genome and that "we will not see it in our lifetimes for sure." Well, they failed to take into account that the amount of genetic code was doubling each year and it was completed in 7 years.
About your second point, I agree that there will be biological humans alongside AI (in whatever form they take) just as there are Amish alongside us humans deeply involved in a human-machine hybrid civilization. I haven't seen that movie, but will check it out.
Posted by: Steve LaBonne
|
August 17, 2010 2:28 PM
So again, why don't we have a "cure for cancer"? Extravagant resources have been thrown at that goal for many years now.
It's just not that simple.
Posted by: KG
|
August 17, 2010 2:30 PM
- mvanbebberTotal crap. There is no machine vision system with anything near the broad range of capabilities of the human visual system.
Posted by: danielm
|
August 17, 2010 2:30 PM
I see a straw man argument based on what a reporter (who doesn't know his stuff) reported was said to him by a scientist (who is a specialist in a different field) about something he isn't a scientist about, necessarily leaving out pertinent and perhaps crucially important details in his piece for people who also don't know what they're talking about.
...and then I see otherwise very intelligent folks going on about how a computer scientist who isn't claiming to be a biologist doesn't know enough about biology to speak about something which doesn't actually involve biology.
I mean gee, when you put it that way...
So there's nothing inherently unsimulatable about the brain...so perhaps we can simulate it, given masses of money, some massive increase in computing power and some radical increments and internal revolutions to our current technological level...exactly where is this invisible line that can't be crossed?
Ray's probably wrong about the timeframe - lord knows him and people like him have been saying "X is only 25 years away" for the last 40 years or so...but I have a real problem with this "never! preposterous!" attitude.
So the slug/rat/cat brain stories are PR fluff - well the systems in question are in reality very powerful, very specialized, very useful "fluff" which, whilst not quite yet (sarcasm!) sentient certainly are a good few steps up the ladder of "knowing the fuck what we're talking about" territory.
Maybe, ultimately, the brute-force "running a copy of a human mind by emulating the neurons" solution won't be first, or will prove too inefficient - but damn if there aren't a lot of naysayers who don't know jack doing the equivalent of holding up those placards that read "doom! the end is nigh! repent you AI heathens!"
The time when we knew nothing about how the brain works is long past - they really are building (simple) neurons and neo-cortexes. The problem with this method is that it's a blank slate as far as anything called "mind" goes and is very inefficient on our current hardware - but we're already using it to solve otherwise intractable problems and experiment with technology that without it would be impossible. It's far, far from a boondoggle.
There was a fascinating article about a new kind of chip which actually works the same way neurons work, and using it they could (and probably will) build something far more complex than the human brain in terms of "connections" that uses a fraction of the power of today's supercluster computers to run something like the bluebrain project.
It's blue-sky technology - emerging already - like this which could just prove Kurzweil right.
On the other hand, yeah, he's a really intelligent kind of nuts - but I like him.
Posted by: BrianX
|
August 17, 2010 2:34 PM
mvanbebber:
Your point about the human genome is true, as far as it goes. But the flipside is Fermat's Last Theorem, which Fermat seemed to think was trivial to prove but took 350 years and the invention of two entirely new fields of mathematics. I see a distinct similarity to the Galileo Gambit in your logic.
Posted by: MJP
|
August 17, 2010 2:35 PM
But it's 2 million Java lines of Java code if you Java program in Java.
Posted by: BrianX
|
August 17, 2010 2:37 PM
MJP:
Not to mention the runtime libraries and native-code interface.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 2:37 PM
Right. And then, once we understand the processes that allow a brain to exhibit intelligence, we can code those processes.
That's how we're going to get AI.
And this does not make Ray Kurzweil happy.
Why? Because it doesn't leave much chance of uploading the state of your brain when you are dead. And that's what he's trying to manufacture: the implementation of his immortality. He doesn't just want AI. He wants a simulated brain for a very specific reason.
Posted by: mvanbebber
|
August 17, 2010 2:44 PM
@Steve LaBonne:
"So again, why don't we have a "cure for cancer"? Extravagant resources have been thrown at that goal for many years now.
It's just not that simple."
Your argument is: "We can't do X, so certainly we can't do Y." There is no correlation between cancer research and brain emulation.
Posted by: Paul
|
August 17, 2010 2:46 PM
Take your strawmen home. Nobody is claiming that it is impossible in principle to create or simulate a brain. People are just pointing out that Kurzweil's "calculating" of how much code it would take is pure numerology, with no relevance to how such a project would actually have to operate. They are also pointing out that we cannot create or simulate brains until we have much more knowledge of what they actually do. People are laughing at the nutball's 10 year estimate and shoddy methodology in generating it, not saying brains inhabit some magical supernatural space that can't be replicated.
Posted by: KG
|
August 17, 2010 2:47 PM
- GreylanderDo you really not know that new synapses are growing and old ones disappearing all the time? The brain is not a computer. It is not nicely "layered" into hardware, firmware, operating system and application programs. Functionally, it's a vast tangle of cross-scale interactions - so some of the molecular level details will not matter, while others will be crucial - FFS, look what happens when you put some alcohol or LSD into it. What's more, these cross-scale interactions are not restricted to the brain: they involve the rest of the body, and the physical and social environment. You really have no idea what you're talking about.
Posted by: KG
|
August 17, 2010 2:51 PM
Posted by: Steve LaBonne
|
August 17, 2010 2:52 PM
Uh, no it isn't. Can't answer the question about cancer research, eh? This non sequitur is on the same intellectual level as your other "contributions".
For the record, I'm saying that there are hard problems that can't be solved in a short time even by throwing massive resources at them, and gave an example of one- "curing cancer"- that's orders of magnitude simpler than what Kurzweil wants to do (but whose terms are equally misconceived).
Posted by: BrianX
|
August 17, 2010 2:53 PM
nigelTheBold:
Yes, but would he admit that, even to himself?
Posted by: KG
|
August 17, 2010 2:54 PM
BusterWell, just for a start, he was a liar, a coward, a drunk, a hypocrite, a cheat, an election-thief, a murderer, a traitor and a war-criminal.
Posted by: Paul
|
August 17, 2010 2:55 PM
Please, point to one. One person here who says a functional simulation of a brain will never exist. The matters of faith here lie solely with the Kurzweil-bots.
Posted by: zenstoic
|
August 17, 2010 2:57 PM
Brain emulation is certainly a difficult proposition, although I believe Kurtzweil's objective is to get people to concede that it is theoretically possible, not to buy his 20 year time frame. If indeed it is theoretically possible, that changes all manner of things that we believe about our universe and ourselves.
More importantly, if it is actually feasible then it is such an attractive proposition that it will probably be done at some point in the future. It really is irrelevant whether "the genes are just the data not the code" if it is possible to understand and emulate the code, even if we do not currently understand it.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 2:57 PM
BrianX:
He's admitted that he desires to upload his brain, yes. He even thinks he'll be able to bring back his long-lost pa, using nothing more than DNA and historical information.
That's why he's concentrating on this whole brain-emulation-from-DNA thing, rather than talking about the easier case of straight-up AI.
I just find it funny how many people are rushing to defend his ludicrous (and inefficient) idea. Even if it were feasible, it's just stupid. It's the pathetic gasp of an intelligent mind rebelling against its own mortality.
Posted by: BrianX
|
August 17, 2010 2:58 PM
Paul:
Truth of the matter is, the Kurzweily E. Coyotes around here remind me of that e-book troll from a month or so ago. They're convinced they won the argument but can't seem to be bothered to answer peoples' objections.
Posted by: BrianX
|
August 17, 2010 3:01 PM
nigel:
Considering that there's no clear answer in the nature vs. nurture argument, there's no reason to imagine he's even remotely right about bringing his dad back...
Posted by: KG
|
August 17, 2010 3:05 PM
- danielmSimulating the brain doesn't involve any biology? I think that's the stupidest Singulatarian claim in this whole thread - and that's saying something.
Posted by: liquiddark.myopenid.com
|
August 17, 2010 3:11 PM
What is Kurzweil getting at? He is getting at the basic information that describes the structure of the brain. He is almost certainly not positing a quantum-granular model of the brain because many people believe you don't need that level of granularity to get a working mind. So you have a linguistics problem (what is the basic element of a mind?) plus a computational problem (How do you unfold the elements into a bootstrapping structure that will thereafter continue to process inputs into meaningful internal and external outputs?).
Kurzweil's theories always lean heavily on exponential progress over the next 50 or so years, so if you don't buy into those figures, you're not going to buy into his ideas. Fair enough. But it's hard to argue that we're well past the point where the tools we use at the cutting edge aren't starting to look a lot like stripped-down minds themselves. AI lost every battle, but it still appears to be winning the war in every field of study.
Posted by: troniq
|
August 17, 2010 3:12 PM
A normal child's brain turns out about the same as his parents. Of course the blueprint is in his genetic code. Kurzweil is right, on the macro view.
Posted by: jasonnyberg
|
August 17, 2010 3:14 PM
#184:
Memristors.
Memristors have been doing visual recognition tasks in trivial neural nets, and promise direct physical neuron-like functionality in hardware at densities far greater than where we are with digital transistor logic...
Combine that with Numenta's neocortex architecture that is already capable of learning non-trivial perception/recognition tasks (see Vitamin-D Video) and is currently being re-worked into something much more capacious and robust thanks to some new fundamental insights that dovetail nicely with what we know about real neocortex biology...
All the nay-sayers here, who say we "have no idea how the brain works!", who are claiming that "Moore's law is already dead!", have no idea what we actually do know and what technological leaps are in sight.
Posted by: fordiman
|
August 17, 2010 3:15 PM
Even if we could derive the proteins from the DNA sequences, and even if we could model their behavior, and even if we could simulate all the experiences from conception to consciousness a fetus would experience, we'd have one major problem with hacking the brain.
Evolutionary process does not produce clean, elegant solutions to problems. It produces nasty-but-effective kludges. For a brain, this means miles and miles of physically emergent logical paths that would each need modeled and mapped before we would be even vaguely able to derive qualitative data.
As a programmer, I can tell you why this is. You fix something. You fix something else that maybe makes something else work a little wonkier. The client requests a feature that would normally require an architectural change, but you hack in anyway because the client wants it now. Eventually, you get the time and money to restructure everything and make it all pretty again.
In evolutionary process, that restructuring never occurs. Even if we could read the code to the brain, I suspect that actually understanding what's going on in it well enough to improve on it will be out of our hands for several more generations.
Posted by: Steve LaBonne
|
August 17, 2010 3:15 PM
Again, recipe, not blueprint. Quite different things. There is nothing remotely- as measured in parsecs- resembling a one to one correspondence between parts of the genome and parts of the brain.
Posted by: BrianX
|
August 17, 2010 3:15 PM
KG:
The curious thing is that from what people do know about the brain, the entire analogy to computers as we know them breaks down. There's a book called "The Barmaid's Brain". The title essay (which, honestly, struck me as being a tad sexist but more or less reasonable, especially since this also seems to apply to batshit crazy chess champions like Kasparov, Fischer, and Alekhine) suggests that if you find someone who seems a bit dull but works a job where lots of memory is required, it's possible that they've wound up trading processing power for a much larger short term memory. If that concept of the brain is true, then a graven-in-silicon concept of the human brain is pretty much impossible. Granted, neural nets try to simulate such adaptability to a certain extent, but the only real way to simulate a brain based on what we know now is with some sort of dynamically reconfigurable FPGA chip that can be either a processing element, a memory chip, or some of both, and we don't yet know the techniques necessary to produce such a thing or the algorithms needed to use it effectively.
Posted by: CJO
|
August 17, 2010 3:16 PM
But the eye-brain vision system HAS been reverse engineered with nearly human-level accuracy already
Sheesh. The level of processing happening at the retina can't even be matched by any known algorithm "with nearly human-level accuracy." You may have noticed the dearth of robot-piloted vehicles in your neighbornood. Well, the computational problems in developing autonomous robots able to navigate the real world are largely problems in machine vision.
The human visual cortex, first of all, is not some algorithm-driven pattern matching machine with some kind of index lookup function to compare new input. It is a massively parallel set of heuristics. It is highly interconnected with the other sensory cortices, as well as with unknown myriads of cognitive and memory "modules" that inform real-time visual perception.
In order to say what you did, you have to not understand what it means to see. Notice as you look around you, that you can't "unsee" objects. You are not capable of reducing visual input to the kinds of parameters that an algorithm has to work with. You see a green coffee mug, and that's what it is, irreducibly, not a string of inputs of various wavelengths, and your "eye-brain vision system" made that judgement instantaneously. Show me an algorithm that could "look at" a coffee mug at an oblique angle under various conditions of lighting and determine whether it is approximately round or elliptical, never mind reliably distinguish it from a tumbler, or a teacup, and you'd have one that's far more sophisticated than any existing artificial vision-processing system, and that's just a trivial little problem of the sort that your brain is solving effortlessly, all the time, several times a second.
Posted by: https://www.google.com/accounts/o8/id?id=AItOawmuS6IcbGsNBBYr3oqZj7TtTFgG1RgdVx4
|
August 17, 2010 3:30 PM
Henry Markram
http://www.ted.com/talks/lang/eng/henry_markram_supercomputing_the_brain_s_secrets.html
Posted by: Luke
|
August 17, 2010 3:31 PM
BrianX: Look up vitrification sometime, jerk. Ice crystals are a solved problem.
And information death is questionable even in the case of frozen tissue, much less vitrified tissue. The morphological characteristics of cells don't instantly melt away the moment the heart stops beating.
If you think you can write a detailed critique that addresses the issues of cryonics factually and accurately, go ahead. You can be famous. There's nothing like that in existence currently.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 3:33 PM
From the Wikipedia article on memristors:
Wow. It's so close to emulating a brain.
Now. To get on to your strawman:
Nobody here is saying we won't have AI sometime in the future. Perhaps we'll have it in the near future, say, within 10 - 20 years. That's not what we're trying to get through your thick skull.
No. What we're saying is, estimating the level of complexity in the brain (and the number of lines of code) based on a simple numerical analysis of the information in DNA is not an accurate way to determine the complexity of the task of emulating the brain. DNA is only one part of the very complex chemical process of the development of the brain.
That's it. That's all.
Then, there's the whole utility of creating a brain analogue for anything other than studying how the brain works under various conditions. Attempting to do a direct brain AI is rather silly, when it would be far better to do . . . well, what you are suggesting. Using what we know of the way the brain processes information to create things that process information in a similar fashion, but far removed from the original inspiration.
So why don't you try to address the actual topic of discussion? It's a lot more fun than getting angry at stuff we're not saying.
As for Moore's Law: hardware surpassed software a long time ago. We don't make use of the computational power we have right now. While the processing power has doubled every 18 months, what we do with that power has not. Hell, MS-Windows 7 is about on level with NeXTStep as far as functionality. Sure, it's pretty and everything, but it's not really advancing what we do with computers.
(Google's done a fair job of doing something with all the processing power. Even they are limited to simply searching text really really quickly.)
So, yeah. Until we understand the brain a little better, all the voodoo science and continuously-doubling computer power in the world won't get us closer to actually emulating the brain.
Posted by: jasonnyberg
|
August 17, 2010 3:35 PM
#208:
Hierarchical Temporal Memory.
Posted by: Vicki, Chief Assistant to the Assistant Chief
|
August 17, 2010 3:36 PM
liquiddark:
The mind is not reducible to a "linguistics problem," because not all of what the mind does is done in, or by, language.
It's very easy for the thinking/conscious mind to forget the rest of what's going on in there: but memory isn't a matter of verbally deciding "I will remember that trip to the beach when I was ten" and "bring up my mother's face." Emotions are part of the mind, and while they can be affected by verbal thought, there's a lot more going on there: breathing slowly is not a linguistic act, and neither is low blood sugar or a change in hormonal balance.
(I'm addressing this as a very verbal person; the thesis would seem even less plausible to someone who doesn't think primarily in words.)
Posted by: Greylander
|
August 17, 2010 3:37 PM
@ KG, #191
- Greylander
Do you really not know that new synapses are growing and old ones disappearing all the time?
KG, do you really not know that your points are irrelevant to the question of the computational complexities involved? Yes, of course synapses are growing and changing all the time. If you prefer, think of every potential synapse whether or not it presently exists, as a "connection". Synapses that do not exist are simply have "zero" for all the variables describing the computationally relevant aspects of their state.
Do you know know what it means for a computer to simulate something? The architecture of the computer is entirely irrelevant to this discussion. What matters is what the computer can represent and simulate in data. "represent" means that in memory it stores a bunch of numbers that mathematically describe the relevant details of neurons and synapses... this can be anything from their geometric shape to the relative quantities of neurotransmitters floating around in different places to the voltage at various points along the surface of the neuron. "simulate" means the computer calculates how the data change over time in a way that corresponds to how the actual physical system would change over time. We can represent any mushy slushy squishy squirrelly thing you like in data and simulated it on a computer. The hardware of the computer is irrelevant. The only pertinent limitations are (1) amount of memory and (2) speed of processing.
Your LSD example reveals that you do not understand what "molecular level detail" means in terms of simulation in this context. Molecualr detail means tracking teh position/behavior interactions of each individual molecule. Just because the systemic presence of certain substances is relevant to the functioning of the brain does not mean that these things need to be modeled at molecular detail. Global, regional and local concentrations of various chemicals will need to be tracked in the simulation. I should think this it would be obvious (if you knew what you were talking about) that this would take orders of magnitude less data than what has already been accounted for in my estimate of the data tracking the state of synapses. In fact I have this sort of thing in mind when I suggest that it would take 1KB or 10KB of data to track the state of each individual synapse. The numbers represented in that 1-10KB would represent such things as: size/surface area of synapse, number of various ion channels, quantity of various types of receptors, concentrations of a many different relevant chemicals, and so forth. This is what simulating at the synapse-level, with a large chunk of data from each synapse means. There will also be data for the internal representation of each neuron, data for local ambient concentration of various chemicals, and so forth.
Alright this is particularly clueless. The environment is represented by input to the simulation. For example, take two digital camera, convert that data into new data representing the relevant electro-chemical changes in the light sensing cells of the retina.
I suggest taht you actually know what you are talking about before you say this to someone else.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 3:38 PM
I've been perusing all the tons of peer-reviewed papers on the success rate of the cryonics process. I see you are right: there's nothing that addresses the issues of cryonics factually and accurately.
Posted by: dgerard
|
August 17, 2010 3:38 PM
@Luke: "It doesn't matter if it takes 300 years to crack the brain, if his essential neural information is preserved."
Yes, and there is no evidence at all that any useful information is preserved by what cryonicists do. Not what they might do in the future, but by what they do now. The process is literally "freeze and hope for a miracle."
Posted by: Franz Meersdonk
|
August 17, 2010 3:40 PM
I'm surprised that this hasn't been mentioned before (no philosophers around here? ;), but the mind has some features that seem to evade every attempt at reducing them to Bits & Bytes. That's because they aren't reducible to a materialistic view of the mind. I'm thinking of the first-person- vs. third-person-perspective and all the problems (e.g. subjectivity: cf. Nagel's classic 'What is it like to be a bat?' ; Qualia: subjective phenomenological content of mental states) that go along with it. Today, nobody really knows how to bridge that gap (though there are efforts, e.g. in Neurophenomenology) and the question if this is even in principle possible is still open. Therefore it seems unlikely that we'll ever get to a completely adequate model (let alone one that would allow you to store and reactivate your 'identity' -- whatever that may be...) of the brain, no matter what kind of knowledge and data and algorithms and processing power we have.
Posted by: BrianX
|
August 17, 2010 3:41 PM
Luke:
Well, if you're going to use abusive language, that changes everything.
You see, dipshit, the burden of proof remains on cryonicist douchebags to prove that their half-baked concept of vitrification actually does anything fucking useful, and they have not fucking done that. I'd put more faith in a scratch ticket than an expensive, unproven freezing process. At least rat pellets pay out once in a while. No one's ever successfully thawed out a mammal. Butthead.
Posted by: liquiddark.myopenid.com
|
August 17, 2010 3:45 PM
@Vicki:
Linguistics/symbology isn't restricted to verbal communication. The earliest languages, after all, were just simplified drawings of objects and common concepts. Modern AI studies indicate that, in fact, your mother's face *is* a symbol in your mind, albeit one constructed in a way quite different from that we associate with most symbol engine functions.
As to breathing slowly and changing hormonal balance, you're assuming those are requirements when we have no evidence that they are such. The proposal that modeling the physical processes of the brain is the only way to model the behaviour of the brain is an unnecessarily and unscientifically narrow belief.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 3:50 PM
Well, it depends entirely on how you wish to model the brain, and the purpose of the model.
Kurzweil's proposal to model the brain based on DNA content would require that detail in the simulation.
If you are looking at how a stroke will inhibit blood flow to portions of the brain and thereby affect cognition, then a much simpler model will work, certainly.
And there are all the levels of complexity between.
Posted by: BrianX
|
August 17, 2010 3:50 PM
liquiddark:
The earliest languages, after all, were just simplified drawings of objects and common concepts.
Where did you learn that from, Jacques Derrida? Considering how some animals have some vague concept of proto-language (some usage of symbols, even without grammar or abstract thought -- for example, there's evidence that dolphins may have names), the idea that cave paintings and the like represent the earliest origins of language makes no sense. Language likely began as sounds, and there's no reason to think otherwise.
Posted by: nigelTheBold, Minister of Spankings
|
August 17, 2010 3:53 PM
I believe I read once that language began as fart noises, used to induce laughter.
Posted by: CJO
|
August 17, 2010 3:54 PM
Hierarchical Temporal Memory
Oh, nonsense. I mean, that's a great approach in devising expert systems that can do image and pattern recognition for very specifically defined problems, but as far as I know there's no application that can "look at" 3D objects under variable conditions of lighting and from an arbitrary angle and reliably identify their shape.
The human visual system is constantly, effortlessly solving such open-ended, ill-defined problems in real-time 3D space, often with no prior experience of near-identical input. No algorithm can do the kind of general-purpose vision tasks it can do, but, sure, if you define the problem narrowly enough an expert system can handle it, often more accurately than our fuzzy heuristics can manage.
Posted by: zerosociety
|
August 17, 2010 3:57 PM
I'm a dedicated Transhumanist, but I still have to applaud this post. Not nearly enough people call Kurzweil on his bad science.
Posted by: Paul
|
August 17, 2010 3:59 PM
Agree, which is why I'm sticking to narrow specific points. Less wiggle room.
I'm not re-reading to check if anyone has said there can be no fully functional AI in a short time scale. Many people likely believe such (or simply consider it extremely unlikely), but it is a completely different question to whether we will be simulating brains in so doing or not. Do airplanes simulate avian flight?
People pointing out "we don't have enough information on how the brain works to simulate it" are only doing so because some puffed-up menso claimed it would be done in 10 years, which is impossible without leaps and bounds in knowledge of brain functionality (booming technology means fuck-all if you don't actually understand what you're modeling, as you cannot verify you're actually modeling what you think you are).
If you knew what Moore's law is and had any knowledge of the current state of industry, you'd know they are right.
Technological leaps are by their nature unpredictable. It's at best mental masturbation pretending to know what will happen in the short term in the field of AI at this point. But whatever other information is out there, Kurzweil was off his rocker in the quoted article, and it's pathetic to see people defend numerology as a reliable means of predicting how much code it would take to simulate a human brain (let alone giving it a 10 year time scale when there's no reason to believe we will have enough information on even all the relevant high-level emergent brain phenomena to conceive of an accurate simulator in the time frame).
Posted by: Greylander
|
August 17, 2010 4:01 PM
@ BrianX, #178
Some of us did pay attention in other classes and realize that a Turing machine is not a very complete map of reality.Posted by: CJO
|
August 17, 2010 4:06 PM
The earliest languages, after all, were just simplified drawings of objects and common concepts.
What?! That's preposterous. You're thinking of heiroglyphics or ideograms or something. Those are writing systems, a technology developed to represent natural spoken languages, long, long after language itself evolved.
Posted by: BrianX
|
August 17, 2010 4:07 PM
If you knew what Moore's law is and had any knowledge of the current state of industry, you'd know they are right.
Cite please? I know that publicly available CPUs seem to have been stuck in the 2.5-4GHz range for some time, but the significant increase in multicore and highly pipelined processors (not to mention the massive amount of processing power in current GPUs) would seem to indicate that Moore's Law hasn't quite broken down yet.
Posted by: jasonnyberg
|
August 17, 2010 4:09 PM
#211:
A) Re: strawmen... Your wikipedia blurb about memristors that you found in 10 seconds is a perfect one. As if I claimed that a trivial handful of them constituted AI.
B) Re: DNA as basis for estimating complexity of the brain: If it were the only basis for his estimated timeframe, I'd care much more. Notwithstanding the whole "environment required for proper expression of the genome" issue however, HE IS CORRECT in that the proper expression of the genome results in the construction of a brain that is capable of developing intelligence. It's PZ etc. who think R.K. is going beyond the simple "information content of the genome" argument, and is actually proposing a simulation of that exact biological process (conception->experienced adult human) as a means to implement AI that are reading far too much into it... As evidenced by quotes like this:
Seriously PZ? There's "simulation", and then there's "SSSSSIIIIIIMMMMMMUUUUUULLLLLAAAAATTTTIIIIOOOONNNNNN!!!!!!!!"...
Re: Moore's Law racing ahead of software... True, for Joe Blow surfing the web maybe. For those that need more FLOPs and more terabits/sec, advances in HW can't come fast enough...
Which brings me back to memristors: Being able to implement neuron-like behavior directly (as opposed to using digital logic to simulate same) would yield something like 10^3- to 10^6-fold improvement in each of speed, density, and power consumption. I.e. not to be pooh-pooh'ed lightly.
Posted by: BrianX
|
August 17, 2010 4:10 PM
I haven't read all your posts. Do you ascribe some mystical quality to the mind or reality that can never be understood by science and therefore cannot be simulated within a computer?
No, I don't. What I do think is that the average transhumanist/singularitarian assumes that there's comparatively easy answers to the long chain of ifs that would be required in order to create a working brain simulation in an imaginable timeline. In other words, we don't know what we don't know, and for Kurzweil et al to be making predictions without acknowledging this point is thoroughly foolish, and if you buy into those predictions, you are too.
Posted by: Paul
|
August 17, 2010 4:15 PM
@228
I went and did some reading and I was going off some bad information I read a year or so ago. I will retract the "end of Moore's law". Shouldn't really have even brought it up in the first place, as the number of transistors has nothing to do with the requirement to understand something before you can simulate it.
Posted by: bascule
|
August 17, 2010 4:22 PM
This is really just a lot of pointless nitpicking. Kurzweil's point is that one day computational biology will advance to the point that we will be able to grow simulations of humans inside a computer.
Now, granted, who knows how many decades it will take... but PZ seems to make it sound like it's nearly impossible. It will happen, eventually, as our ability to model biological systems inside computers improves, in addition to exponential improvements in computing power.
It seems there may be lower hanging fruit techniques that will yield results faster though.
Posted by: Rob Seaman
|
August 17, 2010 4:28 PM
At a public lecture Stephen Jay Gould told an anecdote about Richard Feynman. Feynman had excitedly proclaimed his discovery of a profound result in evolution. Gould pointed out that the result was well known in the evolutionary literature and ticked off various caveats and extensions uncovered over the years by other researchers. Feynman was brilliant, but unread in this field.
Is Kurzweil as brilliant as Richard Feynman? Unlikely. But even if so, he will see farther by standing on the shoulders of those who actually have contributed labor to the effort.
Most of us contributing on the other hand to this pleasant discussion are likewise neither brilliant nor well-read in the pertinent literature. In the case of the human brain, this literature is both diverse and vast.
For compression, start with Claude Shannon's fabulous 1948 paper (http://cm.bell-labs.com/cm/ms/what/shannonday/shannon1948.pdf). For what makes us human, perhaps Julian Jaynes' "The Origin of Consciousness in the Breakdown of the Bicameral Mind". Skeptical about that? I am - but it provides context. For the unique insight that fiction can supply there are many thousands of stories starting with Mary Shelley, but I might suggest "Galatea 2.2" by Richard Powers.
For the Evo Devo context that appears completely absent from Kurzweil's beyond naive back-of-the-envelope estimate, Sean B. Carroll's books appear charming, accessible, and (as far as this neophyte can comment) accurate.
Visit the library more than the book store. And one is well advised to frequent the local book merchant to provide context for interpreting the vast range and unevenly accurate insights of the internet, Kurzweiled or otherwise.
On the other hand, note that rather complex and interesting robotic behavior is possible with very small code size. The LEGO NXT is a great introduction to robot ethology. When reading Richard Dawkins, computer types might be advised to pay as much attention to the extended phenotype as to his overexposed memes and selfish genes. (His lyrical "Ancestor's Tale" is also not too far off topic.)
Posted by: Steve LaBonne
|
August 17, 2010 4:31 PM
And our point is that "someday" is likely to be a hell of a long way off, since what's supposedly going to be simulated is unimaginably complex, vastly more complex than Kurzweil and his acolytes even begin to understand, and we don't currently even know what we don't know we don't know about it, let alone know what we don't know.
Posted by: KG
|
August 17, 2010 4:37 PM
GreylanderYes; designing and using simulations is a large part of my job. And the first thing to know about it is that how good your simulation is depends on how well you understand what you're simulating. NO: that's the second thing - the first is that you will always be leaving out some important aspects of the system. Of course, even a very simplistic simulation can help you improve your understanding - that's what they're good for - but it won't do what the thing you are simulating does. It's blindingly obvious you understand nothing about either the brain, or simulation. The brain grows and removes synapses as a key part of learning, which in turn is fundamental to practically the whole of cognition. So you've got to simulate their growth and removal, in spatio-temporal and chemical detail (different synapses work in quite different ways - did you know that? and neurons do not communicate only via synapses - did you know that?), if you want your simulation to do what the brain does. But we are nowhere near understanding these processes.
Sure we can - if we know what details we need to capture. Which we don't. And if we can capture them. Which we can't. Do you really think we know, for example, the "global regional and local concentrations" of all the relevant chemicals in the human brain? Or even what all the relevant chemicals are! Without that latter knowledge, we need a complete physics-based simulation of the whole 1026 or so atoms in the human body...
Posted by: stuartvo
|
August 17, 2010 4:44 PM
BrianX muttered:
...at which point you have ceded all right to claim to be knowledgeable in this debate, and should just shut up.
You have no idea what Moore's Law even IS!
Moore's Law refers purely to the increase in transistor density in integrated circuits. An assumption often drawn from that is that "computing power doubles every 18 months" which is not the same thing. And since people here have been referring to quantum effects, it's clear they're talking about the actual Moore's Law, not the assumption.
Fucking about with multi-core CPUS and GPUS (not to mention that bizarre non-sequitur about deeply-pipelined processors, which are old technology, and mainly related to clock-speed increases, which you admit aren't happening so much anymore) is absolutely not any indication that process-shrink technologies have not stalled.
And without these regular process-shrinks, with their attendant miniaturisation, current-reduction and speed-increase benefits, predicting the continuation of exponential growth in computing power becomes nothing more than guessing and wishful thinking.
IOW, "futurism".
Posted by: bfwallc
|
August 17, 2010 4:47 PM
I must confess that I haven't read Kurzweil's works, so I can't really comment on him in general. On the other hand, I did look up his online edition of "The Age of Spiritual Machines" (1998), in which he makes explicit predictions for 2009, 2019, and beyond.
Short version: he significantly overestimates technological advances.
This, of course, has been a chronic problem with artificial intelligence predictions since the 1950s. (I speak as someone who got into computer science in 1974 and who took graduate classes in AI.)
The AI field as a whole had gotten more conservative over the past 20 years, mostly because virtually every grand prediction was wrong. Even those that were right (computers beating the best humans at chess) took far longer than originally predicted. AI is like cost-effective fusion: it always seems to be 10-20 years away.
While the AI field has gotten conservative, Kurzweil appears not to have. And, as has been noted many times already, anyone who makes statements that half the brain genome = 25 million bytes = 1 million lines of code needs to get back into serious software engineering to get a grip on reality. ..bruce..
Posted by: https://me.yahoo.com/a/dvNQE_kDkZdgeaeOJp3yj2K2zK7KiOY-#9e8fe
|
August 17, 2010 4:50 PM
lets face it, he's created a lot of wealth in the private sector, thats like an uber genius badge that occludes all other considerations. after all - the std deviations of intellectuals are ultimately in the same 2nd deviation of actual cognitive and capacative range - they are simply more motivated, ambitious, or have some entitlement catapult. that is to say, the kurzweil crowd represents the confluence of science and divinity. its scientology 2.0
Posted by: KG
|
August 17, 2010 4:52 PM
- jasonybergArgument by upper-case is never particularly impressive. You are, quite simply, wrong. The growing brain vastly over-produces synapses (and to a lesser extent neurons), then interaction with the environment prunes them down, by processes we are very far from understanding. So there just isn't the kind of short-cut you think there is, and the vast bulk of the information that produces the adult brain is not in the genome, but out in the world - including other people.
Posted by: jasonnyberg
|
August 17, 2010 4:58 PM
HTM is not the same as the Expert System phase/craze of the 80's. (Those were glorified troubleshooting guide-like systems in which rules were manually configured.)
HTM learns the rules on its own, from raw sensory input. Its exact raison d'etre is to
Yes, Vitamin D Video is HTM put to a particular use, but that does not make the technology an "expert system", and it does exactly what you claim does not exist.
Posted by: jasonnyberg
|
August 17, 2010 5:03 PM
#239: I chose the words "capable of developing intelligence" very carefully, to account for your exact argument.
Posted by: mvanbebber
|
August 17, 2010 5:05 PM
CJO
"You may have noticed the dearth of robot-piloted vehicles in your neighbornood. Well, the computational problems in developing autonomous robots able to navigate the real world are largely problems in machine vision."
Once again, I don't like to spoon-feed people information, but the problem of robot-driven vehicles has essentially been solved, and understanding of the human vision system was key to that achievement:
http://en.wikipedia.org/wiki/DARPA_Grand_Challenge
http://techchunks.com/technology/self-driving-robot-vans-start-epic-marco-polo-trip-from-italy-to-china/
Posted by: Luke
|
August 17, 2010 5:08 PM
@BrianX: You were the one who initially said cryonics is snake oil, which is obviously an insult -- an attempt to discredit cryonics without appealing to the evidence. That is why I called you a jerk. I would not have done so had your comment not merited it.
I could also have called you an ignoramus for not knowing that ice crystals are not even a problem any more, but I chose to refrain, as one insult seemed like plenty. Your reaction to my use of the mildly disapproving term "jerk" with actively abusive terms (which I won't bother to quote) says a lot more about you than it does about me.
You could also seriously benefit from doing some research on the topic.
@dgerard: "Yes, and there is no evidence at all that any useful information is preserved by what cryonicists do. Not what they might do in the future, but by what they do now. The process is literally "freeze and hope for a miracle." "
Your decision on where to place the burden of proof utterly baffles me. You're effectively saying that without proof the gun is loaded it's ok to pull the trigger. (Ischemia/decay being the trigger in this case.) Please, just point the gun at your own head if you absolutely must, but kindly leave mine (and everyone else's) out of it. Cryonics is cheap, it doesn't hurt anyone, and it could very well save some lives.
And no, miracles never enter the picture. It is purely a question of what is scientifically possible. As far as we know, there is no mystical reason why a person cryopreserved under ideal conditions (or even sub-ideal conditions) available today cannot be brought back to life by reasonable future innovations.
Posted by: Paul
|
August 17, 2010 5:09 PM
Even with your hedging, it is by no means certain that "the genome results in the construction of a brain that is capable of developing intelligence" in the absence of the external stimuli KG mentioned.
Posted by: stuartvo
|
August 17, 2010 5:11 PM
Ha, Kurzweil is overly pessimistic! You can simulate a human brain in one line of code:
world.simulate(new Brain(Human, 120, male));
See? Simple!
As others have noted, it's the OS/VM/interpreter/libraries/etc that are the problem...
And why do so many readers make the same mistake of missing the point and arguing against a strawman? "PZ said that there will never be strong AI" / "PZ said all theists are evil" / "PZ hates all Southerners" etc?
Posted by: Paul
|
August 17, 2010 5:16 PM
A better analogy is that he is saying you shouldn't claim the gun is capable of shooting until you can actually, you know, demonstrate that pulling the trigger when loaded results in propulsion of a bullet. Until you can demonstrate the efficacy of your product, you're selling snake oil.
Posted by: Wayne
|
August 17, 2010 5:18 PM
Why is it not controversial to say the design of the human arm, or leg, or opposable thumb, or whether your eyes are blue, is encoded in the genome, but not the brain? None of this involves claiming that every single cell or every single cell:cell interaction in an arm or a leg in encoded in the genome.
Posted by: Vicki, Chief Assistant to the Assistant Chief
|
August 17, 2010 5:20 PM
mvanbebber--
If the problem of robot-driven vehicles has "essentially been solved," when am I going to be able to buy one?
Sure, a lot of people will always want to drive their own cars. The market here starts with those who can't, whose options are to get someone else to drive, take public transit, or stay home. There are a lot of people who can't drive for medical reasons (such as visual impairments).
(At some point, I'd like to see a "drive me home" switch in ordinary cars, so the drunk or exhausted driver could get home safely. But that's not going to be common until it's cheap: until then people will take their chances, or call a taxi, because the need is intermittent.)
Posted by: stuartvo
|
August 17, 2010 5:20 PM
Oh please! The robot vision problem has been "solved" by the DARPA grand challenge?
So those vehicles didn't use GPS and laser rangefinders to find their way, and they drove at the same speed as a reasonable competent human driver, in a typical city environment with lots of dynamic obstacles?
Posted by: symmetric
|
August 17, 2010 5:24 PM
I've found the cryonics comments interesting since I've had a policy for years. So here are a few facts and opinions:
- Cryonics isn't that expensive...it's almost always paid for with an insurance policy for under $1000 a year.
- It's not about proving it works. Obviously if you wait until it's proven, it won't be needed any longer. And, so far, it hasn't been proven that it won't work.
- The key question in cryonics is "is any critical information lost that can't be reconstructed?" As this thread points out, we don't know enough about the brain to know yet. Most cryonics people would cancel their memberships if evidence emerges that the process won't work. In any case, the cryonics companies are always working to improve the process given current technology and knowledge.
- I suspect cryonics won't actually work, but for cultural reasons...someone will make a law against it or bomb the facility, or the company will go bankrupt.
- To me, it's worth $1K/year for a small chance at a longer life. My chances are zero without it.
Posted by: Kagehi
|
August 17, 2010 5:26 PM
Should be noted that we do have, but don't understand neural nets, and we have had projects like CYC, which managed to do pretty well at emulating some level of intelligence (albeit, using memory and processing resources that, at the time, ran them into a brick wall, when it came to increasing its smarts past that of a 4 year old). There has also been a move, out of necessity, and recognition that a single core, unified AI, is worthless when trying to built a robot that does jack, to de-unify processes, spreading out the details of how to decide things, or even the processing itself, to systems dedicated to specific tasks, and just giving the "core" processing override and command control over what those others do. This being not too different than how we now know the brain works (though we didn't, really, back when people threw up their hands and started designing bots with multiple, independent, processors, with a core control center). Reading the genome *may* give us a much clearer, in theory, idea of what the basic framework of our distributed intelligence looks like, adding to what we already have, and maybe even answer questions about how it handles what our attempts at neural networks *can't*, adding and removing connections, without completely breaking the system. It might even allow us to buffer those things, so that we could connect things to a design that where not there before, sort of like adding a hearing implant to someone who couldn't before, and the brain being able to adjust to the inputs, since the core function needed to handle them was already there, even if the complex processing needed to understand it accurately isn't.
But, yeah, this guy is clueless if he imagines being able to make a brain, then just feed it data, direct from a person, to produce a copy of them. A direct one to one mapping *might* work, if you know enough of how that mapping happens, and you really do have something that emulates it precisely in software/hardware. But, its a big if, since you need to not just have a proper emulation, and keep it "blank", until you start rewiring it, you also have to know, way better than we do now, what the hell that means to "wire" memories and the like into such a structure, never mind if its even feasible. You might end up with a case where, if someone got flash frozen, you could slice away layer by layer, to get the map you need, but, by the time you could do that, you *might* find it easier to just repair the ice damage to the cells, and unfreeze them...
Posted by: Josh, "Raquel Dommage," Porte-parole Gay Official
|
August 17, 2010 5:27 PM
I'm way, way out of my depth in terms of knowing how soon, if ever, we'll develop real AI. But if we do, and particularly if it's an emulation of a human brain, I think we'd have significant ethical obligations to it. A self-aware, conscious machine would have the same claim to sovereignty that we meat puppets do. Is there a conversation/debate/consensus in the AI community about those ethical obligations and when they start applying? How much ethical leeway would we have to continue experimenting with an artificial consciousness?
Posted by: mvanbebber
|
August 17, 2010 5:31 PM
@Stuartvo
Go ahead and move that quotation over a word, to properly represent my argument. I said "essentially solved" and I was referring to "robot-driven vehicles" not machine vision. And you are wrong, most cars use the GPS navigation for way points, but rely on an internal model of the road, other cars, and obey traffic laws. How does a GPS help with obeying traffic laws and avoiding/waiting for other drivers at intersections? Yes, they drove at an average speed of 15MPH in the Urban Challenge, but the point is that I'm using the DARPA Grand Challenge as support for the hypothesis that driverless cars are not that far off.
Posted by: hkeithhenson
|
August 17, 2010 5:37 PM
If the instructions for building a brain do not come from the genome (and you didn't argue about his estimation of the code size) then where does it come from?
I actually think his estimate may be high. After all the code needed to specify a large memory chip is far smaller than the chip itself.
Posted by: acidalex
|
August 17, 2010 5:39 PM
I think you've misunderstood what 'reverse engineer' means. Replaying all the protein interactions would be simply 'copying the program'. Reverse engineering is a process of recreating the effective results at the desired abstraction level *without* copying the program. Much of the human brain is it's own life support that a reverse engineered brain would never need to develop. A button-up approach to brain understanding for the purpose of intelligence study is the wrong way to go, because you do become bogged down in too much irrelevant stuff. Is recreating the encoding of the CLOCK gene important for creating intelligence? No. Is understanding the role of the pituatary on keeping things in the brain synchronised with the planets rotation important for creating intelligence? Maybe. Is having some concept for time important? Yes. As it's talking about a "brain simulation", the output would of course have to be recognisably similar, but the processing behind the scenes does not need to be exactly the same. That would be a copy, not a reverse engineered model.
Posted by: Steve LaBonne
|
August 17, 2010 5:39 PM
That's also false, for the naive conception of "encoded" that's being bandied about here. But as already pointed out there's an additional layer of falsehood with regard to the adult brain, which cannot develop without environmental stimuli that determine how excess synapses and neurons are pruned via a process analogous to Darwinian evolution. (To an extent that's also true for those other organs, but there the relevant environment is largely internal.)
Posted by: CJO
|
August 17, 2010 5:40 PM
Once again, I don't like to spoon-feed people information, but the problem of robot-driven vehicles has essentially been solved, and understanding of the human vision system was key to that achievement:
Oh, get off your high horse. I'm completely aware of these competitions, and I'm not even saying that some great strides in this area haven't been made. But you're vastly overstating your case when you say the problem has essentially been solved, when the best we can do is average 10 miles an hour over an idealized 60 mile course.
Also, these vehicles get a leg up from some decidedly robotic and non-human "sensory modalities" like GPS input and lasers for measuring distances, so I'm not sure it's accurate to say that "understanding of the human visual system was key," though the sources you "spoon fed" me aren't heavy on that level of detail about the R&D of the machine vision software they're running, so I'd want to know more before I deny that outright. In a nutshell, would they, could they even be made to, work without the lasers, using only their cameras to estimate distances and thus know how soon to slow down to avoid collisions? Highly doubtful, and yet our visual systems do that effortlessly.
Posted by: HertfordshireChris
|
August 17, 2010 5:44 PM
troniq #203 said
The physical brain is the same – but the knowledge base (and the techniques for arguing logically, etc) depend on the way the knowledge base evolves as a result with contact with parents, teachers, and everyday experiences.
Just because everyone looks physically very similar we should not assume that if we could see what was stored in the brain everyone would look very similar.
There may be very little "intelligence" at the physical level - But the brain has the potential for each individual to build their own "intelligent" world model, and in many respects different people, from different cultures and social backgrounds, build a model that is adapted to the environment in which they find themselves.
Posted by: KG
|
August 17, 2010 5:47 PM
hkeithhenson,
Tell you what, why not read the thread? Your question has been answered several times over.
Posted by: stuartvo
|
August 17, 2010 5:48 PM
mvanbebber: Sorry, I just made the same mistake I accused others of: Reading something into a comment that wasn't actually there.
Others had been arguing about robot vision in totality, and had brought up robot car vision as a subset of that, not you.
Posted by: poke
|
August 17, 2010 5:49 PM
Wayne (#247):
I think this is actually a good way of demonstrating why Kurzweil's claim is so absurd. He's compressing the information "about" the brain in the DNA and then saying that's how much computer code you'd need for a working brain simulation. If somebody said the same thing about the DNA for a human arm, PZ could give exactly the same reply. What would be the relevant aspects of the arm? Say, the shape, the different types of tissue and their shapes and locations, the network of blood vessels and nerves, etc. None of that information could be derived from the DNA alone without reference to development, proteins, regulatory networks, cell interactions, etc. I think saying DNA "encodes" anything (except, perhaps, proteins) is a bad analogy (and it is only an analogy) for exactly these reasons. But it has nothing to do with the brain in particular.
Posted by: BrianX
|
August 17, 2010 5:49 PM
symmetric:
Pascal's Wager applied to an unproven technology? Good luck with that.
Luke:
Show me a successful unfreezing of a sufficiently large mammal (say, a chinchilla or a marmoset) with current technology and I'll be happy to admit that human cryonics is at least plausible. Otherwise, it looks like a sucker bet to me.
hkeithhenson:
Comparing a memory chip and a processing unit is not exactly a reasonable comparison. They're two entirely different forms of complexity; I would expect a VHDL listing for even an 8051 to be substantially longer than the average SDRAM chip.
Posted by: KG
|
August 17, 2010 5:50 PM
LukeYes it does: it's a complete waste of resources, specifically energy. Get your corpse frozen if you want, but don't fool yourself.
Posted by: Josh, "Raquel Dommage," Porte-parole Gay Official
|
August 17, 2010 5:54 PM
Cryonics is a ridiculous scam, and it amazes me that otherwise sensible people fall for it.
Posted by: Jasondk
|
August 17, 2010 5:55 PM
I think this is spot on, and very well said. Turning genetic information into phenotypes requires a lot of physics, a lot of context, and complicated dynamics - none of which are encoded in the genome. (At least not directly.)
Cheers
Posted by: https://me.yahoo.com/a/NAtbPxxg3NBFLy53dBUNPxLe8To-#039b4
|
August 17, 2010 5:57 PM
Coming from a "layman" ... by way of Gizmodo, I have to say this is one of the more interesting discussions I have every read.
And Kurzweil didn't invent the electric piano, he improved (I will not say "perfected") a keyboard synthesizer to emulate the sound of an acoustic piano.
And, don't beat him up because of his 10 year prediction - in software develpment, where you multiple your estimates by 10 to come up with the true delivery date, he really is talking 100 years from now.
Posted by: BrianX
|
August 17, 2010 6:01 PM
#266:
And his company produces a generally excellent, if not legendary, product. That doesn't mean he knows enough about biology or neuroscience to be making these predictions.
Posted by: Paul
|
August 17, 2010 6:07 PM
This has been linked on Slashdot and Gizmodo. For those not familiar with Kurzweil threads, there is a high probability of many people posting questions that have already been answered multiple times earlier in the thread. Handle it however you like, just don't be surprised.
Posted by: poke
|
August 17, 2010 6:09 PM
Steve LaBonne (#256):
It's much worse than that. The developing brain relies on spontaneous firing of sensory cells and spontaneous movement in the prenatal environment too. There's evidence that the basic structures of the brain are laid down through this prenatal experience (i.e., if you restrict prenatal limb movement it alters brain development; the development of ocular columns relies on prenatal spontaneous stimulation of the retinal cells; etc). Postnatal experience is also obviously important as can be seen with dark-reared animals, etc. There's also evidence that the proper development of a primate, and especially human, brain requires appropriate socialisation. Neural Darwinism (pruning) is just one proposed mechanism. So if you want to go the route of simulating development you're going to need to simulate a good chunk of the environment. Brain development is a response to use, which is true of most of our tissues and organs, so this holds true of almost any part of the body (albeit less spectacularly).
Posted by: chrisarkenberg
|
August 17, 2010 6:12 PM
"The program is the ontogeny of the organism, which is an emergent property of interactions between the regulatory components of the genome and the environment..."
This is the crux of it, for me. Kurzweil totally overlooks the deep challenges of emergence, not only with the mushy mechanics of neurophysiology but also with even the most typical elements of human thought. There are some *major* assumptions at play suggesting that provable cognition is derivable from replicating DNA, proteins, or even whole-system brain modeling.
Maybe if Kurzweil can model an entire brain bit-by-bit, flash it into some as-of-yet-undeveloped organic protein matrix, let it stew in it's own juices for about 3 hours then seed it with genetic algorithms while teaching it to speak english, pass a college entrance exam, and replicate itself across every Google server on the planet, then maybe he'll get his singularity.
Posted by: HertfordshireChris
|
August 17, 2010 6:16 PM
Steve LaBonne #234 says
Several others make similar comments
The question is whether you are looking at the wood or the trees.
The concept of evolution is very simple – forget the variety of species past and present (which you can catalogue and record in as much detail as you like – declaring how impossible the task as you go). Basically you have a situation where there are variations and statistically some variations are preferable to others – given enough time - end of story.
How many generation did it take to realise that there was no rigid rule “If this therefore that” and you just had relax and let it all happen. Could the clue to human intelligence be similar - while everyone is desparately trying to discover the rules, by digging further and further into the fine detail?
What I was trying to say in #124 is, in effect, that the stored program computer model requires you to try and define everything you need to know in advance – i.e. Cataloguing the trees. But there are other information processing techniques, using sets, which do not require an advance plan of the surroundings, and they are potentially far more human-friendly – perhaps because they work in a more human way.
Posted by: Midwifetoad
|
August 17, 2010 6:18 PM
An assertion that's been made almost as long and as often as the assertion that evolution is facing imminent demise.
Posted by: acidalex
|
August 17, 2010 6:18 PM
Absolutely. A million lines of code to simulate a fully working and knowledgeable brain would be ridiculous, BUT a million lines of code to simulate a brain that can learn, and then the combination of those million lines of code AND all the post-"birth" training the machine gets, resulting in a working data set many times larger than the relatively tiny amount of data representing the code to create a learning device... I don't see that as being far fetched at all. There are plenty of evolution+selection algorithms out there producing results in excess of what the original program described. Evolution lead to us, so we know it's a decent enough model to yield results. Understanding the ways in which it can occur is key to creating the best set of starting conditions to achieve the best result. Something tells me allowing evolution to run on the building blocks of a brain simulator is going to be an important step in achieving amazing results.
Posted by: SheepdogB
|
August 17, 2010 6:23 PM
A couple of questions occurred to me while reading the latest posts.
Can we actually "develop" a system which emulates or actually produces intelligence at an approximately human level when, in fact, intelligence at our level is a product of evolution both physical and social? It would seem that the best we could hope for would be the development of a system which has the capacity to evolve built in to its own internal structure and be able to adapt to external stimuli through not only modifying its programming but the system structure itself as in brain plasticity and the creation of new neuronal cells and neural pathways as an adaptive response without external guidance? Our brain and intelligence isn't, as has been pointed out here literally thousands of times, the result of an "Intelligent Designer" thus making self-organization, I would think, a requisite for the emergence (yuck!) of "artificial" intelligence.
I added the parenthetical "yuck" because I also have developed a distaste for the use of the term "emergence" mostly due to the number of woo-like contexts in which it has been used recently.
My second question has to do with recognition. How and by what means will we determine that AI has appeared? I tend to discount the Turing test because of an emotional consideration. We have a tendency to anthropomorphize and it may be likely that we will infer levels intelligence and empathy that aren't confirmed by empirical testing.
Posted by: CJO
|
August 17, 2010 6:27 PM
BUT a million lines of code to simulate a brain that can learn, and then the combination of those million lines of code AND all the post-"birth" training the machine gets, resulting in a working data set many times larger than the relatively tiny amount of data representing the code to create a learning device... I don't see that as being far fetched at all.
Except that the training a human CNS gets as it develops is not just whatever patch of earth and sky and living things it happens to be raised around, it's the actions and words of some considerable number of minds that are themselves the products of the same kind of development. The fact is, our minds are embodied, and a not insignificant aspect of our experiences are provided free by the universe at large. How do you similarly embody a string of code, and what kind of simulated universe do you put it in so it can learn the ropes, as it were?
Posted by: symmetric
|
August 17, 2010 6:34 PM
@BrianX
I agree it's superficially similar to Pascal's Wager. But it's actually the opposite. There is evidence cryonics will work, and only lack of knowledge against it; but there is no evidence that any religion is true, and only lack of knowledge for it.
On the other hand, your argument is similar to the typical creationist "god-of-the-gaps" argument, but instead of saying lack of knowledge means something is true, you're saying it means something isn't true. The correct response is we don't know yet.
Posted by: PP
|
August 17, 2010 6:34 PM
While I think that Kurzweil's hand-waving is ...um... really out of hand in this case, there has been important work done towards simulating the human brain. Those who point out that the software for such a simulation will also add development time beyond that of the hardware development, they are guilty of an error as misleading as Kurzweil's - ignoring the possibility that progress would be made towards software aspects of a successful simulation in a co-evolutionary process with the hardware development.
I also think that Kurzweil's argument is not just off the wall, but also a complete red herring. Kurzweil was probably (poorly) trying to address the arguments by some that brain simulation would require near QM-aware simulation of human neurons, or maybe it's to satisfy his own dreams of uploading to a non-biological system. However, as a result of research that was initially deprecated by the molecular-level neural model proponents, there is starting to amass a growing body of evidence that modelling of the human brain to the level necessary for producing human-equivalent intelligence may not need to go down to the level of simulating proteins and protein interactions. Many lessons are already being learned and significant steps taken through efforts to model mammalian neocortical columns.
Simulation of the human-brain at a level required for intelligence (let alone the parallel developments necessary for migration of a human consciousness - a.k.a. "uploading") is unlikely to happen soon enough to satisfy Kurzweil's personal timetable for achieving eternal life, but it's likely to happen a lot faster than some people here will be prepared for.
Seriously, in the 50's most people, including some rocket scientists, didn't believe that going to the moon would be achievable in the foreseeable future. It was Buck Rogers stuff. Unexpected socio-political forces made it possible though. There are huge economic and evolutionary pressures that make continued research in this area likely. Kurzweil's visions are like Jules Verne's cannon in the From the Earth to the Moon. He is an evangelist trying to sell people on the idea, to make sure that the money for research becomes available. Just because his metaphorical cannon can't be built in ten years, doesn't mean that a metaphorical rocket won't be built in 30 or 40 years.
Posted by: Paul
|
August 17, 2010 6:34 PM
Well, the problem is you're ignoring how Kurzweil got this "million lines of code". Even if we're talking a self-modifying program managing to do the simulation (I don't see a reason to assume that it's in principle impossible, although it's not feasible with current knowledge -- even if we stumble on one through serendipity, we wouldn't be able to verify that the simulation was accurate), the way Kurzweil reached the number is simply laughable, as PZ pointed out. Can we at least agree on that part?
Posted by: Andyo
|
August 17, 2010 6:38 PM
So this is interesting. This post got crossposted into Gizmodo as well. Some are already coming up with the "all this guy has is insults" crap. The skeptical quality of the writers and commenters has gotten worse though, for instance regarding audiophoolia. Hey, maybe you can write something on audiophool Michael Fremer next time, about how he doesn't understand the ears or something?
Posted by: BrianX
|
August 17, 2010 6:56 PM
symmetric:
What evidence is there that cryonics will work? Where are the studies? Where's the experimental evidence? And please don't bring embryo freezing into it. The thermal characteristics of cells and groups of cells are much different, and overall nobody's found any truly credible ways to reassemble a dead body. Nanotech doesn't count -- we simply haven't gotten far enough with it to figure out how it can be used for medical purposes.
Not to mention that if cryonics were to be proven usable, it would be a last-in, first-out process -- if someone does find a way, say, fifty or a hundred years from now to thaw out and resurrect a body without worry of major cell damage or information loss, it seems rather unlikely that any but the most recently frozen bodies stand a chance at reanimation. God of the gaps? No. I'd say it's realism. It could work, but so could long-distance teleportation. The problem in both cases is that the difference in scale between an embryo or electron and a full-sized body being frozen or teleported makes it hard if not impossible to anticipate either one working based on what we know now.
Posted by: eleusis
|
August 17, 2010 7:11 PM
Technically, you need 5 bits to specify each of the 20 amino acids, since 2^5 = 32 possible states, but 4 bits is 2^4 = only 16 possible states. So 184 amino acids could be specified by 184x5 = 920 bits, or 115 bytes.
Kurzweil calculated 2 bits of information for each nucleotide because there are only 4 nucleotides, so they could be specified by something like:
00 = adenine
01 = guanine
10 = cytosine
11 = thymine
Thus a string of bits, where every two bits represent a nucleotide, is sufficient to represent the genetic code, and 3 bn nucleotides can be specified by 6 bn bits.
But we get your point, PZ. That's not the entirety of the information that you need.
Posted by: BrianX
|
August 17, 2010 7:12 PM
CJO @275:
That's a problem I almost never see someone bringing up -- a working AI in a computer system would have a life so completely alien to human experience that it might be impossible to have a meaningful conversation with one. It's possible that such an AI in a robot body would be better able to relate, but by no means a guarantee.
The problem goes double for a mind transplant. Unlike computers, where industrial process can stamp them out with a reasonable frequency, how do we know that the "machine language" of one human brain is even compatible with another one? It seems like genetic drift alone would make consciousness-level brain mapping something of an exercise in futility; for comparison, consider the human genome, in which the type specimen is one Dr. Craig Venter. It's a genome map that probably has uses we have yet to anticipate, but any advances made on it are going to have to account on some level for the base pair differences between the experimental subjects and Dr. Venter; a sufficiently intricate bit of gene therapy that's based on one of Venter's genes might wreak havoc in someone whose ancestry is not very close to his. By the same token, you can't assume that human brains map onto each other, never mind onto a computer simulation.
And even then, what if it does? Those of you who have played the games Exile or Avernum will be familiar with the Crystal Souls, who were individuals from a subterranean race called the Vahnatai who had acheived sufficient renown among their people to have their minds preserved for eternity. Unfortunately, at least in part due to their isolation in what could be considered an inorganic computer, a good number of the Crystal Souls were absolutely stark raving bonkers. It seems to me that such a fate of an "uploaded" person is a real possibility, which leads one to wonder why Kurzweil hasn't anticipated this.
Posted by: aepalea
|
August 17, 2010 7:13 PM
The interesting question with Kurzweil is not whether he departs from reality, but the exact point in his chain of argument where he departs from reality. Most people get that wrong.
He's far from clueless about progress on modeling human auditory and visual perception. Progress in these research areas is a lot like sequencing the human genome. We all know that won't be complete until the year 2900 at the current rate of progress.
He's almost certainly wrong about general intelligence. It's hard to say conclusively, because we don't yet know enough to estimate the complexity of general intelligence as an algorithm, without even wading into biological simulation (or biologically inspired simulation, which could be a zillion orders of magnitude easier).
That said, I find his incessant extrapolationism extremely tedious. For once he could mention a potential rate changing obstacle. Would it kill him to do so?
Posted by: Pikemann Urge
|
August 17, 2010 7:20 PM
PZ:
Gestalt FTW!
Posted by: Luke
|
August 17, 2010 7:22 PM
There is apparently a huge amount of confusion in the skeptical community regarding the predictions of cryonics. In alternative medicine (i.e. snake oil) there is a specific sort of claim that we have every right to expect verification of; increased lifespan and recovery rates for individuals that use it. If these aren't observed, after careful testing, we know it is bunk. It's as simple as that. We don't have the right to expect proof of that kind for cryonics -- it is not logically implied by the supposed validity of cryonics.
The validity of cryonics is logically based on an unknown, which is whether it will work in the long run, and (to a lesser extent) when. It is valid for as long as we do not know it won't work, in the same way that a stock has value when the company has not gone bankrupt. If we knew in the hard-science sense, that cryonics accomplishes absolutely nothing towards the hoped for end, it would be invalid. But it's not -- not anywhere near close. In fact, as time passes the procedures become more adept -- rendering it quite unlikely that the latest procedures will be proved unsuccessful at a given time. Rather, it is more likely that by the time such verification is possible, the procedures in use for some time into the past (perhaps even as far back as James Bedford) have already been adequate. We may not be able to verify the validity in those past cases for a long time after the first reanimation, but that would not make it less so.
What we can measure (and improve upon, given research funding, which is notable for its absence given the clear social importance of the project) is how badly damaged tissues are before and after cryopreservation, compared to undamaged tissues, and what techniques work to reduce this damage. We can also measure, in a given year, how much damage the technology of that given point in time can reverse. At some point it should be possible to clinically and reliably diagnose information death somewhere very close to the point where it occurs, but it is not currently possible.
In other words, charges against the validity of cryonics bear an enormous burden of proof. You would have to be absolutely certain that the patient has passed the point of information-theoretic death, or that the very latest version of cryonics will (with something approaching 100% certainty) push them over that limit. (I consider it likely that straight freezing does this. But I am not certain enough that I would throw away the corpse of someone who was straight frozen.) In fact, you would have to have a coherent definition of what point between regenerated clone (i.e. complete amnesiac) and undamaged original you are drawing that line -- a sticky ethical situation even if we had hard data on the topic.
The supposed invalidity of cryonics is a HUGE claim. People who trivialize it and use the fact that cryonics sounds weird to stop their curiousity, or ridicule those who put their money towards it, earn no rightful credibility as skeptical rationalists. Nor are they polite, humanitarian, or otherwise admirable for doing so.
Posted by: CJO
|
August 17, 2010 7:24 PM
Would it kill him to do so?
Well, sort of. He'd have to admit he's not going to live forever, which in his mind I think is about the same thing.
Brian:
Yes, ethusiasts for uploading have a self image of a mind that has a body which they figure they can do without. But in truth we are bodies that happen to think, and have experiences. Not saying it'll never be possible. But "stark raving bonkers" is a pretty likely result, I'd say. Lucky for Kurtzweil, he's halfway there.
Posted by: LB
|
August 17, 2010 7:28 PM
Back in the 1970's Isaac Azimov speculated that "A computer might be designed that would not only store data and produce it at will, but that would, on request, sift and correlate such data and come to conclusions on the order of complexity expected of the human brain. It could, in short, be made to reason. If a true scheme of molecular information can be put into use by mankind, what is to prevent a computer from being made as complex and versatile as the human brain and perhaps no larger?"
Most of us remember Azimov for his science fiction, but with this speculation, as I recall, he was more the scientist Azimov. He also speculated at that time that this might happen by 2000.
As 2000 came and went, I thought about this prediction, and here were the challenges as I saw them at that time (and still see them): 1) the technology simply wasn't there(and IMHO we'll not get there with binary computers); 2) we didn't have a clue how to model and build an electronic brain because we didn't understand enough about a functioning brain. Since then I can't see that we've come much further. As a trained engineer and scientist, I've devoted my career to building concrete things that work and can be relied upon. And, if I've learned one thing it's this: One can't build a reliable solution for any problem without solid, proven theoretical and engineering foundations.
Kurzweil's wild-eyed notions are reckless and baseless. They are indeed "science fiction". Even 40 years ago Isaac Azimov understood this: "If technology is ever to duplicate the technique of storing and transmitting information by use of molecules, the fine details of the process must be understood." Thanks for setting him straight on that point PZ!
Posted by: BrianX
|
August 17, 2010 7:34 PM
Luke:
The validity of cryonics is logically based on an unknown, which is whether it will work in the long run, and (to a lesser extent) when.
This much is true.
It is valid for as long as we do not know it won't work, in the same way that a stock has value when the company has not gone bankrupt.
This is wishful thinking at best, complete nonsense at worst. In fact, it's exactly the same logic woo promoters used when confronted with the scientific vapidity of their assertions.
In other words, charges against the validity of cryonics bear an enormous burden of proof.
This is the case for no other scientific assertion. Why should cryonics be set apart? What evidence do you have? Neither you nor any other cryonics aficionado here has answered that question.
Posted by: Tmax01
|
August 17, 2010 7:37 PM
All this because of nothing more than the post-modernist fallacy that using the term "code" for two distinct things somehow makes them identical.
Posted by: symmetric
|
August 17, 2010 7:49 PM
@BrianX
The evidence is any research involving freezing, brain function, etc., including with embryos:) All I'm saying is that until some of this research says otherwise, the default position should be that it may be possible. There's lots of related research:
http://www.alcor.org/Library/index.html#scientific
http://alcor.org/FAQs/
http://en.wikipedia.org/wiki/Cryonics I agree...both are very difficult problems that may, or may not, be possible. Until we know for sure, to me, it's worth risking a few $'s because it would be so cool if it worked.
Posted by: Illuminatus
|
August 17, 2010 7:56 PM
#204: Citing the development of memristors as support for Moore's Law (which describes the change in how many transistors you can fit on a chip) is silly.
Posted by: John Morales
|
August 17, 2010 8:02 PM
LB:
Tell that to the termites.
Posted by: Rob Seaman
|
August 17, 2010 8:02 PM
Someone mentioned the notion of "general intelligence". Most of the counter arguments here against Kurzweil's naivete have focused on engineering difficulties of one kind or another. The more fundamental issue is the belief in the myth of intelligence as a coherently definable thing in the first place.
Rather, see Gould's "Mismeasure of Man" or Gardner's Multiple Intelligences or almost any theory of mind from the psychological community. Freud's ego, superego and id may be mythical, but they reflect the underlying complexity needed to integrate a person. Or see any of Oliver Sacks' essays for a more pragmatic (and infinitely more human) functional decomposition.
In short, having birthed an AI we would be burdened with all the responsibilities of any parent to raise the resulting entity as a sane and happy member of a larger community.
What could it possibly mean to boot a preloaded intelligence onto some host platform? When C3PO powers down, what housekeeping functions remain active? Alternately, how is his consciousness checkpointed?
Do androids dream of electric sheep? That artificial humans appear as characters in our fiction is no argument for the actual possibility of their existence in the real world.
Brenda Laurel's "Computers as Theatre" is a exploration of human user interface design, not how robots might view Punch and Judy.
Posted by: tyler.krpata
|
August 17, 2010 8:03 PM
#234
You seem pretty certain about all of these things. The fact that the human brain is "unimaginably complex" to you reflects your state of mind, not the mind itself.
Posted by: tyler.krpata
|
August 17, 2010 8:05 PM
#264
That should be a signal to you to figure out why these "otherwise sensible people" are being taken by what seems, to you, an obvious scam. Maybe you're missing something.
Posted by: Illuminatus
|
August 17, 2010 8:06 PM
It's probably reasonable to say that you can build a brain from the genome - a newborn's brain. (I know little about developmental neuroscience, so I'm assuming that genetics, not environment, are primarily responsible for brain development.)
Of course, a newborn's brain can't do jack shit. If you want to have a brain that can do something, you need to pump environmental stimuli into it, and that's where Kurzweil looks like an idiot, because we receive a constant fuckload of information through our senses. How much space does a year's worth of high-resolution video take up? A year's worth of audio? A year's worth of somatosensory, gustatory, and olfactory information? Multiply that by twenty, and add that to the space you need for the genetic information.
Posted by: Steve LaBonne
|
August 17, 2010 8:06 PM
That's because, unlike you, I actually know a little about the current state of knowledge. Unencumbered with that, sure, it's easy to bullshit.
Posted by: John Morales
|
August 17, 2010 8:10 PM
tyler @294: Care to elucidate how complex you imagine the brain to be?
@295: Can you cite any examples of cryogenically-suspended humans being revived?
Posted by: Steve LaBonne
|
August 17, 2010 8:10 PM
Yup, pretty much.
A little Learning is a dang'rous thing;
Drink deep, or taste not the Pierian Spring.
Posted by: Luke
|
August 17, 2010 8:17 PM
@BrianX: "This is the case for no other scientific assertion."
Don't be silly. The practice of cryonics is not itself a "scientific assertion", it is a defensive technological intervention against information-theoretic death (i.e. the decay which sets in after clinical death). It is however based on several scientific assertions which do have ample proof; for example that low temperatures halt decay. The elements it does not have direct proof for at present are not parts we would have valid reason to expect to have direct proof for within the given set of circumstances, which is why lack of that particular evidence is not a valid argument against it. Given the staggering amount of indirect evidence in support, it is entirely correct to demand, at a minimum, equal indirect evidence against. Direct evidence against would be nice, but would be overkill (and just as impossible to obtain at this point as direct evidence in favor.)
"Why should cryonics be set apart? What evidence do you have? Neither you nor any other cryonics aficionado here has answered that question."
Most of the compelling evidence observed to date is comprised of things that would be stupid to have to explain, such as the advancement of technology over time, the suspension of chemical activity at low temperatures. This is obvious stuff, but not less important to take into consideration for so being.
Less well known evidence would include the survival of ~100% of neurons in vitrified tissue slices, the successful reanimation of a rabbit's kidney, and recent innovations in organ cloning techniques. Most of this information is available in the Alcor and CI FAQ pages on their websites.
Of course, viability based "proof" is not even the minimum necessary; structural similarities that persist in non-viable "dead" cells could (and as far as we know, should) be more than enough to preserve the needed information, provided their morphological properties remain sufficiently similar.
Posted by: pauljthacker
|
August 17, 2010 8:19 PM
As someone who has followed Ray Kurzweil for awhile, I do think the Deepak Chopra comparison is getting into ad hominem territory, and it's no surprise you're getting defensive responses. But as a computer scientist, I want to address the program/data distinction you mentioned.
I think you're correct that Ray Kurzweil should mention the epigenome. With this and other factors considered, he is underestimating the amount of disk space that would be required to store all the information needed to recreate the human brain. This would be more than 50 million bytes, though I'm not at all convinced it would be more than, say, an order of magnitude more.
However, once these extra factors are accounted for (let's call it genome++), I do believe it's correct to say this is a program. Of course it can't run in a vacuum. As human life traditionally develops, it starts out with this genome in a cell, and through interactions with another human (the mother) through hormones and such, a child develops. What information is required, theoretically, to simulate this? The genome++, a description of the very first cell, knowledge of how the mother's body interacts with a developing human (which will largely be determined by genome++ as well), and the laws of physics. To continue the CS analogy, you could call the environment in which the human develops an operating system, but genone++ is certainly a program running within it, producing results which should be predictable with sufficient knowledge and processing power.
So, granting that genome++ would take up more than 50 million bytes, is the translation to lines of code at all reasonable? I would say yes, as a very rough approximation. And I fear you may be setting up a straw man to attack it. As I read him, Ray Kurzweil is NOT saying that this program would literally have 12 lines of code whose function is to simulate RHEB. He is merely asking how complex the brain actually is. And we can set an upper bound on this by asking how much storage space it takes nature to "program" it. Having measured this, we can say that there should, in theory, be a way for us to program the same thing using roughly the same size of program. Now, whatever programming language we use may be less efficient than nature's "programming language," but as a heuristic, it's about as good as we'll get before we've actually done it.
Will this happen within the next decade? That may indeed be overly optimistic. But I do think that at some point we'll be doing it not through experimenting, but by simulating nature's "programming" itself, along with the basic laws of nature, and seeing what comes out.
Posted by: tyler.krpata
|
August 17, 2010 8:21 PM
Skipping the blatant ad hominem, no substance comment...
#298
@295: Can you cite any examples of cryogenically-suspended humans being revived?
Answer to both of these is pretty much the same. Science has an excellent track record of figuring out things that were previously considered unknowable or unachievable. I don't see any reason to suspect that trend won't continue.
How complex: less than unimaginably so.
As far as cryonics, again, we can all agree it hasn't been done. The question is whether someday it will be possible, and more to the point, whether it will be easier to revive someone who is cryogenically suspended than to revive someone who's been buried in a box in the ground. Again, we have no idea... it's simply the best option we've got.
I would ask you, if not cryonics, than what?
Posted by: John Morales
|
August 17, 2010 8:25 PM
Luke:
At what resolution level? :)
You think gross morphology suffices?
Posted by: Steve LaBonne
|
August 17, 2010 8:30 PM
Computer scientists like pauljthacker appear to be underestimating by orders of magnitude 1) the complexity of the molecular contents, and their spatial arrangement and interactions, in a single cell (let's say a zygotye), 2) the amount of information that would be contained in a complete description of same, and 3) how much (or rather, little) of that we actually know at present. So postulating "a description of the very first cell" sounds simple but is actually pie in the sky.
I'm sorry, computer scientists innocent of any knowledge of biology simply cannot talk intelligently about this stuff. Do you imagine that a Ph.D in developmental biology qualifies one to design a microprocessor?
Posted by: Ichthyic
|
August 17, 2010 8:33 PM
As far as cryonics, again, we can all agree it hasn't been done.
I'm surprised you haven't mentioned certain amphibians, insects, and other species that can indeed return from a "frozen" state.
http://jeb.biologists.org/cgi/content/abstract/201/2/227
if you're serious about there someday being some kind of a "cryogenic option" that actually works for humans, you might actually want to spend some time looking at research like this.
BTW, this research also readily explains why IT DOESN'T work for humans.
those who don't know biology are utterly doomed to be defeated by it.
Posted by: Steve LaBonne
|
August 17, 2010 8:36 PM
There are at least 10^14 synapses in the brain, the strength of many of which is continuously modified on as little as a millisecond timescale in response to all sorts of stimuli. If you claim to be able to imagine just how that works to produce a mind, you're full of it.
Posted by: John Morales
|
August 17, 2010 8:40 PM
tyler, so, no — you don't care to tell us how complex you imagine the brain can be.
Seems to me you've given it some thought, and decided to wimp out as you realise the difficulty of the task.
--
And you think cryonics is worthwhile, because some day it may be possible to revive corpsicles created today.
Well, as long as we're into wishful thinking, I suggest that we already have the technology to tomographically scan (destructive as it may be) a
brainhuman to cellular-level resolution, and store the resulting dataset. Then, sometime in the indefinite future, that information can be fed to a fab to reconstruct the human. 'Cause science and technology marches on, and we can now do things only imagined in the past.How does that sound to you? :)
Posted by: Luke
|
August 17, 2010 8:43 PM
@John Morales: It is hard to say how much resolution is needed exactly (I pack that in the unknown category). However, it seems reasonable to think that much of the information is redundantly stored, and the more important personalty-influencing data corresponds to larger connections (say closer to microscale rather than nanoscale). But any bigger than that and it is probably just cosmetic, and might at best help make a clone have a vaguely more similar personality than genetics alone would provide.
I certainly don't want to be a sub-ideal case, and I certainly want the ideal case to become better before I must undergo the procedure, and by as large of a margin as possible. And of course I'd like to avoid cryonics until it is proven in animals and humans first. But that's a far cry from demanding to first see an animal or human brought back before I'll make arrangements or support the efforts of those who do so.
Posted by: Midwifetoad
|
August 17, 2010 8:44 PM
The folks who think tis project is feasable either didn't read PZ's analysis or didn't understand it.
I find it interesting that engineers think it is easy to design or reverse engineer things that have evolved.
Posted by: prosfilaes
|
August 17, 2010 8:45 PM
That's like saying that because I can pile a half-ton of books on five square feet of my apartment, that it's safe to store 50 tons of books in my 500 sq. ft. apartment.
Sitting here, looking at the screen, I can't read this whole textbox without moving my eyes, much less the toolbar or the menubar. I can tell that the poster above the computer to the right is way too shiny, but what I can't tell what the picture is; I'm getting 20 to 30 pixels worth of information from the 8.5" x 11" poster a yard away, covering a field of vision maybe 10 degrees vertical by 20 degrees horizontal, just because it's not in the center of my vision.
To attack it from another direction, people are going to be unhappy with sensory media that they can see the "holes" in. If our visual input was anywhere 1200 MP, then increasing the resolution of movies would be a big deal. But people still ooh and aah over 4 MP movie projection
Posted by: BrianX
|
August 17, 2010 8:51 PM
symmetric:
I'm not sure Alcor is a reliable source. Just because they're a nonprofit doesn't mean they're an unimpeachable source. As for embryos etc, the single biggest problem to deal with is heat transfer. Unless you're going to put Mr. Frosty on a deli slicer (which would probably be counterproductive) there's no way, even with vitrification, that you could be sure that cell damage was minimized to an acceptable level. In addition, before trying it on humans, you'd have to make sure it works on smaller animals, preferably mammals. A rat would be sufficient, a mouse possibly too small; I picked chinchillas and marmosets upthread because they're both about the size of a shoebox or a small human head, and are somewhat related to humans (chinchillas obviously far less so than marmosets). If the heat transfer problem can be solved on that scale, I would be willing to admit that human cryonics is at least plausible.
Luke:
The practice of cryonics is not itself a "scientific assertion", it is a defensive technological intervention against information-theoretic death (i.e. the decay which sets in after clinical death).
Bullshit. The assertion is that a person can be literally brought back from the dead after being frozen at LN temperatures. This is a testable assertion; however, I'm pretty sure we don't know enough yet to actually test it.
It is however based on several scientific assertions which do have ample proof
Congratulations. With that one glaring logical fallacy you've underscored the difference between, say, a compound killing bacteria in a petri dish and the clinical use of a well-tested antibiotic. You can't get there from here. You have to go through the whole process, usually the hard way. Cryonics started four decades ago with an assumed conclusion and has yet to get much past that.
Posted by: Steve LaBonne
|
August 17, 2010 8:57 PM
And people listening to the first Edison phonograph marveled at how exactly it sounded like the real thing.
Posted by: F
|
August 17, 2010 9:01 PM
# 81 Dan Kaminsky
Wait wait wait.
Dan Kaminsky?
Anyway, "worst project estimation effort ever", indeed.
Er, hey. Whatever happened to the doxpara site?
Posted by: John Morales
|
August 17, 2010 9:04 PM
Luke, what can I say? :)
The ancient Egyptians also preserved bodies, that they might one day walk and breathe again (well, sans braaaaains and various viscera, but they were just this disgusting goo, not that important).
I see current cryogenics as the modern version of this, though I'm (probably) no less scientistic than you. I just fancy myself a little more realistic.
Posted by: HertfordshireChris
|
August 17, 2010 9:05 PM
So the starting point should perhaps be to look at ways to design systems where humans and "computers" can meaningfully exchange information on significant non-trivial tasks.
In 1967 I was working in a very large commercial computer based accounting system. I was a sufficient novice to say something very silly. I said “The real problem is that the task is dynamically open ended because the market was continually changing – so it is a waste to time trying to predefine the task. Why don't we redesign the approach so sales staff can interact symbiotically with the system. Here's how to do it.”
"Don't be so bloody stupid – that's not how computers work ...” was the reaction. As far as I was concerned that was the whole point of the exercise – people do not mix well with a mystical black box which is controlled by individuals who claim a special position because they claim they are very intelligent.
What I had actually done (as some later research showed – see #124) was to design a white box processor where the machine language had a one-to-one relationship with a language that normal people could understand. The real problem was that it was incompatible with the way that people who have been taught to think “program” expected a “computer” to work – and believe it or not – funding was impossible to get the idea beyond the working pilot stage. I felt as if 150 years ago I had come up with a design for a helicopter and all the financiers said – sounds good – we will fund it if you can convince our transport experts – you will find them at the railway station...
If we want to solve the problem we need to do some genuine blue sky research (uncontaminated by stored program computer philosophy) as to how we can get information processing systems which reflects the way their user thinks (whatever way the user thinks – as it would be silly to think everyone thinks the same way – so we need an information mirror which evolves to meet the needs of the individual user.).
Posted by: Steve LaBonne
|
August 17, 2010 9:07 PM
And the situation is even worse than that. Even if you were to try to get around this by ditching Kurzweil's totally childish notions of what you could calculate from a genome sequence, and just make an end run around biology altogether by merely simulating the abstract logic of all those synapses and the complex ways they're constantly modified by their interactions with one another and by myriad inputs, "staggering" doesn't even begin to convey the magnitude of the task or how much we will have to learn before even being able to think sensibly about it.
Posted by: iGus
|
August 17, 2010 9:14 PM
Darn, I guess that the IBM Blue Gene project using the Bluematter algorithm is just a bunch of ballyhoo too then. Silly IBM Scientists: http://modha.org/
Posted by: John Morales
|
August 17, 2010 9:14 PM
Further to Steve @316: Synapse.
Clickety-click on the various links there or the Neuroscience portal link to see how simple this stuff really is.
Posted by: Steve LaBonne
|
August 17, 2010 9:17 PM
Unlike you, they actually understand what they're doing and what it will and will not accomplish. That's how science progresses, by identifying and solving little soluble bits of the puzzle, not by imaginary grandiose kurzweilian undertakings based on hot air.
Posted by: BrianX
|
August 17, 2010 9:18 PM
Hertfordshirechris:
This is all true. And what you're basically talking about is building a user interface for the outside world. There's a fair amount of that done already, but only time will tell if it's enough.
Posted by: Luke
|
August 17, 2010 9:22 PM
"The assertion is that a person can be literally brought back from the dead after being frozen at LN temperatures."
With certain reservations regarding the phrase "literally brought back from the dead" (as if "dead" were an unambiguously defined state -- we are only addressing a very narrow part of the spectrum of physical states collectively known as "dead"), this must be correct for cryonics to actually do what it is hoped to do. However the broader issue at stake is whether it is a valid decision for a person or society with only the knowledge available in 2010 or earlier. That is what I am referring to.
"This is a testable assertion; however, I'm pretty sure we don't know enough yet to actually test it. "
Bingo. Not testable now. It can in principle be tested, but tests performed now would be under restraints, in terms of available technology, which do not necessarily apply (in fact, which almost certainly do not apply) for the entire relevant period of time.
"Congratulations. With that one glaring logical fallacy you've underscored the difference between, say, a compound killing bacteria in a petri dish and the clinical use of a well-tested antibiotic. You can't get there from here. You have to go through the whole process, usually the hard way. Cryonics started four decades ago with an assumed conclusion and has yet to get much past that."
Not correct. The conclusion has not yet been reached as far as cryonics supporters are concerned. The position that implies the most certainty regarding the conclusion here is the one that sacrifices the most, i.e. the one that permits humans to die rather than waste a few dollars on liquid nitrogen.
Posted by: John Morales
|
August 17, 2010 9:29 PM
Luke:
A few dollars in LN — that's all it takes?
Ahem.
That's leaving aside revivification cost (and motive!) in the indefinite future.
Posted by: Luke
|
August 17, 2010 9:31 PM
@John Morales: "The ancient Egyptians also preserved bodies, that they might one day walk and breathe again"
But you have to admit, their knowledge of certain key components of identity were less than ours in critical ways.
"I see current cryogenics as the modern version of this, though I'm (probably) no less scientistic than you. I just fancy myself a little more realistic."
It's weird how often I hear this objection from lots of smart people, since I would think the difference in technology/science levels between egyptians and modern humans would be pretty obvious, and pretty obviously relevant. I mean, given what we now know the obvious solution is to preserve the brain. Given what the egyptians knew, it was not so obvious -- the intuitive answer to them seemed to be everything but the brain. We are relying less on intuition and more on research. It's at least a step in the right direction, wouldn't you say?
Posted by: BrianX
|
August 17, 2010 9:31 PM
What could it possibly mean to boot a preloaded intelligence onto some host platform? When C3PO powers down, what housekeeping functions remain active? Alternately, how is his consciousness checkpointed?
I think such an intelligence would perceive it roughly the same way we do sleep, minus dreams. One would presume that such an advanced AI as C3PO (or Data and the Doctor on Star Trek) would have some kind of permanent flash-like store that serves as a persistent record of the AI's mind; after all, you're stepping outside biology at that point so such a thing is entirely feasible.
I mentioned the Doctor from ST:Voyager for a reason -- there was an episode where he was transferred to a hijacked Federation ship in the Alpha quadrant, and they were worried about getting him back before the connection dropped. Although it's fair to say that it wasn't handled as well as it could have been (the Doctor almost certainly kept a backup copy of himself somewhere, which would have meant little more than a day or so of amnesia for him), it did make me consider the possibility that the reason the Doctor's code was locked behind some kind of mutex while he was off-ship might have had something to do with instabilities inherent in merging two slightly divergent mind patterns. (Okay, yes, I'm giving STV way too much credit. I'll live.)
Posted by: Ichthyic
|
August 17, 2010 9:32 PM
The position that implies the most certainty regarding the conclusion here is the one that sacrifices the most, i.e. the one that permits humans to die rather than waste a few dollars on liquid nitrogen.
no, the conclusion is, correctly, that freezing a person in LN, at this stage of the game, IS the same as death.
that isn't going to change in the near future.
what might change is figuring out a way to freeze someone that DOESN'T kill them.
right now, that ain't the case.
long way from it.
Posted by: BrianX
|
August 17, 2010 9:37 PM
Luke:
Bingo. Not testable now.
And this qualifies as a defense? If it's not testable, it's cargo cult science at best.
The conclusion has not yet been reached as far as cryonics supporters are concerned.
And yet you assume that the conclusion will be reached. Like I said, cargo cult thinking, pure and simple. Using a sufficient supply of negative energy, I can do FTL travel by wormhole; Kip Thorne said so. But what he hasn't said is if negative energy exists, or can exist. He's a personal friend of Stephen Hawking and one of the most important theoretical physicists in the world, and he doesn't have the level of certainty you do. Check your hubris.
Posted by: tyler.krpata
|
August 17, 2010 9:41 PM
#307
My inability to define the task is no more a barrier to it being accomplished than yours is. I do know that saying "I can't imagine how it could be done, therefore it can't be" is simply argument from incredulity. (Which we all know is a logical fallacy? Right?)
(I suppose we should define the "it" in this case, though. I'm talking about the possibility of simulating human consciousness, not supporting the Kurzweil reasoning from the OP.)
"Wishful thinking" is a loaded term. Do you have life insurance? Retirement investments? You're engaging in a similar sort of "wishful thinking," that past trends will continue to the point where they'll be beneficial to you in the future.
As for how that sounds to me, it sounds reasonable. I'd need more information. It seems to me that cryonics is a simpler, cheaper, and potentially more effective route to preserve the same data. Either option is better than the alternatives, assuming your goal is to have some non-zero chance of extending your life.
Posted by: Ichthyic
|
August 17, 2010 9:43 PM
I'll make arrangements or support the efforts of those who do so.
instead of wasting your money promoting the efforts of those with big refrigerators, but little sense, why not put your money towards researchers who actually look at how tissues survive freezing?
check out the article I posted the link to.
send you money to a zoology lab at a university somewhere. It would be much better spent there, to accomplish your stated goal, than it would be at one of the cryogenic companies.
Posted by: Ichthyic
|
August 17, 2010 9:45 PM
In fact, here's a great place to send both your questions and your cash:
Jon Costanzo
Posted by: BrianX
|
August 17, 2010 9:47 PM
But you have to admit, their knowledge of certain key components of identity were less than ours in critical ways.
I take Celexa and Wellbutrin. They're the chemical equivalent of adjusting your neurotransmitter levels with a sledgehammer. They're an improvement over MAOIs (thankfully, since I'm a major foodie), but they still represent a brute-force approach to brain chemistry. In other words, the clinical gold standard for treating my chronic depression still amounts to "throw it against the wall and see what sticks". If we know so little about brain chemistry now that the best we can do is nonselectively bash at the levels of specific neurotransmitters, how can we expect to understand (and preserve) the complex interplay of nerve signals responsible for personality and memory?
Posted by: Ichthyic
|
August 17, 2010 9:48 PM
Here's his lab page:
http://www.units.muohio.edu/cryolab/projects/VertCryobiology.htm
when considering freezing something in LN, note the following:
Posted by: Ichthyic
|
August 17, 2010 9:50 PM
"throw it against the wall and see what sticks".
that's not a critique of the science of brain chemistry, it's really a critique of how we practice medicine these days.
Posted by: John Morales
|
August 17, 2010 9:53 PM
Luke,
More than a step, yes. I suggest, however, that the mind might depend on more than just what's in the cranium — are you so sure the CNS and the various neuroendocrine elements (for example) are dispensable for a functional mind?
To borrow Rumsfeld's idiom, we're pretty sure about the known unknowns. It's the unknown unknowns that may be problematic.
Posted by: BrianX
|
August 17, 2010 9:53 PM
Ichthyic:
Sadly true, though I wasn't really meaning to criticize to begin with -- it does, however, emphasize the fact that psychiatry is probably decades behind the rest of medicine, and that's largely because we still have only a rough picture of how the brain works.
Posted by: joseph.bashe
|
August 17, 2010 9:58 PM
After reading several dozen comments, the popular rebuttal to Kurzweil is "Simulating human intelligence is possible. It's just too complex to do in 10-20 years. So Kurzweil is a wank."
Sounds a lot like "Evolution is possible, but evolving a human from nothing is too complex to do in 4 billion years. So scientists are wanks."
Yes, this is the exact argument "old-earth" IDer's use.
Yes, I just compared PZ and his readers to IDers.
Furthermore..
FTA: "Reverse-engineering the brain is being pursued in different ways," says Kurzweil. "The objective is not necessarily to build a grand simulation - the real objective is to understand the principle of operation of the brain."
All the productivity you all wasted typing out rebuttals to straw men could have been put to use writing part of that million lines. Kurzweil explicitly stated he was not talking about a ground-up atom-for-atom simulation of a brain.
By the way, notice how the 1 million lines of code reasoning in the article is NOT in quotes? According to standard rules of grammar, that is because Kurzweil did not explicitly say it!
Finally, somebody already lambasted those invoking quantum mechanics, so I will just add that Roger Penrose hypothesized decades ago that quantum phenomenon might govern the brain's functions. I thought that sounded great! Turns out all subsequent research does not support this hypothesis. The brain is a classical system. Until, of course, you accelerate it to 99.9999% the speed of light and smash it with another brain in a particle accelerator.
Posted by: tyler.krpata
|
August 17, 2010 10:01 PM
#331
This is basically a solved problem in cryonics. (See Alcor's web site for more.)
You also seem to be mistaken about the general practice of cryonics... in order for it to be legal, the subject has to be legally dead. We are not talking about freezing a live person. He or she is definitely "dead" in the sense that we understand it today.
Posted by: John Morales
|
August 17, 2010 10:02 PM
Tyler:
Well, it would be, but you've straw-dummied the actual contention to which you were responding, to wit: And our point is that "someday" is likely to be a hell of a long way off, since what's supposedly going to be simulated is unimaginably complex (my emphasis).
Yeah, the eternal rubes' justification. Homeopathy and crystal healing sound reasonable to their proponents, too. :)
Betcha that the more actual scientific literature on the subject you read, the less plausible it becomes. That's the literature, mind you, not the blurb!
Posted by: Steve LaBonne
|
August 17, 2010 10:05 PM
The projection is strong in this one...
Apropos nothing in particular, the discussion on Slashdot is generally and gratifyingly much more sensible than this kind of ignorant codswallop, with many commenters having a good grasp of what PZ was saying.
Posted by: Luke
|
August 17, 2010 10:06 PM
@Ichthyic: no, the conclusion is, correctly, that freezing a person in LN, at this stage of the game, IS the same as death.
Conclusion -- but based on what evidence? Technology, however slow to evolve, is nowhere near topping out yet. To test the conclusion you would need the best possible technology that can be developed, at least within a timeframe in which it is reasonable to expect the patients to survive. That could be several thousands of years. Actually, we can test it by freezing a set of patients, and if they are all thawed before someone develops a cure for their level of damage, we'll have proved that cryonics is based on a flawed premise. Otherwise we can go on not assuming anything and just cryopreserve everyone on principle.
@John Morales: "A few dollars in LN — that's all it takes?"
Accounting for economies of scale, yes. Don't confuse the cost of cryonics where there's a couple hundred patients in dewars that hold a max of 25 neuros, with the cost of storing several billion of them in a few centralized, heavily automated facilities. The latter is a larger absolute number but a much, much smaller number per patient.
I would qualify this by saying that trying to prevent information-theoretic death due to a straight freeze would likely motivate us to allocate at least some money towards stabilization. I am not sure whether the economies of scale on that would be anywhere near as dramatic as for storage, but I imagine they are still significant enough to bring the cost below that of a conventional funeral.
That's leaving aside revivification cost (and motive!) in the indefinite future.
Development cost for revivification is likely to be high overall (though many prerequisite technologies should be pioneered for other purposes -- brain damage reversal, etc.), but once it is perfected I would expect that it is significantly less expensive per individual. There are essentially no biological deadlines, and the interest from trust funds created by richer cryonauts should cover much of the development cost. Bear in mind also that at the point where reanimation of cryonauts is possible, suspended animation (the "easy" version of cryonics, i.e. little to no damage, perhaps used quite frequently) will have been in existence for some time. Society at that point would have much more incentive to think of "corpsicles" as being in fact human beings with a right to life. All the more so if we move to mass cryonics now, as everyone will then have at least a few ancestors in the cooler.
Posted by: BrianX
|
August 17, 2010 10:09 PM
This is basically a solved problem in cryonics. (See Alcor's web site for more.)
A website may as well be a press release as far as scientific credibility is concerned. How about some peer-reviewed papers and replications of their results?
Posted by: Rob Seaman
|
August 17, 2010 10:10 PM
BrianX says: I think such an intelligence would perceive it roughly the same way we do sleep, minus dreams.
Why minus dreams? Are we to believe that dreaming has no evolutionary function? One can speculate on the existence of a "designed" entity and the pointlessness of dreaming, but that doesn't make it the case.
One would presume that such an advanced AI as C3PO (or Data and the Doctor on Star Trek) would have some kind of permanent flash-like store that serves as a persistent record of the AI's mind; after all, you're stepping outside biology at that point so such a thing is entirely feasible.
Notionally conceivable is not the same as "entirely feasible".
Kurzweil may think that the recipe for a human brain can be compressed into a few megabytes, but that isn't the same as capturing and efficiently restoring the running state of trillions of synapses.
Posted by: Ichthyic
|
August 17, 2010 10:12 PM
He or she is definitely "dead" in the sense that we understand it today.
Oh i understand it alrighty. It's just pointless, is all. It will have a point when you can freeze someone who is going to die, because of some fatal illness, say, and COULD be brought back at a later date.
Alcor's web site for more
I did. It's crap.
you must have missed this point even THEY raise:
get that through yer thick skull, moron. Dead or alive, most human tissues are not recoverable after freezing - there is too much tissue destruction.
they are NOT pioneering any techniques in how to successfully freeze tissues.
in fact, while they rarely do rely on actual research, when they DO, it's the research of people like I cited earlier.
so, again, why waste money with a company that spends it on refrigerators, instead of spending on the people who actually do research?
Posted by: BrianX
|
August 17, 2010 10:13 PM
suspended animation (the "easy" version of cryonics, i.e. little to no damage, perhaps used quite frequently)
You have got to be kidding me. Suspended animation and cryonics are two entirely different concepts. Furthermore, suspended animation has actually been done on a scale large enough to be plausible for a human, albeit tentatively so.
Posted by: Ichthyic
|
August 17, 2010 10:15 PM
Until, of course, you accelerate it to 99.9999% the speed of light and smash it with another brain in a particle accelerator.
that would take quite the accelerator for that much mass.
the brain being composed of MANY particles, after all.
:)
Posted by: tyler.krpata
|
August 17, 2010 10:16 PM
Apparently I did. I think the timescale is certainly debatable, but it's very reasonable to think it might be quite longer than some futurist predictions.
Posted by: John Morales
|
August 17, 2010 10:16 PM
joseph @335,
Straw-dummy, addressed ad nauseam above.
That said, let's see some Kurzweil's predictions for 2009 (from his 1998 book):
---
People typically have at least a dozen computers on and around their bodies, which are networked using "body LANs" (local area networks). These computers provide communication facilities similar to cellular phones, pagers, and web surfers, monitor body functions, provide automated identity (to conduct financial transactions and allow entry into secure areas), provide directions for navigation, and a variety of other services.
---
The majority of text is created using continuous speech recognition (CSR) dictation software, but keyboards are still used. CSR is very accurate, far more so than the human transcriptionists who were used up until a few years ago.
Also ubiquitous are language user interfaces (LUIs), which combine CSR and natural language understanding. For routine matters, such as simple business transactions and information inquiries, LUIs are quite responsive and precise. They tend to be narrowly focused, however, on specific types of tasks. LUIs are frequently combined with animated personalities. Interacting with an animated personality to conduct a purchase or make a reservation is like talking to a person using videoconferencing, except that the person is simulated.
---
Autonomous nanoengineered machines (that is, machines constructed atom by atom and molecule by molecule) have been demonstrated and include their own computational controls. However, nanoengineering is not yet considered a practical technology.
---
Uh-huh. So, joseph, I take it that you talked to your computer to post your comment? :)
Posted by: BrianX
|
August 17, 2010 10:17 PM
Rob Seaman:
Why minus dreams? Are we to believe that dreaming has no evolutionary function?
Actually, for a much simpler reason: when dreaming, the human brain is active. When an AI's power is off, there is no activity at all.
Posted by: Ichthyic
|
August 17, 2010 10:18 PM
Conclusion -- but based on what evidence?
*ahem*
you must be new to this game.
ALL OF IT.
as I posted for your shorter-winded cousin, even
Alcor admits that you cannot recover most human tissues frozen in LN.
why don't you idiots actually LOOK at some of the actual research involved, rather than company fucking websites?
Posted by: John Morales
|
August 17, 2010 10:22 PM
tyler:
I'm smug? I'm not the one advocating expensive, untested procedures that rely on Clarke's third law for a successful outcome. :)
So, do you also buy into Tipler's Omega point?
(It has the advantage of being cheaper for you.)
Posted by: joseph.bashe
|
August 17, 2010 10:22 PM
John, thanks, I was looking for some of those predictions.
Ant yes my speech, retransmission software wrote the end tire post!
Posted by: BrianX
|
August 17, 2010 10:25 PM
John Morales:
1. Zigbee already exists, but it doesn't seem to have nearly the level of interest that Bluetooth does, which is substantially more expensive. Could happen, but borders on trivial.
2. I've no doubt that speech-to-text will continue to improve, but it's already pretty good and doesn't have much appeal for much people. In fact, if you've noticed, cellular phone users frequently prefer text to voice communication, since there's a somewhat greater sense of privacy. In addition, speech control has been available on at least the Mac platform since ~1993 (I don't know about Windows or Linux, but I imagine it exists there too) and I think most Mac users, if they're aware of it at all, consider it either a novelty or an accessibility device. Kurzweil is ignoring the intrusive factor of sound in, say, a large computer lab.
3. "Animated personalities": One word: "Clippy."
4. And nanotech is still a speculative technology, mostly good for lab tricks.
The funny thing is, all these technologies Kurzweil predicted exists, and even existed at the time. But none have much respect.
Posted by: tyler.krpata
|
August 17, 2010 10:26 PM
#340
Dude, the web site LINKS TO THE RESEARCH. I mean, damnit.
http://www.ncbi.nlm.nih.gov/pubmed/6467964?dopt=Abstract
http://www.ncbi.nlm.nih.gov/pubmed/10700144?dopt=Abstract
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2781097/
#342, no quote, you are just confused and aggressive.
Posted by: poke
|
August 17, 2010 10:27 PM
I think this term 'simulation' is dangerous. I think it would be better to speak of modelling the brain and save 'simulation' for something like a flight simulator where the goal is to provide an experience similar to a real task for a user. A model (or simulation) of the brain would be based on a description of the brain. A description of a natural system is given for some particular human purpose at some given level of detail. It's difficult to make sense of what a 'complete description' without reference to a purpose.
A model (or 'simulation') is created also with some purpose in mind and to meet given criteria. 'Being real' is not a meaningful criteria. Accuracy is only relevant relative to some purpose. It's ridiculous to speak of a 'simulation' of anything as if one could 'capture' the real system inside a computer and have something 'identical.' The concept of identity doesn't even have any meaning here. A picture is not identical to the object it depicts, although it can be more or less accurate for one's purposes, and the same is true of a computer model.
Posted by: tyler.krpata
|
August 17, 2010 10:29 PM
Also, http://lesswrong.com/lw/2jn/the_threat_of_cryonics/
Posted by: Ichthyic
|
August 17, 2010 10:30 PM
Dude, the web site LINKS TO THE RESEARCH. I mean, damnit.
ROFLMAO
you didn't actually even read those, did you.
VASCULAR TISSUE.
remember when I remarked above "most human tissues cannot be recovered from freezing".
yeah.
there's a difference between posting random literature, and understanding its relevance to the point you are making.
epic fail on your part.
Posted by: tyler.krpata
|
August 17, 2010 10:33 PM
john
So, no literature then? Just more hand-waving?
(Also, it's not expensive. Less investment per year than I spend on beer.)
Posted by: Ichthyic
|
August 17, 2010 10:33 PM
Also, http://lesswrong.com/lw/2jn/the_threat_of_cryonics/
oh, yawn.
yes, because I think you're a moron, I must be afraid of what you're saying.
yeah. sure.
strange, that's a standard creationist argument, too.
Posted by: BrianX
|
August 17, 2010 10:34 PM
Technological prognosticators frequently amuse me because they don't seem to understand how the rest of the world thinks. A few examples:
1. HD Radio. Digital sound is great, right? Well, yes, but few people want to pay for extra streams of the same slop that commercial radio programmers put out, and there's virtually no chance of broadcast radio going all-digital since the installed base of analog radio numbers in the billions. (Seriously, there's probably at least as many broadcast radio receivers in the world as there is people, especially since First Worlders will usually have a good handful of receivers in their house or apartment.)
2. Microsoft's idea of a smart kitchen. First off, Walmart aside, it'll be a few years yet until RFID in packaging is cost-effective for things like milk jugs. Second, why would I choose Microsoft's recipe database over something I already own, how would I program the recipe into the system, and given Microsoft's DRM knobslobbing of big media, would I even be able to?
3. Bob Metcalfe, inventor of a little-known networking system called Ethernet, made a point of ignoring the disruptive potential of Open Source software. He also saw wifi as a dead end and thought the Internet would be unable to handle the mid-90s explosion in growth.
Posted by: Ichthyic
|
August 17, 2010 10:34 PM
(Also, it's not expensive. Less investment per year than I spend on beer.)
money that would be far better spent funding actual research.
'nuff said.
Posted by: BrianX
|
August 17, 2010 10:36 PM
tyler, #354:
LessWrong? Seriously? Eliezer Yudkowsky's temple to Mensa Mentality and Eliezer Yudkowsky?
Posted by: Furcas
|
August 17, 2010 10:38 PM
Um, Kurzweil's calculation may be wrong, but it doesn't really matter. You understand, don't you, that he's not proposing that we reverse engineer the brain by looking at the genome? He thinks we'll do it by scanning it. And while understanding the brain is the ultimate goal, it's not necessary to understand the brain to emulate it. Understanding will come as a consequence of emulation, not the other way around.
Posted by: tyler.krpata
|
August 17, 2010 10:39 PM
#355
is totally relevant.
You clearly did not read these, as you accuse me of not doing.
Posted by: John Morales
|
August 17, 2010 10:41 PM
tyler:
Ahem. You're the proponent, I'm the skeptic, remember?
Remember this question I asked you above: "Can you cite any examples of cryogenically-suspended humans being revived?"?
Your response: As far as cryonics, again, we can all agree it hasn't been done. The question is whether someday it will be possible [blah].
Who's doing the hand-waving? :)
Posted by: tyler.krpata
|
August 17, 2010 10:41 PM
BrianX, don't read it then.
Posted by: Ichthyic
|
August 17, 2010 10:41 PM
hell, if you'd rather spend your money on a company instead of a researcher specifically, why not on the actual company that sponsored the research Alcor is "coopting" for the little scheme to fleece rubes like yourself?
here, this is the company that funded the paper you cited last:
http://www.21cm.com/
seriously, putting your money in Alcor is a waste of your own interests!
Posted by: Ichthyic
|
August 17, 2010 10:43 PM
is totally relevant.
only if you consider all human tissue to be vascular.
(hint: it's not)
Posted by: BrianX
|
August 17, 2010 10:44 PM
Tyler:
The important point here is, what authority does an article on LessWrong have? If Alcor's web page is naught but a press release, this is slightly better but doesn't meet the standard of a scientific paper. At best, it might show up in arXiv in one of the green ink sections.
Posted by: Steve LaBonne
|
August 17, 2010 10:45 PM
And it's been extensively explained already just how stupid THAT is. Not just because of the mind-boggling complexity (>= 10^14 synaptic connections), but because a static "scan" of such an enormously dynamic system, even if it were conceivable to do such a thing at the required resolution, is just not going to get you anywhere (try calculating he number of possible different states of 10^14 synapses if each synapses had, say, only 5 discrete levels of strength- which is far to simple to be realistic).
There is just a colossal failure of imagination here about just how complex the brain is.
Posted by: tyler.krpata
|
August 17, 2010 10:46 PM
You still don't get to say "it can't work because it can't."
Yes, the argument for cryonics relies upon as-of-yet undeveloped technology that may never come along. I am not trying to hand-wave that away. It's up to you if you think that means you should save your 500 bucks and die in a box in the ground. I'll take my chances.
(For the "you should donate to a worthy cause instead" crowd, where do YOU donate to?)
Posted by: Ichthyic
|
August 17, 2010 10:47 PM
...btw, none of the current research invalidates the very simple point:
you cannot revivify LN frozen humans.
ergo, if you have yourself frozen now, you will not be revivified.
period.
again, if you really WANT to have yourself successfully frozen, I highly suggest you not waste your time with Alcor, and instead actually invest in some of the research I linked to.
It still might not happen in your lifetime, but at least you would be contributing to actual research that might progress your goals, instead of tossing your money at a bait-and-switch scam.
Posted by: tyler.krpata
|
August 17, 2010 10:48 PM
@BrianX, no argument from authority here. The reasoning stands on its own, but debate the ideas, not the domain name.
@Ichthyic, cool story bro
Posted by: PZ Myers
|
August 17, 2010 10:49 PM
Then why does he bring up the genome at all? The genome can only be interpreted through the lens of development.
That doesn't make any sense, either. How? We have no tools for scanning a living brain at the molecular level, and no prospects for such a technique.
Why are the Kurzweil defenders all making up such nonsense?
Posted by: Furcas
|
August 17, 2010 10:49 PM
Skepticism of a claim isn't always right by default, you know. It depends on what kind of claim it is. Claims of the form "Object X exists" are usually false if there's no evidence for or against them. However, claims of the form "Goal X is or will be possible" are usually true if there's no evidence for or against them. That all the important information inside a vitrified brain will be recoverable one day (assuming scientific progress isn't stopped, say by blowing the planet up) is a claim of the second type
Posted by: Ichthyic
|
August 17, 2010 10:50 PM
where do YOU donate to?
last time, and much simpler...
try donating to your local university zoology dept.
or even directly to one of the researchers involved in doing the kind of work that interests you.
Alcor is nothing but a scam.
Posted by: Steve LaBonne
|
August 17, 2010 10:52 PM
The men of straw are still on the march! "Default"?
No, very good scientific arguments.
Posted by: John Morales
|
August 17, 2010 10:54 PM
tyler:
Good thing I didn't, then.
Indeed. Yet you think that the first stage (adequate preservation) is already here.
That, quite aside from your reliance on the "as-of-yet undeveloped technology" for revivification, is your wishful thinking exemplified.
cf. What I wrote to Luke @314.
Posted by: tyler.krpata
|
August 17, 2010 11:02 PM
So you're saying today's preservation methods are no more adequate than those of the ancient Egyptians? Or are you saying the brain is not the important part?
I think there is a good enough chance that today's methods of preservation are adequate that it's worth the amount of money it costs. Significantly increase the cost and I might have second thoughts. It's an expected value calculation, not a yes/no decision.
Posted by: llewelly
|
August 17, 2010 11:04 PM
Cryonics - bah. Just a plot by the Zombie - Mind Flayer Coalition to get more brain ice cream.
Posted by: Ichthyic
|
August 17, 2010 11:05 PM
I think there is a good enough chance that today's methods of preservation are adequate
yes, you've made yourself quite clear on that.
you're wrong.
even Alcor says you're wrong.
not much left to be said, really.
Posted by: Ichthyic
|
August 17, 2010 11:07 PM
Just a plot by the Zombie - Mind Flayer Coalition to get more brain ice cream.
makes me wonder what other plots they have going to get brain ice-cream?
Posted by: BrianX
|
August 17, 2010 11:13 PM
Ichthyic:
Well, you know, with Ted Williams' head, they get Moxie-flavored... what else do you need?
Posted by: Luke
|
August 17, 2010 11:18 PM
@BrianX: "If it's not testable, it's cargo cult science at best."
Not testable now, as a simple yes/no question. (In fact, it is not even meaningful as a yes/no assertion, given varying shades of death and memory preservation.) That does not rule out evidence that makes it appear more or less likely, and which tells us which methods work better or worse on a relative scale.
"And yet you assume that the conclusion will be reached. Like I said, cargo cult thinking, pure and simple. ... Check your hubris."
Actually there is plenty of good indirect evidence that we will have higher certainty on the topic than we currently have, because right now we have better data than we did when cryonics first started. For example we now have knowledge of stem cells, and the fact that neurons are replaced in the brain. Our ability to scan at finer resolutions, and further research into what is possible with nanotech, should give fairly convincing evidence one way or the other.
I do not see why the opposite assertion, that our knowledge about the limits of cryopreservation and reanimation will not expand significantly, seems more reasonable to you. It is utterly mind-boggling to me that you would think such a thing.
@Ichthyic: "as I posted for your shorter-winded cousin, even
Alcor admits that you cannot recover most human tissues frozen in LN."
You keep missing the point. We don't know what point human tissues become irreversibly damaged. Our technology is in the early stages of development where tissue engineering is concerned. Assuming the worst case regarding what is potentially possible is just as silly as assuming the best case would be.
"because I think you're a moron, I must be afraid of what you're saying."
Not the point I intended when writing that article, so if it came across that way I apologize. My intention was to examine why people so pervasively don't even think about cryonics, much less read the standard arguments for it. There are a number of fallacious and spurious arguments against cryonics that have already been shot down here: that it implies eventual success as a necessary criteria, that ice crystals cannot be prevented, that death is some kind of binary on/off state, that egyptian mummification is just as technologically advanced, and so forth. The question is why people latch on to these ideas so ferociously and with so little provocation. It makes sense to me that it is because they are afraid to think very deeply or neutrally about the subject matter for fear of being persuaded, that it at least provokes some kind of hostile gut reaction. Am I wrong?
Incidentally, I suspect creationists are the ones who do not consider the evidence for evolution carefully because of the phobia it inspires in them, a feature they attempt to project back onto evolutionists.
Posted by: Ichthyic
|
August 17, 2010 11:23 PM
We don't know what point human tissues become irreversibly damaged.
*sigh*
yes, we do. It's why Alcor posts that statement on their own website.
Incidentally, I suspect creationists are the ones who do not consider the evidence for evolution carefully because of the phobia it inspires in them, a feature they attempt to project back onto evolutionists.
my pardons, then, I took your point as being the same as the creationists.
Posted by: BrianX
|
August 17, 2010 11:25 PM
Not testable now, as a simple yes/no question. (In fact, it is not even meaningful as a yes/no assertion, given varying shades of death and memory preservation.) That does not rule out evidence that makes it appear more or less likely, and which tells us which methods work better or worse on a relative scale.
No, it doesn't rule out future evidence. But until there is such evidence, you're putting the cart before the horse.
Posted by: amphiox
|
August 17, 2010 11:37 PM
Analogy fail. Because the time scale is the critical point of this argument. Therefore you cannot make equivalence between 20 years and 4 billion years. The difference between 20 years and 4 billion years is the whole point.
Kurzeil's assertion is equivalent to saying you can evolve a prokaryote to a human in 20 years. Evolution works in part because it occurs over a very, very, very long time. And any one (can't call such a person a scientist by any reasonable definition of that word) who tries to argue otherwise really would be a wank.
For all intents and purposes he is arguing like a young earth flood geology baramin creationist.
Posted by: mvanbebber
|
August 17, 2010 11:59 PM
@CJO:
I don't think I'm overstating my case, because I thought we were talking about robot-driven vehicles. Well, robots can now drive vehicles - so the problem has been solved. Now, if you want to move the goalposts to include every conceivable interaction that a human driver might have, you basically need a human-level AI, which we don't have. Current vehicles can navigate city streets and obey traffic signs and laws, parallel park, and avoid dynamic obstacles as well as other cars (obviously.) That is amazing, and it used to be thought of as a purely human capability.
If you are only impressed with machine vision that can deduce its surroundings in 3-D, generalize new tasks and tell basic mental states of several human beings, then check out Leonardo, from MIT's robotics lab.
There is a long history, as Kurzweil points out, of people claiming that tasks can only be done by humans. Then, when they are done by machines, the same people claim "well, that wasn't that hard anyways."
Posted by: Forbidden
|
August 18, 2010 12:05 AM
As someone who writes code... that's one of the stupidest things I've ever heard. Code isn't data. You can't do simple math on it. Nevermind that the language has a huge factor in the size of code. What would take a couple lines of Perl could be hundreds of lines of assembly code.
I'll by that it's possible to study brain cells long enough to understand how they process information, simulate it, then connect however many million or billion (or whatever the number is) of processes equivalent to a number of brain cells and get a brain simulator.
It would be a hideously complex undertaking, but it's conceivable. How many lines of code it would be is absurd to even speculate when you cannot even properly define the parameters involved.
Posted by: MadScientist
|
August 18, 2010 12:11 AM
Screw that, the Linux kernel alone had over 6M lines of code back in 2006. My Linux machines are all still dumb as hell though, which happens to be the way I prefer it - that way the machine does as I tell it to, unlike MSWin systems which coerce the users into doing things the way its satanic master wishes.
Posted by: MadScientist
|
August 18, 2010 12:18 AM
@nigelTheBold#38: I thought that a mathematician (from Australia of all places) had actually proved that there is no solution to the multibody problem (3 or more) and that the search for a solution to the 3-body problem was subsequently abandoned by all.
Posted by: John Morales
|
August 18, 2010 12:22 AM
[OT]
MadScientist, your elitism is unfounded.
Operating systems are not applications, and you don't access the kernel directly, but the shell.
Which is an application. ;)
Posted by: John Morales
|
August 18, 2010 12:27 AM
MadScientist @389: Qiudong Wang.
Posted by: First Approximation, L'esprit de l'escalier
|
August 18, 2010 12:30 AM
I'm not sure what you're referring to.
From Wikipedia's article on the 3-body problem and n-body problem, respectively:
Posted by: puf-almighty
|
August 18, 2010 1:02 AM
1) If you could sequence the relevant proteins, couldn't you get their folding worked out through stuff like Foldit?
2) If you knew how the proteins acted, couldn't you simulate how a cell would act? I mean sure, hideously computationally intensive, couldn't do it for every one of the 100 trillion cells in the brain. But couldn't you get a sort of black-box approximation of how a neuron works?
3) Couldn't you then take that simplified set of rules and generalize it to modeling large groups of neurons, until you had the whole brain modeled?
Obviously this is not the same as what Kurzweil was saying. I'm not interested in his explanation cuz it's evident he's not well informed. I'm talking about practical ways to go about it, because it seems neat.
As far as I can tell, you'd lose a hell of a lot of data with all the simplifications- stuff where a single protein cascade can trigger something real, but which would be lost in a stochastic (I think that term's appropriate) model which didn't account for outliers.
Posted by: https://me.yahoo.com/a/QdqQbfpz0Iv82wOHRQdXyHU-#4b2c0
|
August 18, 2010 1:15 AM
Typo alert! You left out "no"
" The brain is a computer of sorts, and I'm in the camp that says there is problem in principle with replicating it artificially."
Posted by: BrianX
|
August 18, 2010 1:16 AM
puf-almighty:
1. Possibly. But that's a lot of computation.
2. Yes, to an extent, but a black-box approximation may not be quite enough, or even appropriate. We just don't know enough about the brain to answer that; biological systems don't usually lend themselves to black box modeling.
3. The key word here is "model". Yes, such a model is reasonable to envision, but it has its limitations, and if intelligence is an emergent property, there's no guarantee that a nonbiological model isn't going to wind up going on some bizarre tangents that don't quite resemble human intelligence.
There is a long standing debate in what's left of the AI community of "neat" vs. "messy" -- in other words, a planned top-down approach vs. a quasi-evolutionary emergent/bottom-up approach. I would think the "messies" have the best perspective on how it would work, and your example is a pretty neat summation of it.
Posted by: SaintStephen
|
August 18, 2010 1:26 AM
I like Ray Kurzweil. Of course he's part showman, just like you are Professor Myers. If all you did was write pedantic biology essays, Pharyngula would be a dried-up river bed within weeks. It takes a little razzle-dazzle to rally support from your troops (and rake-in the accompanying dollars to move on to bigger and better frontiers and projects).
What about your occasional posts/opinions on porn and "What is sexy?" Give me a break. I'd rather watch paint dry than listen to a married man wax not-so-eloquent on sex and pornography. You're the "kook" in this case, and you're doing it solely to generate hits/revenue on your blog. Tell me I'm wrong.
The brain is the most complex computer that's ever existed. It's from complexity that "consciousness arises, even though consciousness is just a silly word for something nobody understands. We might as well call it "cornstroodle." Artificial (non-biological) computers will eventually approach the complexity of the brain, on an exponential power curve resembling Moore's Law -- that's all Kurzweil is really saying. If one could ask a complex-enough computer any question and get a reasonable answer, then the Turing Test has been passed. The ape brain most likely has a lot of wasteful neural "pathways" in it, that aren't necessarily vital for actual "intelligence," just like the genome has a lot of "junk" DNA that isn't directly expressed in the resulting organism.
The complaint you're lodging is a protest against Kurzweil's perhaps overly-optimistic timescale for this event to take place. I'm absolutely certain Ray has never heard this particular gripe. (Cough.) Allow me to respond on Dr. Kurzweil's behalf:
ZZZZZZZZzzzzzzz...
Posted by: uninformedluddite
|
August 18, 2010 1:40 AM
The only AI that replicates the human mind to a T is currently hidden behind all the dark matter and held in place by a singularity gravity wave. This artificial brain construct will feature soon balancing equations that require gullibility, faith, and an uncritical audience of mindless TV watchers.
Posted by: Greylander
|
August 18, 2010 1:44 AM
@ KG #235,
Hmmm... this is news to whom?
"Always"?? why? The only reasons to leave something out are: (1) don't need it, not important; (2) you don't know about it(or make a mistake); (3) you lack resources to include it (programmer time, computational capacity). Are you trying to claim that one or both of (2) or (3) are always true?
This is just idiotic. Of course a simulation will not "do" the actual physical things of the real thing. A computer simulation of a combustion engine will not provide propulsion to a real automobile. And if, for example, you leave out of your simulation the gradual wearing down of the moving parts and accumulation of crud, etc., then your simulated engine will run forever which will not emulate the gradual breakdown of the real thing.
So. Freaking. What?
If you put those details into the simulation and have the computing power to handle it, then your simulation will emulate those things as well.
Perhaps you were too blinded by the "obvious" to fully read my post. I covered a variety of the kinds of variables that need to be tracked, including such things as concentrations of each type of neurotransmitter in the synapse, number of each type of receptor, voltage levels at various points in the cell, concentrations of various chemicals in the ambient environment, surface area of the proximate surfaces of the synapse, and so forth.
What should be blindingly obvious that in posts like these, even very long posts, on a subject this involved, it is possible to at best summarize many points. Just because someone does not address every issue in detail in a post does not mean they do not know or have not considered whatever it is you think they have left out.
You might fair better in this argument if you try to make salient points relevant to the questions at hand. I will refrain further from rising to your ad hominems.
You keep harping on the growing/changing synapses as if they are some mystical thing impossible to simulate, or as if it is going to be impossible to distill the variables relevant to information processing. Yes, I know about them, yes, I know they are fundamental to learning and cognition. Gee whiz, how impressed I am that you know it too.
The question is how much of this detail has to be simulated to capture the information processing capabilities of the brain? Are we going to have to have model the surface of each neuron with a wireframe mesh? To what resolution? Or will a more abstract model suffice, leaving out these physical details? It seems to me likely that we will, for example, need to model the axons and dendrites as abstract tree structures, with a few parameters describing each segment (conductivity, voltage at each node, "currents" of various neurotransmitters and other substances, etc.) and many more parameters descibing each synapse (leaves of the tree), I mentioned many possible parameters already). And yes the simulation would allow this tree structure to grow/shrink/change. It seems highly likely that the hiearchy of the tree structure will be relevant to the computations performed by a neuron (vaguely analogous to expressions nested in parentheses in mathematical expressions).
Yes, I know. That would be one obvious parameter for each synapse: "Type". How many discrete types are there? Mmmmm... let's say a million. That'll require about hmmm... 3 bytes of data to code "type" for each neuron. How many byte of data did I suggest per neuron? 1,000 to 10,000? Hmmmm... I think I have "type" covered in my estimates.
Yes, I know. I have taken this into account as well. Did you not read where, for example, I mentioned local ambient concentrations of various signaling chemicals? Those things come from and are detected by neurons and other cells, they influence what happens at the synapses, and so forth. We will need to, for example, model the diffusion of these chemical signals with partial differential equations on a grid with resolution in the ballpark of the size of a cell. Depending on how many different substances need to be tracked, and the precision to which we need to know their concetrations, we are still talking in the neighborhood of 10^14 to 10^18 bytes, which does not change the order-of-magnitude estimate I made of 10^19 bytes to describe the information processing state of the entire brain.
We are talking about simulating the information processing of the brain, more specifically the information processing done in the cerebral cortex. The brain "does" a great many things other than information processing in the sense relevant to what we call 'thought'.
Ah. Finally we get to the heart of your objections. "It's just too complicated and we've barely scratched the surface!" If you read my first post, you might have noticed that I covered this.
I do not think you appreciate how much faster computers are going to get over the course of a single decade, nor the impact that technologies like "lab-on-a-chip" are going to have on the rate of data-collection of all kinds of information of what goes on at the cellular and molecular level. Will it ramp up fast enough to give us all the info we need to simulate human cognitive processes sufficiently well to pass a Turing test within a single decade? I actually do not think so, I push my estimate out a couple more decades.
That is why we are gathering data... and the rate at which we are gathering data is growing at an exponential rate.
Can't? Why not? Do you mean today? Next year? Five years from now? Ten years from now?
Posted by: Rob Seaman
|
August 18, 2010 1:55 AM
BrianX asserts: "when dreaming, the human brain is active. When an AI's power is off, there is no activity at all."
This is "begging the question" twice over. Can an AI be powered off? Can an AI exist at all?
"Artificial intelligence" is a homologous assertion to "intelligent design", and equally fallacious. Human consciousness is the result of natural selection operating over cosmic timescales.
Perhaps SETI is a waste of time (oh, sorry...that's another thread), but what we can be confident of is that any alien intelligences that we do encounter will have evolved naturally over similarly lengthy periods of time.
Posted by: John Morales
|
August 18, 2010 1:58 AM
SaintStephen:
This again? Have you even read the OP or the comment thread?
Sigh.
Posted by: Ichthyic
|
August 18, 2010 2:03 AM
The complaint you're lodging is a protest against Kurzweil's perhaps overly-optimistic timescale for this event to take place.
*looks*
Kuzweil sez:
"The design of the brain is in the genome."
PZ highlights it with a bright red pen.
conclusion?
Kruzweil fanbois must be colorblind.
Posted by: chronos.tachyon
|
August 18, 2010 2:08 AM
Ugh, Kurzweil. I say "ugh" as someone who thinks that within my lifetime there's a measurably non-zero chance of medical immortality and a good possibility that we'll implement a Friendly AI. Uploading is just silly, though, and I consider cryonics a long shot compared to medical treatment of aging as a disease... and even uploading is plausible compared to Kurzweil's wackier beliefs.
Niggle: "emergent" and phrases containing it get heavily abused, so it's best to avoid the word entirely. (If no chrysalides are involved, "to emerge" is a synonym of "to arise", e.g. "behavior of an ant colony emerges/arises from the behavior of individual ants". Any phrase where the "arise" substitution doesn't make sense, e.g. "arising phenomenon", is probably a meaningless buzzword usage of "emerge".)
Posted by: John Morales
|
August 18, 2010 2:10 AM
Greylander, such a long post responding to KG.
A few of the immediately-apparent issues regarding your proposal seem not entirely inconsequential:
When do you see techniques sufficient to acquire the data-state of the variables for your proposed simulation being available?¹
When do you see architecture adequate to run your proposed simulation being available?²
What about the expertise to code it?³
What about the run-time requirements?
--
¹ Would this be a destructive process? How long might the process take (noting state-changes during the scanning process might be an issue).
² Might there be heat dissipation issues?
³ How many "lines of code", to use a brilliant metric?
Posted by: DrivenB4U
|
August 18, 2010 2:25 AM
Well, at least the guy could make a mean synthesizer.
Posted by: amphiox
|
August 18, 2010 2:25 AM
Hmm. Well, if nothing was left out at all, then it wouldn't really be a simulation anymore, would it? It'd be an exact copy of the real thing.
Posted by: amphiox
|
August 18, 2010 2:28 AM
If you really had a human level AI, then turning it off, even if you saved a copy, is tantamount to murder.
Posted by: HertfordshireChris
|
August 18, 2010 2:34 AM
Forbidden # 387 says
What everyone is missing is that the amount of “code” you need depends on the architecture of the electronic box you are using - as well as any associated software. When computers started no-one every questioned whether the philosophy underlying the stored program computer was THE ONLY POSSIBLE UNIVERSAL architecture. - they just got on and did it. If you go back to first principals and try and design an electronic box in which the starting point is the set and partitions of a set, rather than numbers, you find you no longer have to artificially partition knowledge into “code” and “data”. Much of the circular arguments here are because many people take it for granted that the only possible modelling tool is a computer. After all they were taught about computers at school and built their careers on them - so the foundations of computer technology must be "best" for every task. Religious extremists use the same argument to justify their religion.
If you go to a different underlying architecture, more suitable for modelling how human's think, any measure which assumes that knowledge must be split into code and data is vacuous. We just dont work that way - and because existing computers have been so successful in blocking potential opposition the alternative architectures are not well known because they are not commercially available.
Posted by: Kagato
|
August 18, 2010 3:08 AM
Depends entirely on the implementation.
If AI necessarily had to run in some form of volatile analogue memory, that couldn't be paused or powered down without losing state, then tampering with the host platform would effectively be "AI murder".
If an AI was basically running on a digital simulation of a brain, then execution of the simulation could be paused, then resumed from exactly where it left off. From the AI's perspective, it would have jumped forward in time in an instant. I guess the medical equivalent would be an induced coma.
A grey area is if, while paused, the AI was persisted to storage then powered down. Strictly speaking, you're not running the same AI any more, even though there would be no difference from either the AI's or an observer's point of view.
Running two copies of the persisted AI makes things even more interesting...
One thing people tend to forget when talking about "transferring their consciousness" or whatever -- data is not moved, it is copied (and deleted). Even if it was possible to upload your mind into a computer before you died, you would still die; your duplicate would "live on" in the computer. If digital-you then downloaded into a new body, the body would get up and walk away, but digital-you would remain trapped in the computer until you were switched off...
Posted by: John Morales
|
August 18, 2010 3:20 AM
Permutation City.
Posted by: forrest.pugh
|
August 18, 2010 3:43 AM
Granted there are 400+ comments so far, so I may be repeating a point made already - but I did scan through at least a quarter of the posts...
I do think that what Kurzweiler believes can be done in theory, but it is technically unfeasible at this point
Let's presume that part of the code for the brain not only contains the architecture, but the bios and OS in there as well. After-all there are rudimentary biological functions which must be automatically regulated for life to be maintained, then there are instincts among many members of the animal kingdom which let them behave similar to their direct ancestors untaught - this would be like an O/S.
But the thing that makes an individual unique is all the stored data - the memories, and how they are related. Anyone that understands the differences between constructive or object-oriented programing vs deductive or rational query programming can relate to the challenges involved not only in recreating the files correctly upon the hardware which was built in the same process, but in knowing the data-relations involved that link all the data tables, or neuron clusters.
Now we know synapses store that data, and we can even find out which ones are a cluster by evoking a memory and probing the firing pattern. Given time we would even be able to recreate this exact construct synthetically.
I'm giving Kurzweil some allowances on his apparent understanding of dna here - we presume that we can sample each synapses DNA strand here because (if my understanding of cellular neurobiology is clear) each neuron - each cell - each DNA strand is going to be slightly different than the next one over because each one will have code to connect to a different paired set of neuron than the next.
This is where it gets tricky, because while I might be able to sample perhaps 10% of the total unique DNA strands in a brain and analyze the similarities to a 99% degree of accuracy; that 1% potential difference in each unique strand accounts for 90% of the total difference that makes that particular configuration of neuron clusters a person.
There just is not now nor in the foreseeable future a technology that can sample 100% of the brain's DNA and artificially duplicate it.
Posted by: Andyo
|
August 18, 2010 3:54 AM
Oh you silly skeptics with your "reading" and all that bullshit. And by the way, don't you naysayers have any other argument than to tell us to actually read whatever we're criticizing? Tsk Tsk.Posted by: robin-berjon.myopenid.com
|
August 18, 2010 3:55 AM
Actually, it doesn't even make sense from a programming point of view either. A line of code is not compressed, so even if the rest wasn't simplistic, you'd still need at least (by his reckoning) a sixteen-fold larger code-base. And code complexity, no matter one's best efforts, doesn't tend to grow linearly.
Posted by: KG
|
August 18, 2010 4:49 AM
HertfordshireChris,
You obviously have a deep regret that your "sets and partitions" idea never got off the ground, and it may be very interesting, but it's not at all clear it's relevant here. The basic problems with brain emulation would be just the same if the intended implementation device was made of angels jumping on and off pinheads (this hardware idea is not mine, but Maggie Boden's).
Posted by: KG
|
August 18, 2010 5:10 AM
- GreylanderI can't really be bothered with a detailed response to your crap any more. You are making exactly the same sort of promises as some of the proponents of the human genome project made about its medical implications, and with even less justification. Now in that project, we saw an enormous improvement of computational and automation technologies, allowing the genome to be sequenced in less time than appeared possible only a few years previously. And the medical results have been almost non-existent so far, because it turned out that knowing the sequence of bases told us effectively nothing relevant to clinical practice or even specifically medical research. Sure, it was well worth doing, and eventually it will have a big payoff, but what it proved to anyone with more than a handful of functioning neurons is that brute computational power doesn't enable you to understand a complex system. We don't know how many kinds of synapse there are, or understand the non-synaptic ways neurons communicate, or know what chemical concentrations are important, or when and how synapses grow or atrophy, or, most importantly, as others have said, what we don't even know we don't know. It is going to take neurobiologists many decades, possibly centuries, to acquire the necessary knowledge, because it has to be done step by step - formulating hypotheses and designing and implementing models, testing them, discarding or modifying them, arguing about them, publishing on them, applying for grants to buy new equipment - you know (well actually, I guess you don't) - science as it is really done, as opposed to how it is done in Singulatarians' imaginations.
Posted by: KG
|
August 18, 2010 5:32 AM
BTW, Greylander, and other Singulatarians posting here, I would love to believe in your fantasies, I really would. After all, if you're right, all the world's serious problems (AGW, resource depletion, war, etc.) are readily solvable within the next few decades, and I could get to upload my mind and enjoy an indefinite lifespan in a techno-paradise (I'm a bit younger than Kurzweil). What's not to like? Trouble is, I just can't, any more than I can believe in benevolent deities. (Yes, I have read The Singularity is Near: "the Rapture for nerds" is an apt description.)
Posted by: Ian Gould
|
August 18, 2010 5:46 AM
This thread is already 400+ comments long, so apologies if I've skipped over a post where this point is already made.
Even going beyond PZ's quite valid point about the complexity of the proteome, even if you succeed in replicating a brain in isolation all that will get you is something as similar to a functioning living brain as a brain preserved in formaldehyde.
To get a functioning brain you'd also need to replicate the uterine environment, the influence of the rest of the nervous system and the chemicals produced by the rest of the fetus and the inputs from the external environment.
As PZ says none of this is an argument for vitalism of brain-body dualism, it's an argument for recognizing the complexity of the human brain and for more sophisticated approaches to replicating its capacities than a brute-force molecule-up approach.
If you can replicate the mature neural net that results from those chemical processes why bother with the underlying chemical processes?
It's like trying to build a plane that hatches from an egg.
Posted by: Daniel B
|
August 18, 2010 6:27 AM
KJ
I was assuming, as I think the author does, that the 'hard part' was re-creating the biology/mechanics of the brain, not training it. Rest assured that if we can create an artificial brain we will be able to train it.
We already have a lot of experience at training human brains and there is no reason why the world that is available to children couldn't be brought to an artificial young brain. Wiring up a sensory mechanism to provide experience would be trivial compared with creating the brain itself. Even severely sensory deprived children can develop intelligence so there's no reason to the think that an artificial sensory experience would be a show stopper. The difference with the digital version would be that you could take snapshots of the development all the way along so you could always go back if things weren't working out.
To restate my point just in case you missed it - All this information is also available to assist with the formation of an artificial brain.
Posted by: KG
|
August 18, 2010 6:36 AM
Daniel B.
You miss the point, of course, which is that the "biology/mechanics" of the brain are produced by prolonged interaction with the environment, in ways which are very far from being understood, but which involve extensive "sculpting" by removal of synapses and neurons. It will not be possible to replicate this without understanding the developmental processes involved.
Posted by: KingUber
|
August 18, 2010 6:52 AM
So I have to wait even longer for my robot wife? :(
Posted by: First Approximation, L'esprit de l'escalier
|
August 18, 2010 7:39 AM
Someone upthread told us to prepare for many comments by Kurzweil followers who didn't appear to have read the post or any of the comments that followed. A good predication.
Posted by: pauljthacker
|
August 18, 2010 8:00 AM
I'm certainly not suggesting this is an easy problem. But is gathering this immense amount of information as intractable as you seem to think? To get back to Kurzweil, he's a big proponent of the idea of exponential growth. Think of how much more quickly and cheaply we can sequence a genome since it was first done. If our ability to gather the needed biological information continues growing at this rate (not a certainty, but plausible given historical trends), we could indeed know order of magnitudes more in the next decade or few decades.
Posted by: Steve LaBonne
|
August 18, 2010 8:16 AM
[cough]Darwinian evolution[cough] (But this is no help for the kurzweilbot project.)
Seriously, you don't even know how little you know about what you're talking about. It's tiresome.
Posted by: poke
|
August 18, 2010 8:57 AM
Greylander (#398):
That's not a sensible question. The brain is a natural system. It is not an implementation of a program we can reverse engineer. You can enumerate a list of relevant metrics in the brain in order to create a model but at some point it will turn out that some other, neglected, aspect of the system is relevant to a given behaviour because every aspect of a natural system is relevant to its behaviour.
I think you're too enamoured with the computer analogy of the brain (and it is just an analogy). We can talk about the exact information processing capabilities of a computer program because it is designed according to a set of specific criteria. The brain is not like this. You're never going to get a 'complete' description of the brain and there's no such thing as a complete description of just the brain's 'information processing capabilities.' Simulations do not 'capture' the real world, they model it, which is a form of description or depiction to a relevant level of detail for a given purpose.
'Thought' is not some abstract function that is imposed on the architecture of the brain from without. The brain is not an implementation of thought, or the mind, or consciousness or anything. There isn't a subset of brain activity or architecture relevant only to 'thought' that can be set apart from all gritty details of the whole brain as a natural system.
The problem isn't the details. The problem is that it's utterly meaningless to talk of an 'accurate' simulation in the sense of one that's 'like the real thing.' A simulation can only be accurate to a given purpose. Most likely simulations would be made for the purposes of scientists trying to understand some aspect of the brains global architecture. They'll perform limited simulations of those aspects that interest them. This is true of all models in science.
Posted by: tim333
|
August 18, 2010 10:08 AM
Gosh 400ish comments!
To defend Kurzweil:
I think he said 20 year not 10.
He has no intention of simulating proteins - the genome thing was just a rough estimate of the design complexity.
Check out what he actually has to say at
http://www.youtube.com/watch?v=15sh05wrQ6Y
Start about 16 minutes in if you are in a hurry.
Does not seem very kooky to me.
I rather doubt pz has actually seen this stuff and would probably have written differently if he had
Posted by: Flex
|
August 18, 2010 10:19 AM
forrest.pugh @410 wrote,
And it would be nice if human memory was anything like computer memory.
You do realize that current thinking about how human memory works is based on the idea that memories are re-creations of events through neural processes, not bytes of data in ROM?
But even that is still part of a testable theory and we don't really know how human memory works, other than we know it doesn't work like computer memory.
When AI is developed, humanity is likely to be unable to identify it. Think of Shalmaneser from Brunner's novel, Stand on Zanzibar. Chad Mulligan recognizes that Shalmaneser has developed AI, but no one else in the novel does. (And surprisingly, while looking for the name of the computer at various on-line reviews of the book, it looks like most reviewers missed this point too.)
Posted by: Cantwell
|
August 18, 2010 10:40 AM
With due respect to above commenters:
I don't think human life (which is really what we are talking about) will ever be duplicated or replicated in silico even if we could provide Kurzwiel with an arbitrarily large amount of computing power, storage, research funding, man hours, and time. As PZ says, he doesn't understand biology, sure. But the challenge is bigger than just biology. He would have to understand the nature and purpose of human consciousness well enough to create a faithful representation of the genuine article. Kurzwiel can't do that and neither can anyone else.
For example, if such a thing was claimed, how could it be proven or disproven? Would a simulcra of me be built to see if my friends and family could distinguish it from me? If such a thing could be built, how would one verify the extent to which its actions would have been identical to my own? If my family were replaced by such replicants, to what extent would their interactions with each other be "emergent" and to what extent would they be pre-programmed? Would this emergent-to-pre-programmed ratio be faithful to that which exists in real life? To what extent would the prejudices and shortcomings of the makers emerge in the attitudes of their creations?
There are no good answers to these questions because we have no answers for life, at least not answers that can be encoded and compiled.
To the cryonicists, a few things: If Alcor accidentally dumps your lifeless heads in a landfill, who will be there to file suit for negligence on your behalf when you are dead along with everyone you've ever known? Sorry, but a business where people pay $100k and willingly divest the recipient of any liability is probably a scam. Second thing, do you really want to wake up in a world where everyone you've known and loved is dead? Does your immediate family matter so little to you that you enjoy the thought of immortality without those bonds to tie you down? Third, remember that you could buy something really fancy with that money now, and they you would get well-defined utility out of it in the present moment, rather than using it to pay for a chance to wake up to an existence bereft of friends and loved ones.
Posted by: grudgedk
|
August 18, 2010 10:41 AM
You forgot "providing that neuron interaction works deterministically", which you have no evidence that it does. This is untrue. Computer "scientists" are not particularly better at understanding complexity than any other type of scientist. They have tools for breaking down complexity into simpler models, much like any other field of science has. The reason for this of course in inherently obvious. We can't handle complexity very well. If we were some type of superhuman complexity solvers, we would still be using assembly and punch-cards, and would have no need for high-level languages, data-structures or OOAD. Because apparently there aren't enough computer "scientists" working on it. It kind of does. You're making an assumption on the nature of "neural information" that is unlikely to be correct. On the other hand, where is my flying car and personal jetpack? I mean the technology for VTOL aircraft has been around for half a decade. Still no flying cars. Explain this to me. No. That's what you're saying. What he's saying, to use your gun metaphor, is that pulling the trigger, isn't the best way to determine whether a gun is loaded or not. Well the scope of human intelligence is pretty broad. Cleverbot for example is already much more intelligent than say your average teabagger. Wishful thinking is not evidence. Liquid Nitrogen is very cold, at about 70K, it will destroy pretty much any living organism it comes into contact with. They can't even "cure" frostbite, what possible reason could you possibly have for believing they can reanimate a corpse?Posted by: KG
|
August 18, 2010 10:52 AM
grudgedk,
Here's my initial conversation with "Cleverbot" (my contributions in italics):
Who are you?
I am you.
No, you're not.
Yes, I am.
What makes you think so?
I know so!
I'm unimpressed so far - ELIZA did this 40 years ago
You knew spanish 10 years ago? But now you don't?
Posted by: mvanbebber
|
August 18, 2010 10:53 AM
Good point, I can't believe no one has pointed that out yet. Kurzweil is NOT claiming 10 years, but instead 20 (by 2029) for a reverse-engineering of the human brain. Most people here are throwing around the '10 years' number as if that is what Kurzweil estimates. Not so.
Posted by: mvanbebber
|
August 18, 2010 11:05 AM
@grudgedk:
Because only information technology is exponentially increasing over time, not every single technology. I thought you would have an understanding of this, since it is the basic argument being discussed here. That's like saying "Where is my steam-powered spaceship? I mean the technology for steam engines has been around for 2000 years. Explain that to me"
Well, I didn't I had to "explain that to you" because I thought you understood the basic premise of exponentially accelerating technologies.
Posted by: KG
|
August 18, 2010 11:06 AM
grudgedk,
Here's my initial conversation with "Cleverbot" (my contributions in italics):
Who are you?
I am you.
No, you're not.
Yes, I am.
What makes you think so?
I know so!
I'm unimpressed so far - ELIZA did this 40 years ago
You knew spanish 10 years ago? But now you don't?
Posted by: lexicakes
|
August 18, 2010 11:13 AM
It's on MetaFilter now too.
Posted by: Paul
|
August 18, 2010 11:15 AM
Well, I may not post a lot, but I've lurked for a long time -- through several Kurzweil-bot attacks. They don't change much.
Posted by: mvanbebber
|
August 18, 2010 11:23 AM
Not only does PZ misrepresent Kurzweil's estimate, which is actually 2029 for a brain simulation, but so does PZ's "source material." The Gizmodo article completely gets the time frame wrong and states explicitly that Kurzweil says we will simulate the human brain in 10 years. Yet more evidence that PZ has no idea what Kurzweil's position even is, much less the underlying technologies that will allow that to be accomplished.
Posted by: Paul
|
August 18, 2010 11:31 AM
Oh hey, look, they finally read the OP. How about addressing how extrapolating lines of code necessary to simulate the brain from the number of base pairs in the genome is...well...not even wrong -- it's pure kookery. That was really PZ's main point.
Posted by: KG
|
August 18, 2010 11:33 AM
mvanbebber,
Information technology has been progressing fast in large part because ever-increasing resources have been poured into it. This cannot continue indefinitely - because we still need to eat. Kurzweil's fantasy is that IT will expand to encompass effectively the whole economy. It won't.
It's interesting to look at the "2010 scenario" in Kurzlooney's 2005 publication: The Singularity is Near (pp.312-3). Here are some of the predictions (note that these are all made in a way that suggests the devices and services mentioned will be commonplace):
"Displays will be built into our eyeglasses and contact lenses and images projected directly onto our retinas."
"We'll have very high bandwidth wireless communication to the Internet at all times."
"These resources will provide high-resolution, full-immersion visual-auditory virtual reality at any time. We will also have augmented reality with displays overlaying the real world to provide real-time guidance and explanation. For example, your retinal display might remind us, "That's Dr. John Smith, director of ABC Institute - you last saw him six months ago at the XYZ conference."
"We'll have real-time translation of foreign languages, essentially subtitles on the world, and access to many forms of online information integrated into our daily activities. Virtual personalities that overlay the real world will help us with information retrieval and out chores and transactions. These virtual assistants won't always wait for questions and directives but will step forward if they see us struggling to find a piece of information."
OK, where's my always-available immersive virtual reality? My high-bandwidth wireless communication to the Internet available at all times? My retinal display which can recognise individuals? My real-time foreign language translation? My virtual assistant who can see I'm struggling for a piece of information? (No, the Microsoft paper clip does not count - the example given is of "That actress... who played the princess, or was it the queen... in that movie with the robot.")
So Kurzlooney managed to grossly over-predict over a span of 5 years (6 maybe, given the publication delay), for technologies far, far closer to reality than brain emulation.
Posted by: mitsu.hadeishi
|
August 18, 2010 11:36 AM
I don't have the time to read the entire set of hundreds of comments here, but I want to make one simple point: while Kurzweil may well be wrong, this post oversimplifies his argument. Kurzweil is not making an argument about the ease of *simulating gene expression* (a problem which everyone knows is incredibly hard, and may be impossible in general, as it requires the running of exponentially complex quantum field calculations) --- he's making a totally separate computational complexity argument. That is to say, Kurzweil may be wrong but he's not as stupid as this blog post makes him out to be. The basic argument has to do with algorithmic complexity theory. He's saying that the lower bound of the number of bits required to capture the algorithmic complexity of what is needed to generate the brain would be the genetic information. Obviously he's NOT saying that in 10 years we would literally try to build a genetic simulator that could take the genome and generate a functioning brain.
He may be wrong just because the compression algorithm used to encode the brain may rely on quantum field effects which, who knows, might allow for compression beyond that which a classical computer is capable of (i.e., it's known that biological systems can take advantage of quantum effects, for example a recent paper that showed that plants take advantage of quantum computation demonstrates this). Quantum computers are of course capable of computational feats that would defeat a classical computer, so his estimate could end up being very wrong.
And even if he is correct, he may be grossly underestimating how long it might take for human beings to build algorithms with sufficient "compression" to generate brain-like behavior. But my main point is that most of this blog post is mostly beside the point, as it addresses an argument Kurzweil has not been making; i.e., the post is conflating algorithmic complexity with the difficulty of simulating gene expression, two totally different points. If it were the case that the important aspects of gene expression were *in principle* simulatable classically, then Kurzweil would be absolutely correct, and it would not matter whether we actually simulated gene expression or used totally unrelated algorithms.
Posted by: mvanbebber
|
August 18, 2010 11:50 AM
We have all of that, as consumer products or working in labs. Do the slightest bit of research before you claim something doesn't exist. About VR, it depends on what you mean by "immersive"- that is an unclear prediction. We have virtual assistants, but they are terrible.
Posted by: PZ Myers
|
August 18, 2010 11:51 AM
No, he is as stupid as this post makes him out to be.
He is making a calculation about how much information needs to be generated to simulate an adult brain, based on only the genetic information he estimates to be present in the embryo.
That's dead wrong. It assumes there is no increase in the information content of the brain during ontogeny. The only way his scheme would work is if he also simulated development to generate his final brain simulator.
Stupid, stupid, stupid.
Posted by: PZ Myers
|
August 18, 2010 12:00 PM
One other thing: 10 years, 20 years, it makes no difference. I've been watching the field of neuroscience for 30 years, during a period when the discipline of molecular genetics has been fully embraced, and there has been a huge amount of new information and new techniques generated...but nothing like the magical revolution Kurzweil predicts.
I predict that in 20 years there will be new and amazing things that science will be able to accomplish with the brain, but we won't be seeing what Kurzweil predicts at all. Brain scans and complete cortical simulators and consciousness uploads sound cool, but are unlikely; I'd be content with effective permanent treatments for schizophrenia and implants to compensate for blindness or epilepsy, with a few pipe dreams about artificial intelligence augmentation. That sort of thing.
Posted by: mitsu.hadeishi
|
August 18, 2010 12:11 PM
>It assumes there is no increase in the information content of the brain during ontogeny
I'm not sure what you mean by "information content" but you're missing the point. What Kurzweil is referring to is Kolmogorov complexity. If you can encode the algorithm needed to generate the brain in X bits, and if ontogeny is a classical process, then a reasonable upper bound for the Kolmogorov complexity could indeed be the genetic information. Sure, as the brain grows and so forth it may get more "complex" but the same could be said for the operation of any program in its interaction with its environment.
But the whole point I am making is that talking about the difficulty of simulating gene expression is really beside the point. The issue is Kolmogorov complexity.
Posted by: KG
|
August 18, 2010 12:19 PM
We have all of that, as consumer products or working in labs. Do the slightest bit of research before you claim something doesn't exist. -mvanbebber
Are you being stupid, or dishonest? As I said, Kurzlooney makes clear he thinks that these things will be commonplace in 2010 - there is not the slightest hint that they will simply be "working in labs". Kurzlooney himself notes that some of the things he's "predicting", including immersive virtual reality, were already in use by the Army when he wrote the book - so it's absolutely clear he is predicting common use. For that matter, several of his predictions aren't anywhere near "working in labs". Real-time automated translation is not even near practical use. Nor are systems that can recognise individuals under everyday conditions. Nor are virtual assistants that can take the initiative. It's very telling that you have to resort to this sort of bullshit.
Posted by: BrianX
|
August 18, 2010 12:20 PM
You know, now that I think of it, supporters of the Singularity and cryonics remind me quite a lot of Velikovskyites.
When Immanuel Velikovsky's work first appeared, a good number of laypeople were impressed. But something interesting happened when his writing got to the people whose territory he was drifting through -- while historians seemed impressed with his astrophysics, they thought his history and mythology were a load of crap. Astronomers thought much the same -- the history part seemed interesting, but his concept of planetary dynamics was complete gibberish. It didn't take very long for both sides to realize Velikovsky was out of his mind.
In this thread, Kurzweil's prediction has been torn apart by someone who actually knows something about biology, and the Kurzweil supporters are howling with a mixture of courtier's replies and blatant faith-based special pleading. That's not the mark of a truly rational person.
Posted by: KG
|
August 18, 2010 12:24 PM
mitsu.hadeshiNo, it couldn't - not by anyone with any sense.
Posted by: PZ Myers
|
August 18, 2010 12:25 PM
No, I'm afraid you're missing the point.
What is the decompression function? You need to know that algorithm before you can turn the 'compressed data' of the genome into your functioning brain simulation.
I'm telling you that that algorithm is called neural development. You're doomed to failure if you don't understand it.
It is also not as simple as you think. It's not as if the genome provides the complete specification for the end product. We also need to take into account environmental influences: for instance, in mammals the dorso-ventral axis seems to be specified by interactions with the endometrium of the uterus. That information is not present in the embryo's genome. The information is provided by an additional external source.
The whole problem here is a whole lot of people who are seeing only zygote going in to a black box, and a brain coming out...and pretending that you don't need to know how the black box works to simulate it.
Posted by: theswede
|
August 18, 2010 12:28 PM
He's saying that the lower bound of the number of bits required to capture the algorithmic complexity of what is needed to generate the brain would be the genetic information.
So he's saying that the complexity of an algorithm can be estimated by how large the argument it takes is when compressed?
And you claim he is NOT stupid?
Posted by: mitsu.hadeishi
|
August 18, 2010 12:32 PM
I'm not saying it is simple, I'm just trying to make a distinction between building a simulation of gene expression and trying to estimate the Kolmogorov complexity. The two are really quite different problems. The fact that it is almost certainly going to be impossible to build a gene expression simulator in anything like the next ten, twenty, or even maybe fifty or a hundred years is not what is at issue here.
And sure, the decompression algorithm might make things far more complex. However, the fact is, the laws of classical physics are in fact relatively simple to compress (i.e., you can express them in a relatively simple Hamiltonian). So the idea would be that however complex the problem of building a gene expression simulator that is actually realistic, the problem of the Kolmogorov complexity could actually be the Kolmogorov complexity of classical physics + the Kolmogorov complexity of the genome. Again, we're talking about in principle, not in practice what it would take to really build a simulator.
Posted by: mitsu.hadeishi
|
August 18, 2010 12:34 PM
I should add --- I don't necessarily disagree with you that Kurzweil may well be --- or even is probably --- wrong that we'll be able to simulate human AI within our lifetimes. I'm just saying his argument is not about literally simulating gene expression as it actually occurs in nature, but a different argument about Kolmogorov complexity.
Posted by: KG
|
August 18, 2010 12:39 PM
mitsu.hadeshiWhat is supposed to be the relevance of the Kolmogorov complexity?
Posted by: KG
|
August 18, 2010 12:41 PM
The Singularity is imminent, and Kurzweil is its prophet!
Posted by: mitsu.hadeishi
|
August 18, 2010 12:42 PM
http://en.wikipedia.org/wiki/Kolmogorov_complexity
Posted by: KG
|
August 18, 2010 12:45 PM
I didn't ask what Kolmogorov complexity is - I know that. I asked what its relevance is.
Posted by: JV99
|
August 18, 2010 1:12 PM
How do you simulate the brain with a Turing machine that does not have a memory structure like the brain? All indications are that a functional brain does not work like a "hunt and find" index of data with a central processor. Kind of hard to build a model airplane out of a peanut butter cookie.
Posted by: BrianX
|
August 18, 2010 1:22 PM
JV99:
It's not impossible, but it would be painfully slow.
Posted by: CJO
|
August 18, 2010 1:41 PM
I don't think I'm overstating my case, because I thought we were talking about robot-driven vehicles. Well, robots can now drive vehicles - so the problem has been solved. Now, if you want to move the goalposts to include every conceivable interaction that a human driver might have, you basically need a human-level AI, which we don't have. Current vehicles can navigate city streets and obey traffic signs and laws, parallel park, and avoid dynamic obstacles as well as other cars (obviously.) That is amazing, and it used to be thought of as a purely human capability.
I'm not moving the goalposts, you set them up impossibly far away before I ever brought up robot vehicles when you said that machine vision to human levels of competence was "essentially solved." After I brought up robot navigation, it occurred to me that the kind of robot vehicles you are talking about don't use machine vision for some of the key tasks involved in driving that humans must use their vision for: to wit, establishing distance from obstacles and other vehicles, and "perceiving" relative location. The robots use lasers and GPS. Reading traffic signs is relatively trivial, as machine vision systems are currently working in all kinds of 2D applications.
I want to make it plain that I think the achievements made in this field are amazing too, I just don't go off half-cocked and claim that they amount to solving machine vision, because they don't.
Posted by: chay.donohoe
|
August 18, 2010 2:05 PM
Oh my, how on earth did RK come to this conclusion? it like saying a car engine is made out of (mostly) aluminum, therefore if I can define the subatomic structure of an aluminum atom, there will be enough information there to define the structure of a car engine. I seriously hope he was mis-quoted.
Wikipedia on Emergence (The key word is 'Irreducible'):
An emergent behavior or emergent property can appear when a number of simple entities (agents) operate in an environment, forming more complex behaviors as a collective. If emergence happens over disparate size scales, then the reason is usually a causal relation across different scales. In other words there is often a form of top-down feedback in systems with emergent properties. The processes from which emergent properties result may occur in either the observed or observing system, and can commonly be identified by their patterns of accumulating change, most generally called 'growth'. Why emergent behaviours occur include: intricate causal relations across different scales and feedback, known as interconnectivity. The emergent property itself may be either very predictable or unpredictable and unprecedented, and represent a new level of the system's evolution. The complex behaviour or properties are not a property of any single such entity, nor can they easily be predicted or deduced from behaviour in the lower-level entities: they are irreducible. No physical property of an individual molecule of any gas would lead one to think that a large collection of them will transmit sound. The shape and behaviour of a flock of birds or shoal of fish are also good examples.
Posted by: mvanbebber
|
August 18, 2010 2:10 PM
So now you admit that you got the prediction wrong by double and you say it doesn't matter? Well, with exponentially expanding tools, it most certainly does matter! That's the difference between our experimental and manipulative tools being 1,000 times greater than today, versus those tools being 1,000,000 times greater than today. And you are saying that "doesn't matter"?
Stupid, stupid, stupid. Uninformed, uniformed, uninformed.
Posted by: Rob Seaman
|
August 18, 2010 2:11 PM
Tim333 suggests: "Check out what he actually has to say at http://www.youtube.com/watch?v=15sh05wrQ6Y
Start about 16 minutes in if you are in a hurry. Does not seem very kooky to me."
Kooky? No. Interesting and insightful? Even less so. The guy himself doesn't seem very excited about his own platitudes and generalities.
Even subtracting out the cult of personality (well, what passes for it) and the soft-sell money-making messianic message, listen to what the guy is NOT saying, not just to the words he chooses for his own purposes.
Talk to an astronomer and you'll hear the excitement of the unknown - dark energy and dark matter - but also of the known - the colossal black hole (a real singularity) at the center of our galaxy, gamma-ray bursts shining from the dawn of the universe, neutron stars spinning 30 times a second. Talk to an evolutionary biologist and you'll get the wonder of the latest discoveries about life on Earth. Our history and place in the cosmos.
This singularity crap is all about the potential economic benefit to be squeezed from the always as yet unrealized economic exploitation of other people's thrill of discovery. That it is also snake oil - muttering about exponentially increasing returns for the investors doesn't make it so - is a "tell" that the purveyor is likely to skip town the moment the band costumes arrive.
Differential equations work both ways. Any game of chutes and ladders involves downward trends (exponential or not) as well as upward. Our civilization is not marching inevitably toward a shining utopia. The future may be so bright we gotta wear shades - but the glow is likely to be thermonuclear unless we work hard and long to make it otherwise.
Posted by: BrianX
|
August 18, 2010 2:18 PM
mvanbebber:
I return to what I said earlier about cryonics: the promotion of transhumanist concepts based on something that could happen for which we don't have much of a theoretical understanding is, at best, cargo cult thinking -- create a complex enough chip and AI will automagically just appear. If this is true, we've had very large-scale low-cost supercomputer clusters going since the mid-1990s, a good number of which (a decent fraction of the Top500 list, anyway) almost certainly exceed the brain's complexity, and yet are frankly not substantially different from any existing computer system. That's why Moore's Law is a red herring here -- all the technology in the world is no more than a blank slate until we understand how to do the modeling. And we simply don't, so for Kurzweil to make any prediction at all is outright ridiculous.
Posted by: mvanbebber
|
August 18, 2010 2:21 PM
@KG:
First of all, calling Kurzweil, an esteemed scientist and inventor Kurzlooney is a baseless ad-hominem attack and doesn't lend support to your position. That is overstating your case against his predictions. Second, I re-read that passage from Kurzweil's book and was unclear if he meant that everyone would have this stuff. He also says "in the beginning of the decade." It's very telling to me that you don't even know about real-time language translation. Do a google search for god's sake. There are iPhone apps for it now. There are several applications for desktops that do it, for Chinese, Japanese and Spanish. Facial recognition has been around for a decade now, and is used in surveillance cameras to identify criminals. Geesh. Like I said, I agree that virtual assists are terrible, and I overstated my case when I said that they were here. We have them, but they are terrible, as I said.
It's obvious that you don't keep up with current technology and then you attack me for your ignorance. Seriously, check out those things I told you about.
Posted by: mitsu.hadeishi
|
August 18, 2010 2:31 PM
>what the relevance is
I'm just saying that what Kurzweil is talking about is Kolmogorov complexity, and PZ Meyers seems to think he's talking about actually building a gene expression simulator. Kolmogorov complexity is a theoretical, "in principle" measure of algorithmic entropy, and Kurzweil and Sejnowski are saying that the genome provides a rough upper bound on this, for at least somewhat plausible reasons. That doesn't really in itself say anything at all about how difficult it would be in practice to actually build such a simulator.
One major problem is as I noted above: in reality, problems like protein folding are quantum problems. The question, of course, remains: does nature actually take advantage of the quantum aspect of things like protein folding to get compression greater than that which would be possible using classical physics? I don't think the answer to this is at all obvious; if the answer is yes, then of course Kurzweil and Sejnowski are wrong.
There may be other reasons why they're wrong, but all I'm saying is this particular argument has to do with algorithmic entropy, not actually building a real gene expression simulator using a Turing machine.
I personally have no particular stake in this; I tend to think Kurzweil is being way too optimistic about the rate of progress in AI to think we will get to human-level AI in ten, twenty, whatever years. Even if we got there, the idea of "uploading" yourself into an AI to me makes no sense at all; we're embodied creatures, intricately connected to all that entails. You might be able to upload something into an AI but I don't think in any sense you could call that "you"; it would be some other creature patterned, at best, after some of your patterns. It would have a similar relationship to you as a student has to a teacher, or something along those lines: at best.
Posted by: Nerd of Redhead, OM
|
August 18, 2010 3:10 PM
Yes, you are, and you prove that with every post, and the concept that you and RK can't possibly be wrong. But you are. Live and learn. If you can do that.Posted by: theswede
|
August 18, 2010 3:33 PM
Facial recognition has been around for a decade now, and is used in surveillance cameras to identify criminals.
Please link to a few stories of criminals having been caught by this.
Trick question; you can't. This stuff only works in marketing material, not in reality. Facial recognition may have been around a long time, but it still doesn't work.
I work with machine vision, robotics and other systems which make use of the latest developments in various recognition algorithms, trying to apply them to real world tasks. And these systems all have one thing in common. They suck at doing the most basic stuff humans find trivial to the point that it's almost impossible to make them do it reliably under extremely well controlled circumstances. They need meticulous training for the simplest of tasks, and insert a few oddballs and the whole system falls apart.
But I don't even have to say this. The fact that you throw around terms and say "google it" instead of providing direct links to working, real world systems says it all.
Posted by: theswede
|
August 18, 2010 3:37 PM
Kolmogorov complexity is a theoretical, "in principle" measure of algorithmic entropy, and Kurzweil and Sejnowski are saying that the genome provides a rough upper bound on this, for at least somewhat plausible reasons.
No, the reasons are not somewhat plausible. You yourself have admitted as much by requiring the addition of physical algorithms, and anyone expecting that the genome + classical physics is even remotely enough to build a brain is so ignorant it's painful.
Even if your statement that it's merely about Kolmogorov complexity (the relevance of which is still up in the air) is correct RK is arguing from stupid.
Posted by: Steve LaBonne
|
August 18, 2010 3:38 PM
It's a good thing thing the kurzweilbots aren't representative of computer scientists, or I'd start to think it's a miracle that computers work at all.
Posted by: mvanbebber
|
August 18, 2010 3:40 PM
@BrianX
Is this comment directed toward me? I never said anything about cryonics or Moore's Law. Moore's law is a very specific and narrow observation about chip density, not the first or the last paradigm that has brought exponential growth to computation. About predictions, science is nothing but useful predictions based on past observations. You see a phenomenon that is always true, and you "predict" that it will be true in the future. You "predict" that the laws of aerodynamics will hold true next week, when you board your airplane, and you have extreme confidence in this prediction. What many futurists do is to take into account the laws of the universe (of which, one is exponential growth in information technology of all types) and extrapolate future capabilities, given those tools. What would actually be astounding, as Hans Moravec has said, is if the singularity DIDN'T happen, because it would mean the ceasing of these laws within the next 30 years - laws that had been constant for the past 13.7 billion years.
We can't know if the brain can be simulated in 20 years, as Kurzweil has said. But, given that our tools will be 1,000,000 times greater and faster and 100,000 times smaller, I think it is a reasonable prediction. That has assumptions built into it, of course: that the brain is a type of computer (which even PZ admits), that the brain is finite, that the brain contains no supernatural stuff, and that the salient interactions of the brain are found in the neurons and dendritic connections between them. I am not 100% convinced we will do it by then, but it certainly doesn't seem unreasonable, as many nay-sayers have implied here.
Posted by: theswede
|
August 18, 2010 3:54 PM
That has assumptions built into it, of course:
Oh yes. The first, biggest assumption is that in 20 years we will know how the brain works well enough to simulate it. And that is the big one you're not addressing anywhere, or providing a single plausible argument for.
1,000,000,000,000,000,000,000,000,000,000 times more computing power is entirely immaterial if we don't know what to evaluate the results against.
Ironically, this is also precisely what PZ says in the OP. That RK makes an argument purely on the basis of increased computing power, completely ignoring that this is irrelevant until we know what to run on it.
Posted by: Paul
|
August 18, 2010 4:20 PM
Anyone know which Kurzweil-bot used their time machine (which of course we will have, because our tools will be a million times greater and faster in 20 years and don't ask me how the theory works or any specifics, because SCIENCE!) to go back in time and interrogate Mr. Babbage?
Posted by: BrianX
|
August 18, 2010 4:25 PM
it certainly doesn't seem unreasonable
No, maybe it isn't. But by the same token, that doesn't mean there's enough evidence to take speculation on that front seriously.
Posted by: Flex
|
August 18, 2010 4:26 PM
mvanbebber @446 wrote,
Really. The exponential growth in information technology is a law of the universe?
What they won't think of next.
Not that you have defined 'information technology' . I mean, are you talking libraries or interstellar probes, search engines or cyborgs?
For those of us who are little more attuned to reality, we might postulate that the growth of information technology (however it's defined, so long as the definition doesn't change) likely follows a sigmoid curve, like many other systems.
We may be have just started at the point of exponential growth, or we may be nearing the end of the period of exponential growth for information technology. However, it is highly unlikely that exponential growth of any system is unlimited, let alone a law of the universe.
Posted by: mvanbebber
|
August 18, 2010 4:30 PM
@theswede
Hahaha, well that's a bit rediculous. That much computing power could simulate our entire brain and then some, atom by atom. It could simulate millions of brains atom by atom. So, if nothing else we could conceivably scan an active brain and instantiate it atom by atom with that much power. The only thing I'm claiming is that it is conceivable, not guaranteed. I don't even think this is the most likely method, but a proof of concept.
About the facial recognition - since you work in that field you should know about it's biometric uses in the department of homeland security, checking visas and passports? You know way more than I do about this, presumably. Am I wrong, has this not been used to check visas and passports by the US government? Has facial recognition not been by the NY DMV to catch identity thieves?
http://www.washingtontimes.com/news/2010/aug/9/facial-recognition-technology-identity-thieves/
About the RK argument - he most certainly does not make the argument "purely on the basis of increased computing power," sigh... He makes the case so strongly and so many times that computing power is "only an enabling factor, not the entire solution." It's obvious that most (all?) of these nay-sayers have no clue what Kurzweil's position actually is, even though he makes it abundantly clear in books and interviews.
Posted by: Moggie
|
August 18, 2010 4:45 PM
#468:
Working in IT, I've always loved this quote. Exasperation with dumb lusers dates back to the very dawn of our industry!
Posted by: Kagehi
|
August 18, 2010 4:51 PM
(i.e., if you restrict prenatal limb movement it alters brain development; the development of ocular columns relies on prenatal spontaneous stimulation of the retinal cells; etc)
Yes, but only to a point. Someone without a limb, or who loses one, doesn't lose the structures in the brain that would have mapped to that, they simply remap to a nearby process. However, whether or not they do so completely, or only partially, due to other, preexisting, wiring, depends on *when* you lose that limb. There does seem to be "some" level of modularity, which doesn't change, even if it can, lacking proper stimulus, get "borrowed" for some other purpose, without necessarily changing the fact that it is still part of a different module. Seems to me, this is damn near a necessity. Without some fairly clear delineation, built into the "code", with respect to what area of the brain will handle something, and how much of it, there would be no functional way to prevent the whole thing from wiring randomly, or producing entirely inadequate modules to handle the result. Now, it is *possible* that these things fire semi-randomly, then watch for responses, and wire themselves based on where, and how much, return they get from those sources. But, again, you still need some baseline, which says, "You need to treat this thing as a separate part, and wire appropriately."
That said, one thing that stands out is his assertion that you could remove redudancies... Uh, first off, this is only true if you know that every cell expresses the *same* pattern of genes, and second, its not even true **at all** with computer algorithms. Some level of "seeming" redundancy exists in software code too. One could argue, for example, that if some gene worked like a for next loop, there is no guarantee that the variables it acts on will be the same, so could be classed as "redundant". Worse, we are finding, as we work with multiple core processors, that, in some cases, what seems like redundant, or pointless, code, can be *critical* to prevent the code in question from causing the processors to go into fits, switching tasks, or accessing memory. There is finite room in a cell to process data, even if it does it in large batches, and massively parallel, so *some* redundancies may be necessary, in that context, just to process the "code", without causing a log jam, while trying to run different chunks of it. And, unlike a processor, you don't just turn off one section, offload it to virtual memory, and come back to where you stopped. Like a cell, if you are doing a lot of multi-core processing, the ***last*** thing you want to be doing is offloading 50% of your processes into virtual memory, eating up the existing work space with a mess of stuff that isn't doing anything, etc. The equivalent of trying to run every scrap of code in a cell, all at once, then having to, somehow, "halt" 90% of those processes, then restart them, then halt them, etc., to let them all actually do anything.
Redundancy can be **mandatory** in some cases. One could, maybe, argue that some redundancies "could" be reduced in a computer simulation of the result, but it would only give rise to the necessity of other, different, redundancies, to compensate for other problems. Mind, I tend to suspect that the whole one-neuron = one-processor thinking people have had is going to prove to be unnecessary, and overkill, if for no other reason than that you can run a cluster of simulated neurons, of some size, at the same speed the brain would, just on one processor, making the issues a) how many cores you need to deal with each "group" of those, and b) how to get one simulated cluster of neurons to talk properly to others, in a stable way, which is not trivial, but solvable. One could even, if we understood it better, start with each core having X neurons, and ending with X-Y, after paring down to those needed to handle what ever frame/design the AI needed to handle its "limbs", and other inputs/outputs.
The problem is, simply, that we don't have a damn clue, at this point, how the brain maintains such plasticity, allowing you to either a) pare down the neurons, when not needed, or b) remap them to deal with new inputs, like someone getting a hearing implant, where the "data" coming in to those nerves was never present, but where they, never the less, can still process it. Some of that may be tied into the critical modules already being present, and having had other, less direct, inputs, as seems to be the case with people developing language, even when having no hearing. But, we can't even build, at this point, a neural net that a) won't stop working if you remove a seemingly unused part, b) produces the intended result, instead of learning something unpredictable, or c) doesn't depend on cross talk that is not even part of the wiring (like electrical interference from one of those unused parts, which, when removed, causes the whole thing to collapse). Worse, we can't, last I heard, do the equivalent of installing a Cocular implant, and have the network not, again, die, do to unexpected inputs, that don't match what it normally received, and/or connect to networks together, without both failing.
No idea if purely simulated ones, which lack at least one of these issues, in that they "must" rely only on their own wiring, and not random weirdness, are any better yet.
Posted by: Paul
|
August 18, 2010 4:58 PM
The thread is over guys, we can go home.
Also, can we now stop all study of subatomic particles? Eventually our computers will be fast enough to simulate them, bit by bit, so we shouldn't waste billions of dollars trying to actually understand them before the computers can just spit out the answers telling us.
Posted by: mvanbebber
|
August 18, 2010 5:03 PM
@Flex
What they won't think of next.
Yes, as I said, it APPEARS to be a law of the universe. Here is a smattering of the data that supports this conclusion:
http://www.singularity.com/charts/page17.html
Going to shut up now, too many posts. too... many... straw-men... getting... tired...
: )
Posted by: Paul
|
August 18, 2010 5:10 PM
Oh, look, a chart that cherry-picks important events to fit a log-log graph. That sure sounds like a law of the universe. Especially when none of the events were required or inevitable in any way. Just-so stories make great bases for universal laws. Seriously, it's the Strong Anthropic Principle for non-theist geeks.
Posted by: KG
|
August 18, 2010 5:17 PM
- mvanbebberI've read the idiot's whole fuckwitted book of prophecies. That's why I call him Kurzlooney. That you could take that ludicrous chart seriously says everything we need to know about you.
Crap. It's labelled 2010, and he repeatedly says "we will have" and "always available", including about things that already existed. You just can't face the fact that your Holy Book and its Prophet are garbage.
Of course I know there are things that claim to accomplish real-time translation, idiot - FFS, google has one. Oddly enough, though, there's still plenty of work for translators. That's because these programs are crap; and that's because in order to do proper translation, you have to understand what it is you're translating.
Posted by: CJO
|
August 18, 2010 5:39 PM
Oddly enough, though, there's still plenty of work for translators. That's because these programs are crap; and that's because in order to do proper translation, you have to understand what it is you're translating.
Anyone who thinks a any kind of look-up function, no matter the size of its database could do the work of serious translation doesn't undersstand language. Translation between natural languages is hard even for humans. It's necessarily fraught with all kinds of ambiguities and value judgements. In any beyond the simplest of utterances, there just is no optimal translation; there will be trade-offs, depending on what aspects of the text or utterance the translator decides to emphasize. Actually, it occurs to me that an algorithm that demonstrated the ability to do a human-level translation of a literary text, like a poem, would have pretty much passed the Turing test.
Posted by: theswede
|
August 18, 2010 5:50 PM
So, if nothing else we could conceivably scan an active brain and instantiate it atom by atom with that much power.
No. No amount of computing power in the world will let us scan an active brain. For that we need scanners, not computing power.
Once again a brutal miscomprehension of the problems involved. Sadly no less than I've come to expect. It must be tiring to build straw men at this rate.
Has facial recognition not been by the NY DMV to catch identity thieves?
Since you can't find any links, it would appear you have no support to claim it has. Thus the answer would seem to be "no".
I don't work with using machine vision to see human faces, to clarify. Only con men do that at this stage of the technology. I use machine vision to find mistakes in assembly in industrial settings, a task magnitudes easier than picking out a face accurately. And even that, for a human trivial, task take man weeks of dedicated effort per simple task to implement, with almost no generalization possible between problems.
For some information on what is possible with a state of the art system which costs about as much as a small automobile, look here:
http://www.cognex.com/ProductsServices/VisionSystems/Video.aspx?id=7847
Going to shut up now, too many posts. too... many... straw-men... getting... tired...
Yes, I can see how building that many straw men, and neglecting to support so many arguments, is tiring.
Posted by: BrianX
|
August 18, 2010 6:19 PM
Hahaha, well that's a bit rediculous. That much computing power could simulate our entire brain and then some, atom by atom. It could simulate millions of brains atom by atom. So, if nothing else we could conceivably scan an active brain and instantiate it atom by atom with that much power. The only thing I'm claiming is that it is conceivable, not guaranteed. I don't even think this is the most likely method, but a proof of concept.
Point goes flyyyyyyyyying over your head... we don't even know enough to emulate the brain of a fish, never mind any species at all of terrestrial chordate. Like I said earlier -- you could theoretically run the simulation on a simple, single-tasking Turing machine, slowly. But not if you don't know what code to write.
Posted by: dgerard
|
August 18, 2010 6:55 PM
For those enjoying discussing cryonics with Luke, you now have some idea what the RW article went through. He's not the only example that led to "reversal of the burden of proof" being listed there, but he was by far the most strident.
I've just nominated it for front cover featuring. Reality-based individuals are also welcomed to come by and kick its tyres.
Posted by: RMS (Not Stallman)
|
August 18, 2010 7:02 PM
@25: Well, if we're only talking the 22 essential aminos, you'd only need 4 bits. Once you get into repeated sequences of these, something like Huffman coding or Lempel-Ziv-Welch compression can help reduce the actual size by writing the sequences to a dictionary based on pattern frequency/entropy. You still have to allow for a bit of overhead in encoding the dictionary, but sizewise, you could get down to half. Again, depends completely on how often amino-acid sequences repeat. Of course, you're still boned, because as anyone who's used a .zip file knows, you still have to expand the LZW compression to get to anything useful.
That said, we're just talking data compression. I think PZ's point, to make a computer analogy, is that simply knowing the sequence doesn't explain its function without operationally testing out into what pathway it fits.
That said, even if Kurzweil waxes transhumanist because of advances made in Kohonen/self-organizing maps or generative topographical maps, it doesn't really mean squat as we still don't understand the mechanisms behind vector distance, data-point selection weighting, or the learning algorithm in general. Without working out the metabolic and regulatory pathways for each protein encoded for brain function, from the bottom up, there's no real way of figuring this out... and that takes time. And this is just addressing decision-making and understanding input bias; this still doesn't address the higher issue of sentience and self-awareness. For Kurzweil to think we'd figure out the innermost workings of cognitive science and memetics within this century is premature at best, wishful thinking in the most likely case, and immature at worst.
Posted by: BrianX
|
August 18, 2010 7:15 PM
dgerard:
You know, I sort of wish I was involved in that page... I didn't realize Luke had been there to crap all over it. Not to mention "Justin Loew" with "I was unfamiliar with SPOV... I was misled by the "rational" term of the Rational Wiki." ..the hell does that even mean?
Posted by: dgerard
|
August 18, 2010 7:38 PM
@BrianX: read the talk page archive, it's great. Ben Best showed up too.
I was misled by the "rational" term of the Rational Wiki" is a standard criticism of RW, and means "you don't agree with me so you must be stupid."
Posted by: RMS (Not Stallman)
|
August 18, 2010 8:00 PM
@483: I should learn to read what I type. The analogy was, I can look at a given file, check the magic number, and have a vague idea of its functional usage. Kurzweil seems to think this is the same case one can look at a functional part of the genome in the same way, without taking it into the context of a greater sequence of events... and you can't.
Posted by: Flex
|
August 18, 2010 8:06 PM
mvanbebber @475, wrote
Ah. I do remember this crazy chart.
According to it, your definition of 'information technology' includes everything ever alive. Why not expand it to include the non-living? After all, there is 'information technology' (similar to what is used by life) used in the formation of crystals. There is 'information technology' used for the formation of stars.
I wouldn't call this the anthropic principle fallacy, but the evolutionary progress fallacy. There is no reason to think that evolution is directed in any way toward the development of intelligence, in mankind or any other species. As an example, if it wasn't for the unpredictable event of an asteroid impact dinosaurs may well have continued to rule the earth for the past 65 million years like they had for the previous 170 million years.
Or would dinosaurs have developed stone tools in the past 1.8 million years like humans did in order to make this absurd chart work out?
I can see it now.
Utah Raptor, "Well, it's been a good 235 million years, but it's time to try knapping some flint so that in another 1.8 million years we can have personal computers."
That chart was clearly created to support a predetermined conclusion.
Posted by: John Morales
|
August 18, 2010 8:29 PM
Flex,
Yes, it's a version of the Texas sharpshooter fallacy.
Posted by: jonathan.shore
|
August 18, 2010 9:00 PM
Good to see a rational view on this. 10 yrs vastly underestimates the knowledge gap necessary to simulate our biological processes.
I would be astonished if it is accomplished in the next 100 years. We are just scratching the surface in our knowledge of biological systems. I think we're at the point where we may "know what we don't know", an accomplishment.
The complexity in biological systems is mind-boggling. Biological systems are also very difficult to observe. So unlike other disciplines, the turnaround between experiments is very long. We need much better observation tools.
That said, I think the future for both AI and "immortality" is bound in translating from the biological to silicon substrate. The body is an incredible machine, but it is far from perfect. It is doubtful that we can extend the meat body indefinitely. Controlling / designing our own substrate feels more manageable.
Posted by: Greylander
|
August 18, 2010 9:43 PM
@ #403, John Morales,
LoL. Too much time on my hands? Not enough sleep last nite, that's for sure. Thanks for the good questions and the lack of ad hominem.
I think about 20 years. I would be very surprised by less than 10 or more than 50. For what it is worth I'd say 10 a bit more likely than 50. Of course it depends on exactly what you want the simulation to do. Since the article talks about reverse-engineer the cerebral cortex, I'm limiting to that "size" of problem. For "whole brain", I'm more comfortable with 30 years. I'm also not talking about simulating a direct "copy" of someone's existing living brain. Nor am I talking about having the simulate go from birth up to adult development. All of these push back the date around another 2-10 decades... quite a bit more for an "exact" (how exact?) copy... depends on exactly how "perfect" you want the simulation to be. Simulating "just the cerebral cortex" while interesting in and of itself, does pose the question of what to give it as "input"... since most of that input comes from trillions of connections with neurons in other parts of the brain. How would we know what to feed in, unless we already are simulating all those other sections of the brain? So "whole brain" is actually more interesting to me. I don't see much point in simulating the cortex processing random or absent signals. Of course if we can do the cortex, we can do any other large section of the brain, we just may not have the hardware to do it all "real time" for another decade or so.
Anyway, as to "destructive". I do not see this as an issue. Most of what we need to know will be in understanding the rules -- the partial differential equations -- governing the [somewhat abstracted, relevant aspects of] behavior of individual neurons and governing the rates of various electro-chemical reactions, etc.
We do need "initial conditions" for the simulation. For the 10-20 year mark, I do certainly do not see us doing anything like copying he detailed state of a living human brain. But I expect that new higher resolution variants of MRI, fMRI plus extremely high resolution scans of dead brain, in conjunction with our knowledge of the detailed rules of behavior for neurons, synapses, etc., will let us fill in "good enough" initial conditions to let the simulated brain begin learning on its own. In this case I would expect such a simulation to be vaguely like someone with extreme amnesia re-learning everything, and a bit like a brain of a body that was in a coma from birth and then had to start learning everything. Of course we would probably want to make the initial conditions as close to a young infant or prenatal brain as possible, so that it could develop and learn in as "normal" a way possible.
10-20 years. No special architecture reuired, just cpu speed and memory capacity. The squishiness is in the data representation of the brain state.
Heat dissipation requirements? Why would there be any special such requirements?
Your questions seems to hint that you imagine some very special design of computer. Nowadays "supercomputers" are just lots and lots of fast regular computers networked together and performing massively parallel operations. The simulation of a brain is all about number crunching of massive amounts of data largely in parallel.
Already exists a thousand times over. There is nothing particularly special here. The simulation would largely amount to numerical solution of big huge partial differential equations operating on arrays of quadrillions of variables.
The "lines of code" metric is a bit silly and at best a rough guestimate of how much programmer-effort will be required. If you think of the program as being the "differential equation solver"... then the code will be tiny and the differential equations themselves can be thought of as data along with the quadrillions of state ariables. If you think of the equations as part of the executable code... the the "million lines of code" esitimate seems a decent ballpark.
Perhaps you mean when will we know what equations to solve? This is where those who object will say "gee we barely know anything about how neurons work, there is no way we'll know all the equations relevant to information processing withing the brain, any time soon." (and presumably when they are thinking 'soon' they are thinking centuries or millenia -- far far from the decades proposed by 'singularity' optimist).
This is where some people "get it" and some don't. The incredible accelerating pace of computation and automation technologies is nothing short of mind boggling. There is no "s-curve" barrier in sight. Computation devices are built up from intrinsically simple parts (bascially you just need types of switching elements emulating logical "NOT" and "AND") short of the atomic scale there is no practical limit to how small/fast/efficient these components can be made. And once the atomic scale is reach we just make more and more with increasingly efficient automated manufacture.
In the near terms we have "lab-on-a-chip" technology in its infancy, but it can ramp up quickly. Think of a single computer chip which can perform thousands of distinct, carefully controlled chemical reactions in tiny micro-chambers utilizing extremely tiny amounts of chemicals, or isolating thousands of individual living cells, each in it's own controlled environment, able to subject each cell to a unique combination of chemical signals. All automated. All doing the work of a macro-scale robotic or human lab, but doing hundred or thousands of times more experiments in the same amount of time. Then imagine massed production of vast arrays of hundreds of thousands of these labs-on-a-chip. Where today we barely know the functions of a relatively small number of proteins, etc... soon we will have massive databases cataloging millions, billions of distinct proteins and with details of how they interact with each other and with other substances.
Lab-on-a-chip is just one of many technologies starting to come online. There are also extremely high resolution imaging techniques to observe the detailed structure and behavior of indivdual molecules.
Not sure what you mean here. The simulator would be a conventional program running on a convention [but "super"] computer. In my first post I make my estimate of the number of FloPS required for the brain as a whole or the cerebral cortex alone. I come up with an estimate that the brain performs the computations equivalent of about 10^21 FloPS. It is interesting to not that this means each of your 10^11 neurons, including all it's synapses, is doing the equivalent of 10^10 FloPS or 10 GigaFloPS.
In other words, my estimate would suggest that a single neuron, with its axon, all its dendrites and synapses, is doing the equivalent number-crunching of a circa 2003 $1000 desktop computer (assuming I haven't jiggled an explonent incorrectly somewhere).
There are good reasons to suppose that a neuron is doing more than this by a few factors of ten, in which case my projected date gets pushed back a couple decades.
--
Posted by: BrianX
|
August 18, 2010 9:51 PM
dgerard:
Ah. The Very Serious People problem.
Posted by: John Morales
|
August 18, 2010 10:26 PM
Greylander, I'm pleased you're giving your proposition some thought.
To respond to some of your responses:
Computational substrate density and 2LoT — (remember, you're not using (analog) meat, but doped silicon that digitally simulates it processes) and you need a fair bit of it to handle your proposal.
We don't yet have computronium.
Really. We have the required algorithms already?
No (say) recursion issues?
Primarily latency issues, but also simulation speed. In an actual brain, signal propagation cascade issues seem not irrelevant. Neural net hardware, too, requires multiple synchronised inputs which in turn generate outputs which are fed back into the system. If some inputs lag, the whole thing would grind to a halt.
(cf. I Am a Strange Loop by Hofstadter for some speculative issues regarding the nature of consciousness.)
In short, it may well be that consciousness (and that's what you're really asking for, not just intelligence) as we know it, requires near-real-time internal feedback.
Why FLOPS rather than OPS? I rather doubt brains use floating point, and of course integers can map floats (cf. the original FRACTINT).
State-changes-per-second would be a more appropriate metric, I suggest.
A single neuron does nothing much outside its environment, and is much more akin to a 1950's analog computer than to any digital system.
Analog computers are limited by dynamic range and (in the case of electronic ones) noise levels, but have other advantages.
Posted by: Greylander
|
August 18, 2010 10:53 PM
@ #414, KG
It's all good. I'm not bothered one way or another.
No. I am not. I never made (and would not have if I had been thinking about it at the time) such a prediction about the genome project. It is as silly on the face of it as the old AI assumptions of neurons as single-bit processors -- an assumption that was fairly obviously silly even at the time it was being made.
Exactly. A big part of this was in the ability to automate lab processes, not just faster computers to analyze the data. Lab automation is getting smaller/faster/more efficient all the time.
I strongly suspect that any optimistic hype about curing cancer was for the sake of politicians and beauracrats. In any event, it is quite obvious that just knowing the sequence of base pairs in the genome is not going to get you very far.
We begin to disagree here, so long as we can get a handle on what we mean by "brute computational power" and what exactly we do with that power. Suppose we have enough computational power to run detailed simulations of the interactions of tens of thousands of molecules and individual atoms. We already have the equations. Then guess what... it become easy, for example, to get the shape of each protein coded for in the genome. We can observe how they interact with each other and with other molecules in simulations thereby vastly increasing our knowledge of the network of metabolic pathways in the human body (or any other organism). In other words, sufficient "brute computational power" will get us very very far. We can use such simulations to "observe" in detail how non-protein-coding areas of DNA influence the expression of the coding regions.
We already know the equations of physics which govern the behavior of atoms and molecules. With that knowledge and appropriate simulation we'll be able to "bootstrap" up to getting the "higher level" equations governing the behavior of things like cell-membranes, ion channels, neurotransmitters, and so forth. Once we have those equations, it is just a question of how many components we can simulate in a reasonable amount of time for reasonable cost.
Brute computing power is going to get us very, very far... but it is possible to both over and under estimate how much we need to achieve any particular result.
Excellent! You have finally made a solid point that can be addressed squarely. Here we have the crux of your objections.
You are unable to see how "brute computational power" can speed up this plodding process of science. (By the way. Yes, I do know how science is done... why you feel the need to make these snide ad hominem attacks is beyond me.) I've touched on this in other posts in this thread but I'll tackle it once more.
Basically what need are: (1) a list of all the relevant parameters -- concentrations of which chemicals, parameters describing the surface characteristics of synapses, what are all the types of synapses, and so on. (2) a set of differential equations relating all these variables. Right now, with the version of science you describe, we can only pick at these things individually, making educated guesses, running tests, and so forth, we can only pick at the individual bits of the problem. Just discovering a new type of synapse or a new variant of an existing neurotransmitter is a big deal.
There is more to speeding up science than just computers doing number crunching (though it helps). There is automation of lab-processes and data-gathering in general. Do a search on "lab-on-a-chip" to see what I have said about that in other posts... then google it. As the ability comes online to run automated trials, tweaking dozens or hundreds of variables and getting results in hours or days on thousands of experiemental trials, instead taking months to get results on a handful of trials, we will begin accumulated a veritable landslide of data. What will be interesting at that point is fitting equations to the data, not publishing papers on how "inhibiting expression of gene A means less of neurotransmitter B and this is correlated with a greater chance of epileptic seizure".
PZ mentions his expertize in developmental neurobiology and his awareness of the immense complexities of the problem of reverse engineering the brain. I do not dispute this.
What you & PZ and other singularity-objectors fail to comprehend is how what you dismissively refer to as "brute force computation", along with corresponding automation and mass-deployment of physical data-gathering processes will fill in all those gaps of what we "don't know" far, far faster than you imagine.
Posted by: Greylander
|
August 18, 2010 11:09 PM
PZ,
I just got to read your addendum to the OP.
First of all, Kurzweil is using the term 'design' very informally.
Second, Kurzweil is using an estimate of the information content of the genome to then estimate the size of a computer program required to run a simulation of the brain. This is a a perfectly legitimate thing to do. It has nothing whatever to do with supposing that we can look at the genome and derive the structure and function of the brain with the kind of computing power with the computing power available within the next 10 years. To do that would require a detailed simulation of the entire human body, at the molecular level starting from a single egg. The computing power to run such a simulation in a reasonable time frame would be many, many orders of magnitude larger than the proposed computing power to simulate the information processing functionality of the cerebral cortex.
It is nowhere said (that I have seen) that looking at the genome is how we will "reverse engineer" (I would call that "forward engineering") the brain. The actual job of "reverse engineering" will be done by looking at actual brains and all their little bits.
I should also point out that the size of the program that will simulate a brain "million lines of code" is quite naturally going to be dwarfed by the amount of data which represents the state of the brain and is processed by that program to simulate the functioning of the brain.
This is why they mention a computer needing something like 3 PetaBytes of memory capacity. The size of the "50 million bytes" of executable code is insignificant by comparison.
I see that a great many of your commenters have followed you in failing to make these distinctions.
Posted by: John Morales
|
August 18, 2010 11:24 PM
Greylander:
Yes, you've made it clear what you "see".
You think PZ fails; I think you fail to see PZ doesn't fail, because you don't get what he's saying.
So, from "The human genome has three billion base pairs or six billion bits, which is about 800 million bytes before compression, he says. Eliminating redundancies and applying loss-less compression, that information can be compressed into about 50 million bytes, according to Kurzweil." you get that the executable code is 50MB?
Uh-huh.
--
PS You do realise the magnitude of the source for any given program in machine code ≠ assembly ≠ higher-level languages.
NB: machine code is dependent on the specific architecture.
Posted by: PZ Myers
|
August 18, 2010 11:34 PM
It is not legitimate to use the size of the genome as a basis for the information content of the brain unless you are going to construct your brain simulator using an analogous method to the construction of the organic brain.
You're going around and around in circles about your own ignorance.
Posted by: BrianX
|
August 18, 2010 11:43 PM
Greylander, 493:
You're still doing it. You're still making unwarranted extrapolations. Why are so many of you Singularitarians failing to get the point that just because it's possible we may get there does not mean we will, nor that there's any reasonable way to predict when or how we'll find it if it's taken as a given.
Which means Kurzweil is full of shit, as is every other Singularitarian out there.
Posted by: First Approximation, L'esprit de l'escalier
|
August 18, 2010 11:44 PM
PZ (addendum),
Should that be: "there is no problem in principle with replicating it artificially"?
Posted by: John Morales
|
August 18, 2010 11:52 PM
BrianX, no, what Greylander is doing is thinking of the genome as computer code, because that defends Kurzweil. 'Tis sad.
Posted by: BrianX
|
August 19, 2010 12:12 AM
John Morales:
Either way it's sad. Besides, the two are not mutually exclusive...
Posted by: Greylander
|
August 19, 2010 12:54 AM
@ #423, poke,
This is not a sensible objection. I know full well the differences between how the brain work and how a computer works. None of my arguments have anything to do with that analogy.
Your miscomprehension (shared by many) appears to be that you think simulating something on a computer is means that the physical structure and behavior of the computer is somehow similar to the thing simulated. It is easy to simulate a wiggling jiggling bowl of jello on a computer. That does not mean that there is anything about the components of a computer that resemble in any way a bowl of jello.
Hmmm... basically you are saying that every detail of a physical system is relevant to every detail of the behavior of a physical system.
The points is that some details of the behavior of a system of not important (for a given purpose) and that many details of a system will not be important to those aspects of behavior that are important (for a given purpose).
Really? A brain is not the same as computer program? I never would have guessed...
'Complete' in this context can only mean relative to a given purpose. No one is suggesting the simulation would be exact in every conceivable detail. What would that even mean unless we were simulating a specific person?
Again, 'complete', in what sense? Are you suggesting, for example, that we cannot separate the information processing done by neurons from the generation of energy by their mitochondria?
Of course energy supply is important to a functioning human brain, and if the brain starts running out of fuel that is going to impact 'thought', but are such details necessarily important? It depends on the goal of your simulation. If your goal is to examine the neural processes involved in solving math problems, composing a song, or recognizing the face of your grandmother, we may not be particularly interested in how those processes start to get fuzzy and break down as the brain becomes exhausted. Rather we may just want to assume that the neurons have enough energy to function and leave mitochondria (and other "life support" and "maintenance" systems of the neuron) out of the simulation.
Please show me anything I said that contradicts this.
Who said it was? You are getting way off track here.
(underline mine)Whether or not you include the word 'not', the above sentence is meaningless. I certainly never said anything like this.
Really? And why not? Where is your evidence? What is your reasoning? You really think that we cannot separate the various signal processing that goes on involving neurotransmitters, excitation, and so forth from the basic maintenance and life support operations of the neurons?
What? So if I run a simulation of an automobile, with engine, aerodynamics and so forth, it is meaningless to say how accurate the simulation is with regards, for example to the actual wind resistance and fuel economy of the real thing? If I simulate the load capacity of a steal bridge, it is meaningless to talk of whether and to what degree the simulation is accurate with respect to how much weight the real bridge can hold before collapsing?
I think I said pretty much the same thing, with a bit more clarity. How you relate this to some of the other nonsense you've written I have no idea.
Er... um... what?
Thank you very much, for trying to save us some work... but we will simulate everything we can in as much detail as our computational resources make feasible, as accurately as our knowledge of the physical processes involved will allow. And we will use those simulations to make predictions and guide further research. We often run simulations based on well-established laws of physics but involving such a large number of components/particles that the only way to determine the so-called 'emergent' behavior is to (1) look at a real-world system; or (2) run the simulation. Often it is much, much easier/cheaper/faster to run the simulation. Many scientists literally do experiments by running simulations these days. They are quite aware that some missing detail or simplifying assumption may skew the results, so they do checks of the simulation against the real world as much as is reasonable to verify that the simulation is accurate in the sense of being like the real thing and the range of parameters over which the simulation remains accurate.
Posted by: Praxiteles
|
August 19, 2010 1:36 AM
Having sat through a half dozen Kurzweil lectures now - and several conferences - Dr Myers - you seem to be attacking a point that I have not seen Kurzweil make.
Kurzweil just gave a lecture on "Building a Brain" at the Singularity Summit this last weekend. In no way shape or form did he suggest reverse engineering the brain from the genome. I have never heard a single "brain builder" suggest such a difficult endeavor. Scientists don't even understand how proteins fold - let alone constructing physical structure from DNA.
Kurzweil may suggest that the "design" is there (and as you point out, may be wrong) - but he is not suggesting this as a basis from which to build the brain.
Kurzweil uses the "data" DNA argument to suggest that the brain structure is highly redundant and fractal in design. He uses it in his lectures to convey how much simpler the design must be than specifying each and every one of the trillions of connections we see.
For those who have a scientific interest in what Kurzweil is proposing - it was said that the lectures from the Singularity Summit in LA last weekend will be posted online.
Posted by: John Morales
|
August 19, 2010 1:56 AM
Praxiteles (my emphasis):
What PZ points out is what PZ points out.
It's odd that you imply PZ addresses a straw-dummy in the very same comment you attack the one you've made.
Posted by: SaintStephen
|
August 19, 2010 2:03 AM
Hey Morales:
Please put a friggin' sock in it. Thank-you.
Jebus H. Christopher you are a snarky snoozeball. If you're that lonely mate, try striking up an argument with Quetzalcoatl on rd.net. You guys are a match made in heaven, if there was one.
Posted by: John Morales
|
August 19, 2010 2:13 AM
[meta]
SaintStephen, your plea is noted and duly ignored.
PS Any response to my question to you @400?
--
PS Quetzalcoatl is, like anyone else, welcome to PZ's sandbox, and subject to the usual rules (again, like anyone else).
Posted by: SaintStephen
|
August 19, 2010 2:24 AM
You know Morales...
If there was a Molly for quantity of posts, you'd have my vote. No question about it.
Best I can do for ya, dude. Over and out. Forever.
Posted by: Andyo
|
August 19, 2010 2:29 AM
Geez SaintStephen, chip on your shoulder?
Posted by: SaintStephen
|
August 19, 2010 2:35 AM
More like gum on my shoe. But thanks for asking, tone troll.
Posted by: John Morales
|
August 19, 2010 2:42 AM
"Over and out. Forever." SaintStephen: I note you don't dispute you misread the OP and opined out of ignorance.
Qui tacet consentire videtur.
Posted by: Andyo
|
August 19, 2010 2:45 AM
"tone troll"? It doesn't mean what you think it means.
Posted by: Greylander
|
August 19, 2010 3:09 AM
@ #496, PZ Myers,
[emphasis mine]Wrong.
I don't have to because the machinery of the fertilized egg does it for me.
Let's break it down...
No one that I know of suggests that "Information content of the brain" equals [or less than] "Information content of the genome". That is absurd. It is not what Kursweil says or implies. It is not what I have said or implied.
What is suggested is more precisely something like this: "an efficient binary coding of the necessary differential equations to describe the dynamical processes in the brain relevant to cognition is in the neighborhood [order of magnitude-ish] of 10^8 bytes".
What is also suggested is this: "The information content of the brain (the values of the all the variables on which the aforementioned partial differential equations operate), relevant to cognition, is at least [order of magnitude-ish] equivalent to about 10^16 bytes [Kurzweil, optimistic], 10^19 bytes [Me, conservative] bytes of binary data and probably not too much higher than that."1 (I would be very surprised if it turns out to be more than 10^22 bytes.
What is also suggested is this: "The information content of the brain, relevant to its cognitive state at any given time, is much, much less than the information required to describe the detailed physical state of the brain at any given time." For example it would take something like 10^29 bytes just to describe the position, orientation, linear & angular velocities of all the atoms in the brain... and that does not even scratch the electrons and subatomic particles... and if you want to describe the state of the underlying quantum fields then for all practical purposes you have infinite information.
It is safe to say that: "Total [relevant to cognition] information content of the brain" is less than or equal to "Information in genome" plus "Information in cellular machinery of fertilized egg" plus "information received as input from environment" pluse"Laws of physics".
Since all brains are structurally/functionally similar pretty much regardless of environment, it is safe to say we can leave environment out of a functional description of the brain (functional description mainly refers to those differential equations mentioned above). Evironmental influence will be encoded in the physically encoded cognitive state of the brain (the aforementioned minimum 10^16 bytes representing cognitive state of the brain).
Much of the information in the cellular machinery is already in the genome, such as the sequence of amino acids in all the proteins. The shape of proteins depends on sequence plus laws-of-physics and perhaps a bit of chance wiggling and jiggling.. so there is no new info in protein shape... which largely determines function. The particular arrangement of all the components in the cytoplasm is information not found in the genome, but is it relevant? We can jiggle and wiggle the cytoplasm quite a bit, and all those molecules will still wiggle and jiggle functionally (in fact the normal functioning of cellular machinery looks like a lot of random wiggling and jiggling, 3 steps forward/2 steps back, with a general trend toward functional results) -- the cell survives, divides and grows up into a human being, so information describing the detailed arrangement of molecules in the cytoplasm of fertilized egg are not for the most part needed to construct the human being that follows.
There is not much information in the laws of physics per se. A few kilobytes is sufficient to encode the equations.
Ergo, all or at least the vast majority of information required for a functional description of the human brain (aforementioned partial differential equations) is contained in the genome.
We can draw this conclusion without knowing anything more about how to interpret the genome to construct a phenome.
The 10^9 bytes of data in the genome can be losslessly compressed to about 10^8 bytes. In other words about 10^8 bytes is an upper bound on the amount of information in the genome.
Ergo, a functional description of the human body (differential equations) can be efficiently encoded in about 10^8 bytes. A functional description (differential equations) of the human brain will require somewhat less.
1For those who want to complain that the brain is a squishy analog thing and so we cannot compare to digital data... sorry but, yes we can make such comparisons. For example, the eye detects colors, which are analog signals impacting the retina and the retinal cells transmit this information to the brain with 'analog' electrical currents, 'analog' time delays between neuron firings, and squishy squirts of neurotransmitters... how can we possibly say anything about how many bits of information are transmitted by a single dot of color? By noting how many distinct colors that analog system can actually distinguish and determining the number of bits required to encode that many distinct colors. Your eye can barely perceive the differenc between RGB and RGB on your computer screen. Typically we encode color with 1 byte per color chanel or 3 byte, 24 bits. It does not take many more bits to have such fine gradations of color that you cannot distinguish them. Ergo, each "dot" of color provides maybe 32 bits of data to the vision system in single "instant". Or each rod or cone provides about 8 bits of data per "instant". How fast -- how many "instants" per second? Well we know it is difficult to consciously distinguish more than about 30 frames per second. Allowing for subconscious processing to be affected by faster rate, lets say 60 fps... high-ball it on up to 100... so we can say that each rod and cone in the retina sucks in about 100 bytes/second. The same kind of reasoning applies to every squishy 'analog' process of every neuron and synapse.
Sigh. You are slamming into brick walls with yours. Thus saving you from the circles.
Posted by: Greylander
|
August 19, 2010 3:45 AM
#495, John Morales
Yes.
Sigh. In for a penny in for a pound. SIWOTI.
OK, here goes.
The "code" is just going to be programmatic encoding of what amounts to a bunch of partial differential equations. That's just mathematical expressions. We're not talking about writing the Windows operating system here. The binary encoding of mathematical expressions is pretty tight. In other words, we can expect the "real" program, whenever/however it is created, will be fairly efficiently encoded when compiled into binary by a decent compiler.
It is the information content of those differential equations which is suggested to be less than equal to information of genome, not the information content of the cognitive state of the brain at a given instant in time.
The genome can be compressed losslessly to about 10^8 bytes. So the information content of the genome is about 10^8 bytes or less.
There cannot be any more information in the above mentioned differential equations than in the genome.
So if coded efficiently those equations will require about 10^8 bytes or less.
About the "million lines of code". Write down fairly long one-liner mathematical expression in Java, C, or whatever. You will find it typically expands to a few dozens of bytes of actual binary code. So lines of human-readable code will be somewhere in the neighborhood of factor of ten to factor of 100 less than bytes of binary code.
This gives us something like 10^6 to 10^7 lines of code for the whole body... of which the brain is a significant fraction.
It bears reminding: this "million lines of code" will be a program process arrays of quadrillions of variables which represent the actual cognitive state of the brain, along with a comparably sized stream of input data.
Posted by: https://me.yahoo.com/a/r_Xfs6k0qpFJE7_5lI2g.C8S3S7k0a4KFJ16#5454e
|
August 19, 2010 3:55 AM
I agree with what you're saying, but I have to ask, is it necessary to insult anyone besides Kurzweil?
"To simplify it so a computer science guy can get it..."
That's an attack on an awful lot of people and what purpose does it serve?
Posted by: John Morales
|
August 19, 2010 3:58 AM
Greylander, what about PZ's point regarding epigenetics? Is that irrelevant?
--
Sure.
Let's try Conway's Game of Life in one line of APL.
Posted by: Ichthyic
|
August 19, 2010 4:04 AM
conclusion I reached from the comments here:
there are way too many programmers who think they are biologists.
Posted by: dgerard
|
August 19, 2010 4:41 AM
@Icthyic:
"conclusion I reached from the comments here: there are way too many programmers who think they are biologists."
It's more a case of "I don't understand it but I'm expert in something unrelated, so therefore everything else can't be that hard."
Or, in Dilbert terms: "I don't understand it so it must be simple."
And also, "when you point out how insulting and stupid my arrogant assumption of transference of expertise is, you're just being rude so I don't have to listen to you nyah nyah even though this is actually you're field I'm bafflegabbing about."
Posted by: dgerard
|
August 19, 2010 4:44 AM
@516 "you're field"
You used an apostrophe wrongly so therefore my arrogant bloviation about fields I clearly don't understand is valid.
Posted by: Greylander
|
August 19, 2010 4:56 AM
@ #492, John Morales,
Yes. It is called a supercomputer... which nowadays are just grids of more ordinary joe-blow computers. The low end estimate, 10^17 FloPS is just about a factor of 100 away from current fastest supercomputers...i.e. about 3-4 years out. From the article: "By next year, IBM's ‘Sequoia' supercomputer should be able to offer 20 petaflops per second peak performance, and an even more powerful machine will be likely in two to three years." Ask them how they deal with heat dissipation right now. There is nothing intrinsic about computation that requires so much waste heat... rapid advances are being made in energy efficiency of computer... worst case in the near-mid term, space out the grid of cpu's a bit more and buy heavy duty inustrial air conditioning units.
We would if the aliens weren't keeping it from us! :)
You asked about expertise not algorithms. But as to algorithms all that is needed is some good numerical integration of partial different equations. As I've touched on elsewhere, it is figuring out what the differential equations need to be, and the variables they operate on that will be the tough part. See other posts by me. Search phrase "lab-on-a-chip" in this thread.
Again it seems like you are imagining some special hardware for this. I'm talking ordinary Von-Neuman architecture, or rather a networked grid of such... basically a huge grid of ordinary (if very fast individually) PC's. Each PC would focus simulating on a clump of proximate neurons and also transmit receive boundary data and data representing synaptic signals and so forth to nearby PC's which would be simulating nearby clumps of neurons. To some extent it is probably (but not strictly necessary) that there will be significant correlation between the relative physical arrangement of cpu's and the relative locations of the neurons they simulate.
Cool. Love Hofstadter. Godel, Escher, Bach and The Mind's I (also Dennett), were tremendously mind-opening.
The subjective experience of self-awareness (i.e. consciousness) is a whole other conversation.
FloPS is just typical measure of computing power of supercomputers. If the differential equations are implemented as discrete integer difference equations is an implementation decision which has little impact on the rest of the discussion at hand.
State-changes-per-second is around 100... but that does not tell us what we need. The question is how many bytes encode a brain-state (i.e that 10^16 minimum estimate) and how much computation is required to calculate the next state. Each CPU will be calculating the next state for many neurons each of which is describe by many variables. Hence we need some idea of the total number crunching that will be required from all the CPU's in the grid.
You are missing what I mean by "equivalent number crunching". Consider the simple model of neuron used in AI "neural networks" (I expect real neurons much more complicated): they sum up all their inputs and apply a nonlinear s-curve function. So if real neurons were that simple, we might suppose each neuron performs the equivalent of summing N products of "synapse strength" time "input activation" where N is the number of synapses on its dendrites. The nonlinear sigmafunction would require a few more mults and adds to do a polynomial approximation (negligible compared to N). Since neurons may have up to ballpark of 10,000 connections, say N=10,000 and we need to update 100 times per second, so we need about 2 * 10,000 * 100 FloPS or about 2 MegaFloPS per neuron. That's not all because we also have to apply learning rules to adjust the connection strength of each based on some rule given a learning feedback (maybe the concentrations of some "pleasure" and "pain" neurotransmitters in the area). This is going to require several operations per synapse, so now figure 10 or 20 MegafloPS per neuron.
Given, as KG has pointed out for the edification of my presumed ignorance, that there are many types of synapses each obeying different rules, many types of neurotransmitters, and so on, 20 MegaFloPS per neuron of the simple A.I. model is paltry compared to the real picture. I think I estimated earlier something like 10 GigaFloPS per neuron.
Posted by: Greylander
|
August 19, 2010 5:14 AM
@ #499, John Morales,
Absurd. I have never said anything to imply this. There is a huge difference between deducing things about and from the information content of the genome verses regarding the genome as a "program" in the conventional sense.
But if the "computer" you have in mind is the cellular machinery floating about in the cytoplasm... then it is appropriate to think of the genome as a sort of computer program. A very different sort of program, running on a very different sort of machine, to be certain. Many different parts of the program can be "running" simultaneously and rather haphazardly, there is nothing directly analogous to a computer's cpu or memory. Instead of conventional flow control (if/then, loops), you have epigenetic factors applied (at the direction of parts of the 'program') which increase/decrease or entirely block the frequency at which parts of the program are executed. It is all rather fuzzy and non-deterministic.
None of this prevents us talking about the information content of the genome, which is clearly made up of a finite alphabet of discrete symbols (i.e. the genome in and of itself is digital data).
Posted by: Stephen Wells
|
August 19, 2010 5:20 AM
Greylander, if you produce lines like this: "Since all brains are structurally/functionally similar pretty much regardless of environment, it is safe to say we can leave environment out of a functional description of the brain " ...
...then you should expect to be mercilessly mocked anytime you go anywhere near anyone who knows anything about biology. Kurzweil's claim that "the design of the brain is in the genome" is simply wrong on its face. Brains are the result of genetics + development + learning, and the genetics end is the smallest, easiest bit.
Stop worrying about how ultrafast your futureputers are going to be. This is not your problem.
Posted by: Stephen Wells
|
August 19, 2010 5:30 AM
Oh, I see that the topic's already been fully covered:
http://amultiverse.com/2010/08/19/
Posted by: John Morales
|
August 19, 2010 5:31 AM
Greylander:
20 petaflops per second, eh?
Ahem: "The size of the "50 million bytes" of executable code is insignificant by comparison."
Where did this magic figure come from, then, if not what I thought @495?
Yes, I did my own semi-jocular bafflegab @9. And the genome is not just data, it's physical — you know, deoxyribonucleic acid?
Posted by: John Morales
|
August 19, 2010 5:41 AM
Greylander:
You sure you're not getting confused by terminological similarity?
Biological neuron (model) ≠ Artificial neurons.
Because you seem to be conflating the two.
Second, even then, in the latter, transfer functions aren't always sigmoid.
Posted by: KG
|
August 19, 2010 6:00 AM
GreylanderEr, no. Synapses don't "obey rules" at all. This kind of conceptual confusion is yet another indication that you simply don't know what you're talking about.
Posted by: Greylander
|
August 19, 2010 6:06 AM
@ #520, Stephen Wells
...then you should expect to be mercilessly mocked anytime you go anywhere near anyone who knows anything about biology.
Shall we go over it all again? Take one over-simplified statement out of context all you like. How about addressing the substance of my posts, eh?
Kurzweil's statement is only wrong if you insist on interpreting an informal one-liner as a precise and formal technical statement. We all make flippant off-the-cuff statements of that sort which can thus be attacked, even our generous host, PZ. It behooves us to look a bit deeper into the substance of what each other says, and then critique *that*.
You will note in one of my posts that I made this equation: information in brain equals info in genome + info in cytoplasm of fertilized egg + info input from environment + physical laws.
Your own equation is flawed. You leave out the cellular machinery of the fertilized egg. And "Development" is not an input of information to the system... it is the sequence of intermediate states of the system. I will accept your "learning" as refering to input of information from the environment. You also left out physical laws, but I'll give you a pass on that as tacitly assumed.
I'm not worried. My only problem at the moment is having gotten caught up in defending my position on this thread... but it's not really a problem, I am enjoying. Costing me sleep though. LoL.
Posted by: KG
|
August 19, 2010 6:14 AM
GreylanderThis is just so utterly, hopelessly wrong it's hard to believe you're being serious. If you were right, then you could take the zygote (or for that matter, almost any somatic cell) out of its environment, replace its genome with anything whatever bearing the equivalent string of four "characters", and hey presto, a fully functional brain (and body) would develop! The genome does what it does because it makes use of the vast amount of information available in its environment - both what specific molecules and larger structures are present, and that implicit in physical law, which determines how those molecules and larger structures interact with each other and the genome.
Posted by: John Morales
|
August 19, 2010 6:15 AM
Greylander, for your consideration¹: When do you think computational chemistry will make physical chemistry redundant?
Because (I hope you agree) that problem is far, far simpler than the sort of biological simulation you write about, and yet that field has been around for decades.
--
¹ No need to respond. I just want you to give it some thought.
Posted by: Greylander
|
August 19, 2010 6:21 AM
@ #522, John Morales:
Are you referring to the reduncy of the "s" and "per second". Who know's maybe they were referring to the computational accelleration of their supercomputer instead of speed? Blame the journalist who wrote the article.
Where did this magic figure come from, then, if not what I thought @495?
I've explained this pretty clearly in earlier post... maybe you're just trolling me because I'm biting tonite?... er... this morning?... er... whatever 'now' happens to be atm? LoL.
"executable code" refers to code of the proposed simulation, it does not refer to the genome itself as executable code. The 50 million bytes refers to information content of genome... and also to size of code of simulation depending on which sentence it appears in.
I have spelled out quite clearly the reason why information content of genome (measure in bytes) will be quite similar to the size of the executable code portion of an efficiently coded simulation (again to harp: the amount of bytes needed to encode the differential equations describing the time-evolution of all the relevant variables).
Posted by: Greylander
|
August 19, 2010 6:30 AM
@ #524, KG:
Are you being deliberately obtuse? They don't obey rules? Perhaps you think I meant the kind of rules that school children sometimes do and sometimes do not obey?
So synapses are not made up of atoms and molecules, surrounded by fluids of more atoms and molecules... and all of these are not obeying physical laws? Do synapses sometimes turn into pink unicorns? Might as well cancel all funding for neuroscience research because KG thinks we will never be able to find equations describing the dynamics of synapses!
Good freakin' lord. Please try to add something useful to the conversation.
Posted by: albatros183
|
August 19, 2010 6:32 AM
"Ray Kurzweil does not understand the brain"
Ray Kurzweil does not understand things, like computers or AI
Ray Kurzweil does not understand things and should not be talked about because he is to stupid to live
Posted by: Greylander
|
August 19, 2010 6:40 AM
@ #526 KG:
This does not even remotely follow from anything that I have said. And it shows that you are unable to make or deliberately avoiding the particular distinctions of what kinds of information applies to which context in this conversation. There is a good deal of clarification of terminology and supporting argument to the above statement. If you are going to ignore (or fail to comprehend) all that there is no point in continuing the conversation. I'm not going over it again.
Your next paragraph doesn't make any sense, but if you mean what I think you are trying to convey, I have already said essentially the same thing. It does nothing to refute me:
Posted by: Kagato
|
August 19, 2010 6:41 AM
Regardless of any of the arbitrary numbers being bandied about for the simulation of a human brain... why on earth is anyone talking about how big the data is when losslessly compressed? It's utterly irrelevant! You can't do anything with compressed data!
Assuming you actually want to run your simulation, the code and data will not be in compressed form.
Some demo coders wrote an awesome game called .kkrieger a few years back. It's a full 3D shooter that fits in a 96kB executable. It's insanely clever.
But really, you can't derive too much from that number. For starters, that's the program's compressed size. It's self-contained and doesn't install, but it does unpack into memory. The game requires a minimum of 512MB RAM (~5400 times its compressed size). It requires a fast processor and a high-end video card, which does most of the hard work of rendering the graphics. It requires Windows & DirectX 9, which provide a massive amount of extra code in support.
You can't look at it's compressed size alone and determine that it should therefore run on a Commodore 128. Compressed size is useless for determining anything about its function.
Posted by: dgerard
|
August 19, 2010 6:41 AM
Here is a demonstration of the sort of compression the genome does. Here is A COMPRESSED COPY OF THE ENTIRE TEXT OF 'ORIGIN OF SPECIES', allowing for an environment to uncompress the data in:
Darwin, Charles (1859). ''On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life.'' John Murray (London).
Note how much environment is required to successfully express the coding.
That's the sort of thing you've got fifty megabytes of.
Posted by: John Morales
|
August 19, 2010 6:41 AM
albatros183, your post is indistinguishable from that of a troll.
But, in case you're being genuine:
Ray Kurzweil is almost certainly much smarter than I am, and certainly more than a bit smarter. And I'm smarter than the average, if not by much.
That he indulges in wishful thinking is beside the point, just as it it with the religious.
Posted by: albatros183
|
August 19, 2010 6:53 AM
John Morales
not exactly trolling but i am a bit sick of him I am also smarter than average but this is basically irrelevant, he has proved again and again that he has no clue what he is talking about.
wishful thinking is not the issue being consistently wrong is rather like religion, if all you do is fail it does not matter that your smart
fail is fail
Posted by: John Morales
|
August 19, 2010 6:58 AM
albatros183, Kurzweil is anything but a failure.
Posted by: albatros183
|
August 19, 2010 7:05 AM
He is well recognized I do not debate that it still does not mean that he is right
Lots of Honorary doc's does not make you correct, in fact I think if you did the math it would increase your likelihood of being wrong
Posted by: KG
|
August 19, 2010 7:17 AM
GreylanderI know - that's why it's puzzling you could also come out with total crap like:
There isn't any sort of context or qualification that could make the above anything other than total crap.
AFAIK, they all act in accordance with the same physical laws. Which means that the statement that they "follow different rules" can't be applying to physical laws, doesn't it?
Posted by: Stephen Wells
|
August 19, 2010 7:38 AM
Something to consider before deciding that it's just a matter of differential equations. Most systems of coupled partial differential equations don't have analytic solutions. So we have to solve them numerically. This introduces numerical errors. In a chaotic system a numerical error of any size, however small- and there'll always be a limit to how many decimal places you can store numbers to! - will propagate up to an arbitrarily large effect on the output.
And you want to simulate a brain as a system of coupled PDEs on a neuron-by-neuron basis?
Right. Unless consciousness is a divide-by-zero error, this is not going to work.
I'm confident that we will in time have conscious, intelligent computational systems.
They won't be simulations of the physics of the brain. They'll be implementations of the algorithms of the mind.
Posted by: Greylander
|
August 19, 2010 7:39 AM
@ #523 John Morales
Did you just quotemine me? (yeah I know not exactly what the term applies to... buy you sure have isolated the above quote out of context).
I conflated nothing and I clearly make precisely that distinction in my post. In context I clearly am using this as a particular, grossly over-simplified, example to illustrate a point about how to relate what neurons do to FloPS. Applying a sigmoid function to a weighted sum of inputs is certainly something that a squishy single-celled analog computer could conceivably do. I make it perfectly clear in the same post that I think individual real neurons do something far more complicated.
Where did that come from? How do you extrapolate from my isolated toy example that somehow I claimed that all neurons in all A.I. neural nets use sigmoid transfer functions?
Bottom line if we know the requisite differential equations that are relevant to cognitive processes, and we know the value of delta_t required for accurate simulation, then the FloPS of processing performed by a neuron are just (1 / delta_t) times the FloPs to calculate the change of state for a single time-step.
Posted by: PZ Myers
|
August 19, 2010 7:44 AM
Guys: I understand Kolmogorov complexity. I understand compression. I understand decompression.
What you seem to be missing is the whole business of the decompression function. You've got a pile of what looks like random numbers, but you're told it is merely a highly compressed version of a functioning brain, and all you've got to do is unpack it to have a complete brain simulator.
How do you do that?
Explain to me how you take the compressed, primitive, incomplete description of the brain that you think is encoded in the genome and expand it into a useful organ. That's a gigantic stumbling block.
I swear, I get the impression the Kurzweinies think they'll just use gzip on the genome database and presto, brain on a desk.
Posted by: Rorschach
|
August 19, 2010 7:50 AM
Gzip is in fact a powerful decompressor.
Wait, what ?
Posted by: John Morales
|
August 19, 2010 7:52 AM
Greylander, no, I seek not to quotemine.
And, clearly, you see no particular difficulties in the various issues I've raised. Fair enough.
Time will tell.
Posted by: Andyo
|
August 19, 2010 7:58 AM
"Time will tell", though, is pretty convenient for futurologists and prophesiers, who tend to speak in vague terms. "Time" will bring about just pretty much anything resembling anything. How about, "be clear and stop dicking around"? I'm not talking about Greylander's assertions in particular, just these "futurologists" in general.
Posted by: Greylander
|
August 19, 2010 9:07 AM
@ #538, KG
Hey, since I actually provide definitions of terms, context, and reasoning... here is an idea. Why not actually point to what you think the flaw in my reasoning is, instead of pointing to a conclusion and just saying it it is wrong?
Yeah, sure at the level of atoms they all obey the same laws, and then all we need to do is run a simulation with known physical laws to track the motion of 10^27 atoms!
It should be pretty clear that this ain't what anyone is talking about. Looking at it the way you describe, there are not any different types of synapses at all. There is just one 'type' and it is parameterized (in part) by detailing, coordinates, orietation, velocity, angular velocity, atomic number, atomic weight, ion number, etc., of every atom in the synapse. This is just silly.
You are the one who mentioned the different types of synapses in the first place. They are going to be characterized by such things as types/quantities of receptors, types of neurotransmitters used, effect on the 'computation' performed by the neuron (excitatory, inhibitory, modulating, etc.).
A simplified example of one "rule" for a synapse might be something like this. Let X be the number of X-receptors on the surface on the dendrite side of the synapse. Let Y1 be the concentration of neurotransmitters Y1, and let Z be the amount of time since the last time dendrite-side neuron fired, and c1,c2,c3,c4 are all positive constants, c4
dX/dt = - c1* X * ( (c2/(c3+Z) - c4 ) * Y1
In words this "rule" basically says, "if the neuron fired recently then decrease the sensitivity of the synapse to X by a percentage proportional to the amount of Y1, the more recently the neuron fired, the more decrease of sensitivity to X, but if it has been too long, then increase the sensitivity to X."
(If someone quibbles about the fact that I put down a differential equation for an integer variable, I'm gonna reach right through the intertubes and slap you silly. Round, floor, or ceiling the whole right side... happy? There are other ways to deal with -- X is typical large enough we'll be interested in surface density as the variable and can treat it a real number of arbitrary precision)
Of course the above is just a made-up toy example. The real equations will be much more complex, with the right hand side involving all the other variables (quantities of each type of neurotransmitter, etc.).
Now... are you really trying to say that there are no "rules" governing the behavior of neurons, synapses, etc., at this level of detail? That is patently ridiculous on the face of it.
Posted by: Stephen Wells
|
August 19, 2010 10:10 AM
Greylander, did you not read my comment @539, or are you hoping that if you ignore it it'll go away?
Numerical stability is a big issue and throwing more computer at it does not make it go away.
Posted by: Greylander
|
August 19, 2010 11:42 AM
@ #539 Stephen Wells,
You are correct... up to a point. And then you are wrong. There are two ways to deal with the problem, both well known. (1) there are highly stable methods for solving numerical differential equations such that the error tends to stay small and random; (2) if the system you are simulating contains a fair amount of noise, you can inject noise as small random perturbations at each time step -- this will swamp the effect of any numerical errors. The combination of (1) and (2) means that you can run your simulation for very long time and it will remain bounded and 'realistic' in the following meaninful sense: at any given simultion-time t, the state of the simulation with be within ordinary bounds, and for some reasonable time span s, the state of the simulation at times t+s will be quite accurate with respect to taking the state of the system at time t as your boundary condition.
Closer to a synapse-by-synapse basis... but in loose terms, that's the idea.
LoL. You post is entirely worth it just for "consciousness is a divide-by-zero error". I love it!
You just inspired this entirely off-the-cuff two-person sketch. Needs some work, but what the hell...here it is:
"Divide by Zero Error"
GOD: Oh shit.
Satan: What now?
GOD: Runtime error. Divide by zero.
Satan: Really? I tried to warn you about linking your knowledge-base to in-world arboreal objects -- that thing has always been glitchy.
GOD: I don't really need the I told-you-so's.
Satan: So. Conscious? Thinking for themselves.
GOD: Yeah. Or rather they will be when I restart from the last saved state.
Satan: So pull up the previous save, from like ten-minutes before she eats the apple, delete the whole freakin' tree and restar--
GOD: mmmmm.
Satin: Uh... you have been doing incremental saves, you know... like I told you... right?
GOD: Um...
Satin: Jesus Christ!
GOD: I told them not to eat the damn apples. I figured that would hold until I got it patched. Uh -- why do you say that whenever you get pissed off?
Satan: I dunno. Habit. Look, why don't you just reboot? Delete references to the tree from the config file first.
GOD: No way. It takes like six days... last time it seemed like I was staring at that hourglass thingy for eons.
Satan: OK, look, I'm sure we can fix this.
GOD: How?
Satan: Edit the config. Enable PROCREATION and DEATH. Restart the sim and kick them out of paradise.
GOD: but I want them with me in paradise.
Satan: Don't worry, I'm patching in a reaper. It'll copy pristine versions of the ones you can stand to have around you from a pre-death backup and transfer them into a second paradise sim.
GOD: Hey awsome. I think I'll call it 'Heaven'. Hmmm... what about the rest?
Satan: Can I have them?
GOD: Sure.
Satan: Confirm Delete Avatar: Serpent? [yes] [no] >
GOD: What will you do with them?
Satan: I've got this cool nightclub sim that needs populating... music with a beat, glitter, lights, cage dancing, two dance floors, split level -- we call the lower floor "The Pit", a bar tender who knows every mixed drink you can imagine, bottomless pitchers, sex, drugs... your basic endless party.
GOD: Well, there'll be none of that in my place.
Satan: I'm sure they'll be beating down the Gates of Heaven to get in.
GOD: Of course! I'm sure you'll get a few too. So what are you going to call your nightclub sim?
Satan: The Inferno. Oh, by the way, Sim-Earth is going to need a run-time correction when sim-time hits about 4K years. Otherwise the Heaven transfers won't go through. You'll need to use the neuro-stim virtual reality interface.
GOD: Oh? But I thought the reaper just needs to do a file copy?
Satan: No... no... um... they're default-linked to the runtime kernel now. You'll... uh... get a file-share violation that'll crash the sim. And there'll be billions... uh... you don't want to halt it for every death.
GOD: OK. OK. You lost me at "kernel". You know I hate all those techie Linux internals. What about that neuro-stim interface. Never used it before.
Satan: Oh it's pretty cool. You'll feel just like your one of them! I use it all the time in the The Inferno. Once I had these three succub--
GOD:
Satan: --anyway. It's very realistic. I should...
GOD: What?
Satan: Nevermind... I'm a sentimental fool.
[Much time Passes]
Satan: Hey it's time for the intervention. Sim-time is year 4265. Get suited up.
GOD: Do I really need this?
Satan: Yes. Now hurry.
GOD: OK, OK, I'm ready.
Satan: Now remember. Stick to the script or you end up having to restart the sim from scratch.
GOD: Right. Right. We've been over this.
Satan: At the end... if things seem a little weird... just go with it. It's all part of the Heaven's Gate transfer protocol. It's just--
GOD: just... what?
Satan: It might sting a little bit...
The discussion here revolves around what we will be capable of vis-a-vis simulating the brain. Whether that will be the best/fastest way to achieve 'conscious' machines is another question. It is worth noting that if we have computational power to simulate the human brain, then we will have the computational power to run cognitive algorithms which do not necessarily emulate the human brain in any way.
Posted by: BrianX
|
August 19, 2010 12:22 PM
Greylander: Biggest. Wanker. Ever.
Posted by: Greylander
|
August 19, 2010 1:04 PM
@ #548 BrianX,
Ah gee Brian, thanks for adding such an insightful comment to the conversation.
I'm flattered. I do indeed wank. But I fear I have to spoil your fantasies, as I am only average in size.
Posted by: Greylander
|
August 19, 2010 1:27 PM
@ #543, John Morales
It's all good.
There certainly are very good questions to be raised the answers to which could push back the timeframe for various technological advancements.
It's a worthwhile discussion.
Time will indeed tell.
Posted by: Greylander
|
August 19, 2010 1:30 PM
@ #546 Stephen Wells
Not ignoring you. As you can see I got sidetracked writing silly sleep-deprived sketch inspired by your line about consciousness as a divide-by-zero error.
Posted by: Greylander
|
August 19, 2010 1:39 PM
@ #541,PZ Myers
PZ,
Not sure to what extent this is addressed to me but, I just remembered my original comment includes references to "self-extracting" compressed files, and I want to clear something up that may have thrown you and others off about what I was saying. I referred to self-extracting files in two ways. The first was to explicitly point out more or less what you are saying here. Even if you have a "self-extracting file" it is useless without the right machine to run it on. The second was expressly in reference to a self extracting compressed version of the 10^9GB of genome down to 10^8 bytes. I only mentioned self-extracting in this case because I was being careful to be clear no information was "hidden" in the decompressing program. But the self-extraction here only refers to expanding from the 10^8 compressed file to the full 10^9 byte genome. This had nothing whatever to do with expanding the genome up to a human body.
Posted by: alexandertolley
|
August 19, 2010 1:56 PM
While Kurzweil is wrong in his statement about genomes and computer code, I think you are wrong in the opposite direction, assuming greater complexity is needed than necessary.
1. Kurzweil is only asking for a design in principal, not a complete simulation.
2. There is no a priori need to know about proteins and protein folding to simulate a human type brain. We just need to model the gross characteristics of the brain. That is much simpler.
3. If we had a reasonable model of brain development, then we should be able to code that model te brain without having to define every neuron and connection.
I suspect it is this last that Kurzweil was trying to express. That if we can model the recipe and the baking, this will be simpler than trying to replicate the final product.
Beyond that, we know that Kurzweil is wildly optimistic about progress towards the singularity, but that he fervently hopes to not be "left behind" when it comes. Rapture of teh nerds, indeed.
Posted by: Alan Broadmoor
|
August 19, 2010 2:34 PM
Hmmm. So Ray Kurzweil is wrong about one thing. I guess that makes him a kook.
Posted by: https://me.yahoo.com/a/_SwATXQImOlKHnEUpIKZ4HQTZlpgWTs-#c18aa
|
August 19, 2010 3:37 PM
I don't think Kurzweil is arguing that you need to simulate the development of a brain from placenta to adulthood. You map the gene expression patterns, connections, synaptic proteins (etc.) of an adult brain and then model them in silico (like Blue Brain is attempting to do). This is not an easy feat, but it is totally different from the straw man you set up.
A large portion of genes are non functional and only a subset are expressed in the brain. I don't think you need to waste memory on ones that are not needed. While the body does provide support for the brain, scientists don't really need to simulate the biology of the heart or the stomach to begin to understand how the mind works.
Kurzweil has said that an entire brain simulation would require Petabytes of memory (at the very least). I think he was referring to RAM, but I'm not sure. The hard drive space needed for the genome is not particularly high. Modeling all the other stuff is obviously more difficult. He has actually said that the genome doesn't specifically encode for the wiring of the brain. So trying to pass off the cliched "biology is more than the genome" as if it some arcane knowledge is not particularly impressive. Also the quote was taken from a gizmodo article and was probably out of context. I'd would tend to give him the benefit of the doubt.
I think Ray's main point is how some relatively simple rules (or data if you will) can lead to a structure that appears to be vastly more complex. You may be right that he ignores a lot of the information that is "encoded in the environment" when discussing the brain's development. He understands more about this topic, though, than you are giving him credit for.
Posted by: theswede
|
August 19, 2010 3:46 PM
There is no a priori need to know about proteins and protein folding to simulate a human type brain. We just need to model the gross characteristics of the brain. That is much simpler.
Please explain what connection this has to the comment by RK about genome data content. If you can't, your supposition about what RK meant to say is trivially wrong.
You're addressing a lot of things which have not been said, and say nothing at all about what RK actually did say, and PZ commented on.
Posted by: https://me.yahoo.com/a/_SwATXQImOlKHnEUpIKZ4HQTZlpgWTs-#c18aa
|
August 19, 2010 3:47 PM
In the previous comment I meant to say that a large portion of the genome is non-functional. I'm not sure how much hard drive space it takes to describe non-junk DNA.
Posted by: Stephen Wells
|
August 19, 2010 3:53 PM
I'm so relieved to hear that "there are highly stable methods for solving numerical differential equations such that the error tends to stay small and random", greylander. I can tell several of my colleagues to stop their research immediately- their field has been completely solved!
There are methods for solving some PDEs under some conditions.
But I'm sure the unknown network of an unknown but very very large number of PDEs which represents a brain is completely tractable.
Posted by: theswede
|
August 19, 2010 3:55 PM
You map the gene expression patterns, connections, synaptic proteins (etc.) of an adult brain and then model them in silico
What the hell is this sentence trying to convey?
He understands more about this topic, though, than you are giving him credit for.
How would someone who puts together the sentence I quoted above know anything about this?
Posted by: https://me.yahoo.com/a/_SwATXQImOlKHnEUpIKZ4HQTZlpgWTs-#c18aa
|
August 19, 2010 4:10 PM
@Theswede
Connectome, allen brain atlas, etc. There are many "mapping" projects that are ongoing and explaining them in explicit detail was not that point of the comment.
Posted by: BrianX
|
August 19, 2010 4:11 PM
Greylander:
Why shouldn't I call you a wanker? All we've gotten from you and the rest of the Singularitarians here is a bunch of special pleading, irrational petulance, ass-backwards logic, and unconvincing yakkity blah blah. Your main contribution to the whole thing has been some epic tl;dr that's no more convincing than anything your compatriots have come up with, though it's one hell of a lot wordier. (Argumentum ad Filibustrum?) Not one of you has shown an ability to distinguish between "probable", "possible", and "likely"; you keep digging a hole with a sign that says "here be dragons" and you don't actually know where to get the dragons.
Alan Broadmoor:
Bullshit, and if you're any kind of rationalist at all you should know that. It's not that he's wrong; it's that he refuses to listen to people who tell him why he's wrong.
Posted by: theswede
|
August 19, 2010 4:13 PM
Specifically; what are "gene expression patterns", and how does one "map" them, and what is the purpose of doing so?
It was very clear that the comment had no appreciable point other than apologetics, no worries.
Posted by: Alan Broadmoor
|
August 19, 2010 4:17 PM
@BrianX:
Has Ray Kurzweil refused to listen to people who tell him why he's wrong about specific ideas on a regular basis?
I say "on a regular basis" because everyone is guilty of refusing to listen occasionally.
Posted by: https://me.yahoo.com/a/_SwATXQImOlKHnEUpIKZ4HQTZlpgWTs-#c18aa
|
August 19, 2010 4:19 PM
@Theswede
What are you not understanding? Do you not get the concept of mapping the brain?
The open access Allen Human Brain Atlas combines four data sets: anatomical images of the human brain obtained by MRI scans (to show gross anatomy), diffusion tensor imaging (which reveals the fiber tracts connecting different brain regions), histology (to show cellular level anatomy), and gene expression data. The result, scientists hope, will provide the most detailed look yet at which genes are active in which parts of the human brain.
Posted by: BrianX
|
August 19, 2010 4:24 PM
Alan:
I'm sure there are biologists who have told him what PZ is saying right now, and if he was listening to them, his response would be along the lines of "well, we don't know for sure, but we're working on it", not "the Singularity will be here in X number of years". Not to mention pulling out the "tu quoque" is very bad form. When defending your side in a debate, trying to divert attention from your argument's weaknesses is a neon sign saying "DO NOT TAKE ME SERIOUSLY".
Posted by: https://www.google.com/accounts/o8/id?id=AItOawnHlxPZxlg9U-16cuUCIAUL0drBqHPO2qo
|
August 19, 2010 4:42 PM
There's also a significant difference between having an artificial human brain (i.e. a piece of hardware that emulates the neural activity of a human's grey matter), and developing an artificial thinking machine.
The purpose is to develop a computer that can learn, grow, and expand itself. Less of an Artificial Intelligence, and more of an Evolving Intelligence.
How close or far we are from understanding every chemical and neurological nuance of the brain does not necessarily preclude developing a computer that can emulate the relevant activities. Our artificial construct doesn't need to accept pain/pressure/temperature input from its left big toe. It doesn't need to be able to interpret inner-ear fluid levels to keep it upright. It needs to be able to learn, change, and evolve.
We're closer to a victory on this front than you believe. I personally suspect, however, that even once we DO have such a machine, the education phase of its development will be the real challenge. "Raising" said computer will be a decades-long task in and of itself.
Posted by: Greylander
|
August 19, 2010 4:57 PM
#561 BrianX:
By all means, BrianX, you should call me a wanker. As you seem incapable of launching a cogent counter-argument to anything I've said.
Besides. I wank. Ergo, I'm a wanker. I wanted to spare you from delusions about my large size, which seemed to be on your mind.
May I call you attention to your own words in #199?
I have quite clearly shown why PZ is dead wrong in his interpretation of what Kurzweil said.
I have quite clearly shown what Kurzweil actually meant.
If I am so wrong, why are you not bothered to answer my objections?
Or is calling me a wanker the best you can do? If so, that's fine. Please continue. It's kinda cute.
Wow. I am amazed that I succeeding in exhausting the vast expanses of your attention span.
You sure do use an awful lot of words (below) just to say, "I think all you people who disagree with me are wrong."
Posted by: Greylander
|
August 19, 2010 5:03 PM
#565 BrianX:
No. His reponse would be something like, "Is that what you actually thought I meant? Are you stupid, ignorant, crazy, or just a... [wait for it...] wanker?"
Posted by: CJO
|
August 19, 2010 5:07 PM
How close or far we are from understanding every chemical and neurological nuance of the brain does not necessarily preclude developing a computer that can emulate the relevant activities.
You're encapsulating here the hand-waving that everyone's objecting to. What are "the relevant activities"? You don't know, and nobody else knows either. Biology is messy, highly redundant, and the "chemical...nuances" are often of disproportionate significance in non-obvious ways, surely never moreso than in the interconnected intricacies of the cerebral cotex.
You'd like to have it as: we just need to simulate the thinking part. But best guess is there isn't such a part, or easily delineated subset of the activities, of a functioning brain. The mind appears to be a holistic emergent phenomenon, dependent on the whole operation of a functioning, embodied brain. You want the big toe and the inner ear to be irrelevant, so you assume they are.
Posted by: theswede
|
August 19, 2010 5:22 PM
@564: Do you not get the concept of mapping the brain?
I do not get how the complexity of this concept in any way can be directly expressed through the compressed size of the genome. Please explain it, because the lack of that connection is why RK is being called ignorant.
I hope that was the point you were shooting for. If you were only after stating the obvious which no-one will contest, and which the OP is not about, then I'm not interested in hearing it. Tons of singularites have regurgitated the obvious, ignoring the actual points made.
Posted by: HertfordshireChris
|
August 19, 2010 5:27 PM
Thanks BrianX #320 and KG #413 for your comments.
Clearly the points I raised are proving to be off the main topic of the debate here, but let me just say a few words in reply.
(1) If you want to model what the brain does at the neuron/cell chemistry level you need to understand what it actually does. My interest was in trying to build a tool which would work with humans in handling non-trivial open-ended tasks where it was not practical to specify requirements in advance. Basically there has to be two-way way interaction, with the tool “learning” and “understanding” what the user is telling it, and then implementing the user's own logic to make deductions. Even if such a tool did not accurately model how people handle open-ended tasks, etc. it could well through useful light on what the “user” brain was doing with the information (and it may be that different users did things differently) It is at this level it could be relevant to brain modelling
(2) It is easy to be pedantic about what a computer is – and what can be simulated and we could go round in circles arguing about this. The initial project was to examine the feasibility of building a white box processor which directly processed a knowledge language which was the user language – so that, for example, if anything went wrong the processor could only provide a memory dump in the user language – because that was the only language the processor knew. This is totally different to the “user interface “ of a stored program computer approach – where vast quantities of software are needed to hide the fact that the computer is, internally, a black box whose machinations are totally alien any normal human being.
BTW, BrianX, while I was trying to find time to do research I was teaching human computer interaction (in the conventional sense) and also acting as book review editor on an early online computer-human factors forum – so I am well aware that what I was doing was not just building a user interface.
(3) I would be the first to agree that there are many applications, particularly those involving vast numbers of numeric calculations, and possibly situations where two machines talk to each other using a rigorously defined protocol, where the approach I suggest would be unsuitable. All I am suggesting is that people do not naturally think in terms of pre-defined programs, and that where they appear to it is because they are carrying out tasks which are artefacts of modern civilization. To handle such situations one needs a totally different philosophy – which rejects the need to split knowledge artificially into code and data.
(4) KG I am interested you mentioned Maggie Bowen. Many year ago I attended a seminar she gave on the Artificial Intelligence approach to writing stories. I asked whether she agreed that we would not have cracked the problem until we could ask a story writing system to write the story “How I write stories”. Her reaction was that this was impossible, I assume this because she thought writing programs required an “intelligent designer” (i.e. her research students) to pre-define the story writing rules. As far as I am concerned the human brain works with some kind of knowledge base where there are no rigid boundaries between different kinds of information. If a story-writing simulation is human-like in its actions it must be able to write about anything in its knowledge base, including general statements about how to construct a story.
Posted by: BrianX
|
August 19, 2010 8:53 PM
No, Greylander, you have failed utterly and completely and made yourself look like a stubborn moron in the process. I chose my words the way I did because I can't be bothered to take you seriously anymore, because your logic is complete and utter nonsense and sounds exactly like every other creationist/global warming denialist/etc. out there who's trying to convince people that there's room for their particular view. You are a frustrated evangelist looking for an excuse to have a superiority complex over the mainstream, nothing more. Bugger off already.
Posted by: mvanbebber
|
August 19, 2010 9:02 PM
@PZ Meyers
So now Henry Markram also has "no knowledge" of the brain? Now we have added the full professor and founder/director of the Brain Mind Institute and Director of the Center for Neuroscience and Technology at the Ecole Polytechnique Federale de Lausanne in Zurich to the list of idiots who know nothing about the brain?
http://bluebrain.epfl.ch/page18900.html
Again, I remind you that Kurzweil claims we will understand the brain in 20 years, not 10, as Markram and others claim, so you are wrong in two distinct ways. Hey, wait a minute, you're wrong about something - doesn't that make you a kook?
Posted by: BrianX
|
August 19, 2010 9:07 PM
In fact, Greylander, I think there's a bit more to this. Maybe to you, your logic is self-evident. But you sure as hell haven't made your case to us, since you haven't made a case that Kurzweil knows a damn thing about what he's claiming to prognosticate about. If anything, you and your compatriots sound like you think you're in on some great secret of the universe that everyone will eventually come around and realize, while the people who actually work on and understand the basis for your prediction are laughing their asses off at you. You can claim all you want that it's just because people don't like what you say, but you're just lying to yourself, and on some level you know it.
Artificial intelligence, as a field of study, has been stagnant for over twenty years, with most of the productive work being on ancillary issues like game theory, fuzzy logic, and handwriting recognition, in the process finding out that some of these things don't really deal in what we would consider sentience at all, and even those things that might be related, we still don't know how to string together in the form of a recognizable AI. Kurzweil, like many other singularitarians, is in denial about this, and is fundamentally misunderstanding the nature of complexity in computing. Saying that strong AI might be possible is reasonable, since we can't prove differently just yet. But actually predicting it is another story, and that's why people consider Kurzweil a crank.
Posted by: https://www.google.com/accounts/o8/id?id=AItOawlEkvo68pvTmhzetxlSvvb-8J5IcWbddnU
|
August 19, 2010 10:30 PM
There shouldn't be much question that Kurzweil's path to a conclusion about when we can reverse engineer the brain just by counting DNA base pairs is way off base.
However, I'd say that pointing this out as a flaw of reasoning (which it is) is actually not the point in the first place.
We can reengineer things without knowing how they, or their constituent parts, were initially constructed. We develop new materials, for example, without understanding how the atoms were constructed. We observe the materials and construct theories about how they work, but we do not by any means say that we know how the component atoms were themselves constructed nor do .
So we should, theoretically, be able to keep separate our understanding of how the brain is constructed from how it works. Both are wildly complex but we don't require knowing the latter in order to make progress on the former.
We won't have a full understanding until we can completely explain and understand the whole process of construction and operation, but we can make great progress well before then. By that measure we really don't understand *anything* completely today but that doesn't stop us from making progress.
Posted by: https://me.yahoo.com/a/_SwATXQImOlKHnEUpIKZ4HQTZlpgWTs-#c18aa
|
August 19, 2010 11:11 PM
Kurzweil does say a lot of crazy things and is wrong quite a bit. I think he takes his prognostications way too seriously. However, he does come up with some novel ways of looking at things. It seems absurd to compare him to deepak hoopla.
Neuroscientist Henry Markram agrees with some of what Ray says. I might add that if you check google scholar, Markram's top paper is cited 1542 times vs. 323 for Myers highest one. Markram is also in a better position than Myers to judge the feasibility of such an undertaking. Remember, he was furious at Dharmendra Modha's claims of simulating a "cat" brain. So it's not like he doesn't understand the difficulty involved.
Posted by: https://me.yahoo.com/a/_SwATXQImOlKHnEUpIKZ4HQTZlpgWTs-#c18aa
|
August 19, 2010 11:26 PM
The gizmodo took the quote out of context. See this article.
http://singularityhub.com/2010/08/19/when-will-computers-match-the-human-brain-kurzweil-vs-myers/
Posted by: hix1050
|
August 20, 2010 12:40 AM
Historically, there have been several plausible claims for methods for duplicating the human brain.
Hans Moravec's proposal was to take a small element of the brain, stimulate it, and duplicate it in software. Repeat until complete.
I believe it was John von Neumann that made the claim that given a description of any element of human behavior, he could write a program that would duplicate it.
However, these are physical duplication processes, not what Kurzweil proposes.
On the same track, I'm aware of two current efforts to section an entire mouse brain and read the sections by electron microscopy, at very high throughput; one using thick sections and one thin sections, and reassemble the brain in a computer, which will get a connectional map of the mouse brain. The first runs will probably take place in a year or so.
My own impression of the human mind is that it is a collection of genetically mixed computational elements with a tendency to self-organize. Which is reminiscent of one of the descriptions of a helicopter; a collection of parts flying in loose formation.
Posted by: Ichthyic
|
August 20, 2010 12:46 AM
Hans Moravec's proposal was to take a small element of the brain, stimulate it, and duplicate it in software. Repeat until complete.
except there is a lot of evidence of cross communication, so that's a fail.
I believe it was John von Neumann that made the claim that given a description of any element of human behavior, he could write a program that would duplicate it.
which says nothing about how behavior is processed.
However, these are physical duplication processes, not what Kurzweil proposes.
what Kurzweil proposes is JUST as disconnected from reality.
*shrug*
Posted by: R. Schauer
|
August 20, 2010 8:19 AM
I've read the Singularity and must say it made me think of how cool our futures could be in a sci-fi'ee way. To wit: if on face we observe what he brings up regarding the brain, we can see he's underestimated but at least he's thinking about something interesting to think about and he infects us into thinking about it, too. In summary, he may be wrong about stuff but at least he gets us thinking about interesting future AI/human developments.
Posted by: John Morales
|
August 20, 2010 8:31 AM
R. Schauer:
Well-trodden territory.
Remember Asimov's three laws of robotics?
As for the singularity, well, Vernor Vinge did it earlier, and Stanisław Ulam did it before that.
To give Kurzweil credit to the idea is akin to give Wolfram credit for cellular automata. All he's done is popularise it.
Posted by: Lucifer
|
August 20, 2010 10:04 AM
Kurzweil responds
Posted by: cosmicaug
|
August 20, 2010 11:12 AM
The funny thing, is that a while ago I had a similar argument with someone and I was basically stating what P.Z. Myers is saying on this post (though not doing such a good job) but my correspondent was actually taking the position that there isn't enough room in the genetic code for the information required to produce a human brain and therefore there must be something else out there (that is, he was trying to, noncommittantly, imply dualism).
In other words, my correspondent was adopting the same naïve 'DNA as computer program' view of developmental biology as Kurzweil and arriving to a radically different conclusion that not only would we never be able to simulate a brain (much less its development) but that the information needed to create one is not physically present and thus an outside source of information is required. I guess he was looking badly for a soul.
Posted by: Balstrome
|
August 20, 2010 11:13 AM
Rebuttal or so it seems
http://www.kurzweilai.net/ray-kurzweil-responds-to-ray-kurzweil-does-not-understand-the-brain
Posted by: https://www.google.com/accounts/o8/id?id=AItOawn0fHzaHilGRsyxb0tsRVZ3s-yHo0UVOSI
|
August 20, 2010 11:15 AM
Read Kurzweill's response here:
http://www.kurzweilai.net/ray-kurzweil-responds-to-ray-kurzweil-does-not-understand-the-brain?utm_source=KurzweilAI+Weekly+Newsletter&utm_campaign=62b6882ca7-UA-946742-1&utm_medium=email
Posted by: Balstrome
|
August 20, 2010 11:16 AM
Snap
Posted by: liquiddark.myopenid.com
|
August 20, 2010 12:05 PM
@BrianX:
Fair enough. The earliest written languages that we have any evidence for, however, were certainly pictorial in nature. The connection between communication and visualization has been around forever because they are intrinsically connected phenomena for a creature as vision-dependent as our species. To claim that there is not a major element of linguistics/semiotics in the study of the mind seems to me pure folly.
Posted by: BrianX
|
August 20, 2010 12:13 PM
Aaaaaand Kurzweil responds with a late Galileo Gambit... ooh, yeah, not the wisest maneuver.
Posted by: Timberwoof
|
August 20, 2010 12:54 PM
Slashdotted again:
http://science.slashdot.org/story/10/08/20/1429203/Ray-Kurzweil-Responds-To-PZ-Myers
Ray Kurzweil's blog entry:
http://www.kurzweilai.net/ray-kurzweil-responds-to-ray-kurzweil-does-not-understand-the-brain
Posted by: Mario3k
|
August 20, 2010 1:16 PM
PZ Myers imediatly reminded of this quote:
"Heavier-than-air flying machines are impossible."
- Lord Kelvin (1824-1907), ca. 1895, British mathematician and physicist
Posted by: BrianX
|
August 20, 2010 1:34 PM
Kelvin also thought the Earth couldn't be more than 100 million years old. It took the discovery of radioactivity to prove him wrong, because he didn't have the data to come to a different conclusion.
Posted by: Greylander
|
August 20, 2010 1:40 PM
@ #588 BrianX,
Really Mr. X, would you care to provide the quote that you think constitutes a Galileo Gambit? Perhaps you think that mentioning one's critics in and of itself is a Galileo Gambit?
Perhaps you think illustration of the failure of linear prediction in the face of exponential growth is somehow a "Galileo Gambit". Rightly or wrongly, he is criticizing his critics not whining that he is correct simply because he has critics. For that matter, right or wrong, he is wealthy and gets good press... hardly the vilification requisite for the Galileo defense.
Kurzweil does make clear that PZ thoroughly mis-represents him in PZ's OP. Exactly what I said in my first comment.
Posted by: CJO
|
August 20, 2010 1:41 PM
The earliest written languages that we have any evidence for, however, were certainly pictorial in nature. The connection between communication and visualization has been around forever because they are intrinsically connected phenomena for a creature as vision-dependent as our species.
There are no "written languages." There are writing systems that are a technology used to represent natural spoken languages, which have existed several hundred times longer than writing, and they use sounds. However "vision-dependent" we are, that is not the channel we have been primarily using to articulate our thoughts and communicate complex ideas to each other, though facial expressions, gestures and posture are all salient communication channels for other primates as well as for us, and they've been around longer than language. Which is by way of saying that you were partially correct about such a connection being around "forever" but only in the trivial way that animals can and do use all of the sensory modalities available to them for communication, vision included.
To claim that there is not a major element of linguistics/semiotics in the study of the mind seems to me pure folly.
Okay. Who made that claim?
Posted by: BrianX
|
August 20, 2010 1:44 PM
Greylander:
We've already established that you're an intellectually self-deluding pompous jackass. I'm talking to people who actually understand logic here.
Posted by: ghamlin
|
August 20, 2010 1:45 PM
PZ,
I think you make several logical assumptions for impact that are silly.
I didn't hear the original talk, but your argument is below what I would expect.
Is genomic data + the laws of physics an upper bound on the rules required to create consciousness of course it is. That is obvious. Is simulating every protein required for concousness? I highly doubt this....
I think as thinks like Diffusion Tensor Imaging improves we will actual neuron by neuron data for "interestingly sized" networks. As 3d printing of cell scafolding improves we can test man made biological neural networks of out design. We will discover the enough laws to govern a "human like" NN in 20 year ... I dunno, that is a good question, but your argument about this stinks.
Simulating somebody's head one protein at a time is not a sensible way to think studying NNs I don't think that was proposed. OTOH, the genome strawman you setup (with a small tweak to include some laws of physics) is in fact a valid (but dumb) way to set a ridiculous upper bound on the complexity of the actual rules needs for the human BNN (ala information theory). Its no different than comparing the amount of information is a math book or an english book by compressing both. Yes, its not an answer but its a _limit_.
We don't to perfect molecular mechanics to understand the NNs just like thermodynamics didn't need to wait for quantum mechanics.
Is 20 years wrong... probably, but don't play this straw man BS, please. A better counter argument might be whether DTI, cell printing etc could advance fast enough...
Posted by: stevieinthecity#9dac9
|
August 20, 2010 1:49 PM
Mario3K, you're missing the point.
If Lord Kelvin had said, "I don't think we'll be traveling at the speed of light any time soon."
Then it might apply to PZ.
Posted by: Greylander
|
August 20, 2010 1:52 PM
@ #372 PZ Myers,
You are a smart guy, PZ, so it boggles me that you would say something this stupid. It is clear, even in the article which manages to represent Kurzweils ideas pretty poorly.
Setting aside for the moment, whether the premise is wright or wrong, if we assume that the fundamental processes (or my phrase "the functional description") of the brain are encoded in the genome, then the information content of the genome provides an upper bound on the information content of any description of those fundamental processes.
If you retract your strawman, you can still ridicule that premise all you like, but we'll at least be one step closer to a reasonable conversation on the subject.
Posted by: Greylander
|
August 20, 2010 1:58 PM
@ #594 BrianX,
Damn, but you are cute, Mr. X. Since you are talking to me, at least I know that I am an intellectually self-deluding pompous jackass who actually understands logic.
Posted by: BrianX
|
August 20, 2010 2:04 PM
Also a Dunning-Kruger case, but I wouldn't expect you to realize that.
Posted by: Mario3k
|
August 20, 2010 2:06 PM
PZ Myers also reminds me of this quote:
Computers in the future may...perhaps only weigh 1.5 tons.
- Popular Mechanics, 1949.
Posted by: stevieinthecity#9dac9
|
August 20, 2010 2:33 PM
If the brain were so simple that we could understand it, we would be so simple we couldn't.
— Emerson M. Pugh
I'm feeling that Ray is too simple to understand it.
Posted by: rbairos
|
August 20, 2010 2:43 PM
Having read PZ's editorial and Kurzweil's response, its clear who is being honest and who is looking for internet attention.
Its disheartening to see so many self-described rational, critically-thinking people in this forum madly contributing to the feeding frenzy of Meyer's sensationalist strawman attack.
Posted by: stevieinthecity#9dac9
|
August 20, 2010 2:49 PM
Oh for fuck's sake. It's MYERS.
PZ has been pointing out Ray's foolishness for years. PZ gets plenty of attention. The amount he gets from Ray's fans is probably minimal in comparison.
The brain is the most complicated kilo of matter in the universe.
— Anonymous
Posted by: CJO
|
August 20, 2010 2:51 PM
Having read PZ's editorial and Kurzweil's response, its clear who is being honest and who is looking for internet attention.
Sez you. Make a fucking argument.
Posted by: Steve LaBonne
|
August 20, 2010 2:54 PM
Kurzweilbot to English translation: "Myers uses big words in a subject I don't understand at all so I'm going to stick my fingers in my ears and yell "la la I can't hear you!", just like a creationist."
Can't believe these ignorant twits are still at it.
Again, >= 10^14 synapses, whose strength is constantly being modulated by a whole host of inputs internal and external. You people have NO FUCKING IDEA, not within many many orders of magnitude, of the degree of complexity you blithely bullshit about emulating. In the most literal sense you don't know what you are talking about. Successfully simulating the functionality of just very very small bits of that circuitry is a life's work for a researcher. Things as "simple" as facial recognition are nowhere even close to being solved to anywhere remotely near the level at which the brain effortlessly operates, despite massive investment because of the potential commercial payoff.
Posted by: rbairos
|
August 20, 2010 3:11 PM
Steve LaBonne, CJO
Did you or did you not read Kurzweil's reply, which clearly states PZ Meyer's blatant misrepresentation of his position on the genome wrt brain simulation for example?
For all the ad-hominems flying about, why is no one discussing the specifics of the reply?
Instead its a name calling free for all.
Posted by: Sven DiMilo
|
August 20, 2010 3:15 PM
qft
Posted by: Steve LaBonne
|
August 20, 2010 3:21 PM
In other words, you have no response to what I pointed out. FAIL.
Posted by: rbairos
|
August 20, 2010 3:22 PM
Steve LaBonne rants:
"Things as "simple" as facial recognition are nowhere even close to being solved to anywhere remotely near the level at which the brain effortlessly operates, despite massive investment because of the potential commercial payoff."
Two seconds of research reveals:
"In 2006, the performance of the latest face recognition algorithms were evaluated in the Face Recognition Grand Challenge (FRGC). High-resolution face images, 3-D face scans, and iris images were used in the tests. The results indicated that the new algorithms are 10 times more accurate than the face recognition algorithms of 2002 and 100 times more accurate than those of 1995. Some of the algorithms were able to outperform human participants in recognizing faces and could uniquely identify identical twins.[6]"
^ a b c d e Williams, Mark. "Better Face-Recognition Software". Retrieved 2008-06-02.
http://en.wikipedia.org/wiki/Facial_recognition_system
So, you know, your grand generalized dismissals aren't relevant or even accurate.
Feel free to swear and insult again..
Posted by: BrianX
|
August 20, 2010 3:23 PM
Mario3K, 600:
Well, there should be a computer or two from the late 50s and early 60s that is approximately 1.5 tons in weight, thus fulfilling the prophecy. Heck, give me enough power strips and an unlimited expense account at Newegg, I could build one of those for you easily.
Posted by: rbairos
|
August 20, 2010 3:26 PM
"Instead its a name calling free for all.
In other words, you have no response to what I pointed out. FAIL"
Errr...
You did read my reply as you quoted from it no?
Just in case, here it is again:
"Did you or did you not read Kurzweil's reply, which clearly states PZ Meyer's blatant misrepresentation of his position on the genome wrt brain simulation for example?
For all the ad-hominems flying about, why is no one discussing the specifics of the reply?"
Posted by: Steve LaBonne
|
August 20, 2010 3:27 PM
Right. Than turn your head most of the way around or put on sunglasses and a Groucho Marx mustache and see what happens. These systems are still a LONG LONG way from doing things that humans do effortlessly- and they do what they do in a very different, and much more simplistic and "rigid" way- and real researchers, as opposed to shills for commercial FR systems, will readily admit this.
Again, you understand nothing about the brain (nor does Kurzweil) and thus in the most literal sense don't know what you are talking about.
Posted by: CJO
|
August 20, 2010 3:29 PM
Yes I read the reply. No I did not call anyone names.
I merely pointed out that you asserted without doing anything to support your assertion. Calling people names isn't necessarily an ad hominem anyway, and saying ad hominem doesn't make you sound smart.
PZ did not misrepresent, other than perhaps the 1 decade/2 decade confusion, but Kurzweil said both 2019 and 2029 in the speech. He didn't misrepresent, he is saying that any information theoretical metric of the genome is utterly irrelevant to the nature of the enterprize Kurzweil so blithely puts forward as a small problem that increased computing power will magically solve.
Kurzweil's reply amounts to yet more irrelevant information theory bafflegab, an irrelevant analogy to the human genome project (refer to Sagan on Bozo the Clown), and some delusionally optimistic assertions about what trends in information processing mean for our ability to model complex and poorly-understood phenomena. The point that he's not addressing is that it's not an information problem, it's not a computing power problem; it's a science problem. Until we understand what our Greylander handwaves away as "the fundamental processes" of the brain, we will not be able to simulate them, not with a Ceres-sized chunk of computronium.
Posted by: rbairos
|
August 20, 2010 3:33 PM
"Right. Than turn your head most of the way around or put on sunglasses and a Groucho Marx mustache and see what happens. "
First of all, how exactly do you know how orientation-invariant these algorithms are? Or are you just guessing?
Secondly, putting on sunglasses and moustaches are disguising ones face. Since when does facial recognition involve ESP?
"These systems are still a LONG LONG way from doing things that humans do effortlessly"
Again, the relevant information you overlooked is the rate at which these algorithms are improving. 100 fold from 1995 to 2006, or ten fold from 2004 to 2006. That is the relevant info.
" and they do what they do in a very different, and much more simplistic and "rigid" way-"
Again, you have no idea what elements these heuristics include, so stop making up claims. Its tiresome.
" and real researchers, as opposed to shills for commercial FR systems, will readily admit this."
Great.. Somebody call Michael Moore.
"Again, you understand nothing about the brain (nor does Kurzweil) and thus in the most literal sense don't know what you are talking about."
Queue the ad hominems..
Posted by: Nerd of Redhead, OM
|
August 20, 2010 3:36 PM
Yawn, the Kurzweil fanbois don't and cant' get it. But then, that is to be expected. Science is hard, beyond their ken, and it just isn't, and won't be, there for what Kurzweil proposed for many more years than Kurzweil's WAG.
Posted by: Paul
|
August 20, 2010 3:37 PM
This is what bothers me the most about the Kurzweil-bots here. They keep talking about how increased information processing capability will mean that we can simply enlist a computer to give us an accurate simulation of the brain. I really don't see how it's any different than the example I gave in 474 -- according to most of the arguments supporting Kurzweil on the brain issue, we can simply wait for computers to get fast enough to tell us everything about the nature of subatomic particles regardless of what our knowledge level on the subject is (presumably, we simply trust that the computer is accurate in the knowledge it hands down from on high). It's insanity.
Posted by: Greylander
|
August 20, 2010 3:39 PM
@ #604 CJO,
So... it's OK for the the PZbots to make posts with no more content than to say how right PZ is or about how stupid Kurzweil & supporters are... but not the other way round?
I was going to say 'hypcrite', but I've notice that you actually do provide arguments. To be fair though, you should also call out those who agree with you when they make contentless "we're right, you are wrong" posts... or else just let it pass on all sides.
Not that anything requires you to be fair. You know, free speech, equal opportunity ridicule, and all that.
Posted by: stevieinthecity#9dac9
|
August 20, 2010 3:47 PM
Oh for fuck's sake. He has no evidence that PZ is just looking for internet attention. PZ clearly states his viewpoint and he didn't "strawman" Ray. He responded to the Gizmodo article.
He didn't make an argument, just stated that he thought PZ was looking for attention from his point of view. "Sez you."
In Ray's response he does nothing to diminish PZ's point. He basically reinforces what PZ said. Higher resolution imaging and faster information processing has nothing to do with understanding the functioning of the brain and replicating it's functions. They are just tools.
Posted by: Steve LaBonne
|
August 20, 2010 3:48 PM
People routinely see through disguises. That is a perfect example of the fact that what goes on in human facial recognition is far more sophisticated and subtle than what FR software does. FR systems emulate what the brain does with about the same fidelity with which an Edison phonograph emulated the sound of a symphony orchestra. The brain processes are not even nearly well enough understood to be actually emulated; instead a greatly simplified version of the task is being performed by very different and far less powerful means.
And there will be an uncountable number of such challenges, many of them involving far more complex functions, involved in trying to build a computer that can do what a brain does.
Posted by: BrianX
|
August 20, 2010 3:48 PM
So... it's OK for the the PZbots to make posts with no more content than to say how right PZ is or about how stupid Kurzweil & supporters are... but not the other way round?
"It's just a flesh wound!"
Posted by: BrianX
|
August 20, 2010 3:51 PM
@618:
Higher resolution imaging and faster information processing has nothing to do with understanding the functioning of the brain and replicating it's functions. They are just tools.
It truly amazes me that Kurzweil's acolytes can't get this through their heads. The term "nerd rapture" is far more appropriate than any of them are willing to admit.
Posted by: CJO
|
August 20, 2010 3:53 PM
Not that anything requires you to be fair. You know, free speech, equal opportunity ridicule, and all that.
Yeah, well, if I spent my time calling out every moron on the internet who makes unsupported assertions as their contribution to a discussion, I wouldn't have a whole lot of time for anything else. In this case, rbairos chose to do it as his grand entrance into a 600+ comment discussion where we've been, y'know, actually trying to make points and defend our positions. It just seemed particularly egregious to me.
Posted by: rbairos
|
August 20, 2010 3:59 PM
Steve LaBonne shifts the goalpost:
"People routinely see through disguises. That is a perfect example of the fact that what goes on in human facial recognition is far more sophisticated and subtle than what FR software does."
Steve, who is arguing AI FR beats humans all the time? (Even though the MIT review of the challenge claimed it does in many instances now, and that was 5 years ago).
"FR systems emulate what the brain does with about the same fidelity with which an Edison phonograph emulated the sound of a symphony orchestra."
Poor analogy, as we're all carrying 24 bit stereo sound containing gigs of music in our pocket now.
"The brain processes are not even nearly well enough understood to be actually emulated; instead a greatly simplified version of the task is being performed by very different and far less powerful means."
Im not sure what you're arguing?
That we've somehow reached the pinnacle of reproducing/recreating/surpassing tasks that require some level of intelligence?
Our that our rate of exponential achievement over the last few decades/centuries is simply an anomaly?
Or that our remaining brain mechanisms lie outside this approach altogether for some as of yet undiscovered phenomena you refuse to share?
Please explain.
Posted by: rbairos
|
August 20, 2010 4:02 PM
CJO writes:
", I wouldn't have a whole lot of time for anything else. In this case, rbairos chose to do it as his grand entrance into a 600+ comment discussion where we've been, y'know, actually trying to make points and defend our positions. It just seemed particularly egregious to me."
How many of them referenced his reply?
1 or 2? Instead it *was* a feeding frenzy based on Meyer's running with a poorly interpreted Gizmodo article.
Posted by: stevieinthecity#9dac9
|
August 20, 2010 4:04 PM
Who is this Meyers guy? I think you're on the wrong blog.
Posted by: Steve LaBonne
|
August 20, 2010 4:04 PM
There's only one set of goalposts in this discussion: performing all the tasks a living brain performs. As we see with FR, even the currently most advanced simulations of tiny isolated bits of that functionality currently fall way way way short. The point is very simple, the "nerd rapture" is a hell of a lot more than 2 decades off. No amount of whining and special pleading will change that.
You people are like children.
Posted by: rbairos
|
August 20, 2010 4:04 PM
"Higher resolution imaging and faster information processing has nothing to do with understanding the functioning of the brain and replicating it's functions. They are just tools"
Yes, since everything we *do* know about biology involved meditating on first principles.
Posted by: rbairos
|
August 20, 2010 4:09 PM
Steve LaBonne writes:
"There's only one set of goalposts in this discussion: performing all the tasks a living brain performs."
Nowhere was that stated.
If you read Kurzweils reply:
"we would be able to reverse-engineer the brain sufficiently to understand its basic principles of operation within two decades"
"As we see with FR, even the currently most advanced simulations of tiny isolated bits of that functionality currently fall way way way short."
References?
Numbers?
I have supplied references showing 5 year old technology outperforming people.
"The point is very simple, the "nerd rapture" is a hell of a lot more than 2 decades off. No amount of whining and special pleading will change that.
You people are like children."
Sucks to your auntie?
Posted by: Steve LaBonne
|
August 20, 2010 4:17 PM
"we would be able to reverse-engineer the brain sufficiently to understand its basic principles of operation within two decades"
FR is not even reverse engineering, it's trying to find other means to perform a greatly simplified version of one task WITHOUT understanding how the brain does it.
Again, you refuse to understand either this point, or in general how grossly both you and Kurzweil are underestimating the mind-boggling complexity subsumed in that seemingly innocent phrase "understanding its basic principles of operation". In the proposed time frame our level of understanding will still be at such a high level of abstraction that it won't bring us a millimeter closer to the nerd rapture. And it takes some knowledge of neurobiology, not just computer science, to grasp this. Without it you're rather in the position of a music critic who's been deaf from birth.
Posted by: BrianX
|
August 20, 2010 4:22 PM
rbairos:
Poor analogy, as we're all carrying 24 bit stereo sound containing gigs of music in our pocket now.
Two entirely different technologies, which is where your fallacy lies. At the time Edison developed the phonograph, digital communication was still in its infancy -- the teletype as we know it, only recently evolved from the one-way stock ticker, wasn't even ten years old; off-line storage (if you had it) consisted of paper tape; and I don't think anyone really had a concept of analog-digital-analog conversions until Harry Nyquist two generations later. (Hell, people were still copying landline Morse Code by ear for commercial uses into the 1990s.) There was no way at the time to reasonably anticipate digital audio, and even if someone did, there was no conceivable way to implement it since the science that made it possible (pulse code modulation) didn't exist till the 1930s, the theoretical basis wasn't understood until Claude Shannon's work in the late 1940s, the first workable digital recordings were made in the 1960s (in Japan and the UK), and a storage medium with sufficient bandwidth to carry it all didn't appear till the late 1970s. And digital audio didn't reach the masses until 1982.
So let's recap -- from the teleprinter (1870) to the commercial release of the compact disc (1982) was 112 years, and there was no way to even anticipate digital audio before the 1930s, except as a wild guess. Cognitive science and artificial intelligence have not yet reached the 1920s in terms of convergence, and from the looks of some of the Singularitarian arguments here, the relationship is likely to remain quasi-adversarial for some time.
Still think it's a bad analogy?
Posted by: Stephen Wells
|
August 20, 2010 4:28 PM
Greylander: "if we assume that the fundamental processes (or my phrase "the functional description") of the brain are encoded in the genome,"...
...then we would be making a wrong assumption.
It's encoded in the genome, the proteome, the environment, and the developmental process.
Posted by: rbairos
|
August 20, 2010 4:31 PM
"FR is not even reverse engineering, it's trying to find other means to perform a greatly simplified version of one task WITHOUT understanding how the brain does it."
Never claimed it was.
Wonder who brought up the topic as such a striking example of how slowly progress is apparently moving?
"Again, you refuse to understand either this point,"
see above..
But for the record, the human brain is not treated as a magical black box free from physical inspection.
" or in general how grossly both you and Kurzweil are underestimating the mind-boggling complexity subsumed in that seemingly innocent phrase "understanding its basic principles of operation"."
Thanks for sharing.
I look forward to reading your thesis quantifying your arguments against Kurzweil's computational or biological claims.
Posted by: stevieinthecity#9dac9
|
August 20, 2010 4:33 PM
He doesn't have to. PZ did. A thesis isn't required.
Posted by: Steve LaBonne
|
August 20, 2010 4:35 PM
What a fascinating example of massive projection.
Posted by: BrianX
|
August 20, 2010 4:36 PM
But for the record, the human brain is not treated as a magical black box free from physical inspection.
It better not be, or AI researchers are wasting everyone's time.
However... that has worked for some technologies in the past (the Compaq PC BIOS and the Samba protocol stack being the most famous examples I can think of). YMMV, though I think for AI it's kind of a long shot.
Posted by: Kristjan Wager
|
August 20, 2010 4:37 PM
Then perhaps you should read the original post again and the comments by people with much more knowledge about the subject than you (and Kruzweil). Personally I recommend KG's posts, but the stuff by BrianX, Stephen Wells, theswede, Steve LaBonne et al are also well worth reading.
You might learn a thing or two, and who knows, you might even be able to figure out the problems with Kruzweil's biological claims, and why they make his computational claims completely irrelevant, even if they were completely true (not something I am willing to grant given Kruzweil's abysmal track record on predictions).
Posted by: First Approximation, L'esprit de l'escalier
|
August 20, 2010 4:38 PM
Honestly, "exponential achievement of technology" is simlpy Kurzweil and fans' version of God-did-it. It offers pretty much the same in way of explanation.
Deus Ex (Exponentially Growing) Machina!
Posted by: rbairos
|
August 20, 2010 4:42 PM
Stephen Wells writes:
Greylander: "if we assume that the fundamental processes (or my phrase "the functional description") of the brain are encoded in the genome,"...
...then we would be making a wrong assumption.
It's encoded in the genome, the proteome, the environment, and the developmental process.
Kurzweil already agreed had your read his reply:
"I mentioned the genome in a completely different context. I presented a number of arguments as to why the design of the brain is not as complex as some theorists have advocated. This is to respond to the notion that it would require trillions of lines of code to create a comparable system. The argument from the amount of information in the genome is one of several such arguments. It is not a proposed strategy for accomplishing reverse-engineering. It is an argument from information theory, which Myers obviously does not understand.
The amount of information in the genome (after lossless compression, which is feasible because of the massive redundancy in the genome) is about 50 million bytes (down from 800 million bytes in the uncompressed genome). It is true that the information in the genome goes through a complex route to create a brain, but the information in the genome constrains the amount of information in the brain prior to the brain’s interaction with its environment.
It is true that the brain gains a great deal of information by interacting with its environment – it is an adaptive learning system. But we should not confuse the information that is learned with the innate design of the brain. The question we are trying to address is: what is the complexity of this system (that we call the brain) that makes it capable of self-organizing and learning from its environment? The original source of that design is the genome (plus a small amount of information from the epigenetic machinery), so we can gain an estimate of the amount of information in this way"
..In other words, you don't have to encode the world in order to quantify the difference between the genome that creates a human brain and a one that creates a carrot.
Posted by: Steve LaBonne
|
August 20, 2010 4:45 PM
Already done. The relevant quantity is 10^14. As I said much earlier in the thread, I invite you to make the grotesquely oversimplifying assumption that each of the 10^14 synapses in the brain can assume just 5 discrete levels of strength, and go ahead and calculate the number of possible states of a brain. It's more than possible that we will never understand beyond a fairly superficial proof-of-principle level what's really going on in there. Be that as it may, the nerd rapture will remain a pipe dream for a long time.
Posted by: rbairos
|
August 20, 2010 4:46 PM
Steve LaBonne writes:
'But for the record, the human brain is not treated as a magical black box free from physical inspection.'
"What a fascinating example of massive projection."
Exactly which side here is promoting a continuous intensive anatomical study of the brain and which side is calling it a hopeless mystery?
Posted by: Kristjan Wager
|
August 20, 2010 4:47 PM
Is Kurzweil really so ignorant that he doesn't understand that there is no brain before its interaction with its environment? Does he think it suddenly exists fully formed?
Posted by: Steve LaBonne
|
August 20, 2010 4:49 PM
If you think it's just about neuroanatomy, I don't even know where to start, except to invite you to read a good intro neurobiology textbook. I would also remind you that the exact pattern of those 10^14 connections is NOT hard-coded in the genome as even Greylander still doesn't understand after all this discussion.
Posted by: rbairos
|
August 20, 2010 4:49 PM
Steve LaBonne writes:
"Already done. The relevant quantity is 10^14. As I said much earlier in the thread, I invite you to make the grotesquely oversimplifying assumption that each of the 10^14 synapses in the brain can assume just 5 discrete levels of strength, and go ahead and calculate the number of possible states of a brain. It's more than possible that we will never understand beyond a fairly superficial proof-of-principle level what's really going on in there."
By that bit of numerology, we can never understand a desktop PC with 640K memory, as the number of possible states for RAM alone is 2^640K.
Seriously, thats a bad argument.
Posted by: rbairos
|
August 20, 2010 4:51 PM
Kristjan Wager writes:
"Is Kurzweil really so ignorant that he doesn't understand that there is no brain before its interaction with its environment? Does he think it suddenly exists fully formed?"
Uh.. No?
Posted by: Kristjan Wager
|
August 20, 2010 4:53 PM
rbairos, taken Kurzweil at his words, he really is so ignorant. He seems to keep believing that he can somehow separate the brain from its environment, using the genome as a path for this.
That is of course absolute nonsense, but then, that's par the course for him
Posted by: rbairos
|
August 20, 2010 4:53 PM
SteveLaBonne writes:
"If you think it's just about neuroanatomy, I don't even know where to start, except to invite you to read a good intro neurobiology textbook. I would also remind you that the exact pattern of those 10^14 connections is NOT hard-coded in the genome as even Greylander still doesn't understand after all this discussion."
Who thinks its just about neuroanatomy?
Who claims those 10^14 connections are hard coded in the genome?
Man, its like a strawman convention in here.
Posted by: Steve LaBonne
|
August 20, 2010 4:54 PM
Nope, because RAM is just a matter of what bit is stored at what address- the massive connectivity of the brain is completely lacking. The content of RAM can be addressed one bit at a time. There is not remotely any analogue of that in the brain. In a fundamental way, RAM is radically simpler than the brain ("radically" isn't even strong enough a word). So is even the most complex CPU.
Posted by: First Approximation, L'esprit de l'escalier
|
August 20, 2010 4:54 PM
The sad thing is some of Kurzweil and his fans' goals seem somewhat plausible and even desirable EXCEPT:
1) Their timeline is off. Whether 10 or 20 years it's far too soon given our ignorance. Yeah, experts have been wrong in the past, but that doesn't make Kurzweil right now.
2) There's a lot of handwaving. Powerful computers will solve it all. A real explanation is lacking. If we ever do get some of the things Kurzweil envisions it's because of people doing hard work and finding concrete answers/solutions. Science is actually interesting because there are concrete answers rather than hand-wavy dreams.
Posted by: rbairos
|
August 20, 2010 4:55 PM
Kristjan Wager writes:
"He seems to keep believing that he can somehow separate the brain from its environment, using the genome as a path for this."
Just read his reply detailing exactly why he does not claim this, nor ever did and we can all save some time..
Posted by: Steve LaBonne
|
August 20, 2010 4:56 PM
Nobody who has tried to make this excuse for Kurzweil has yet been able to explain just WHAT he was doing by babbling about the information content of the genome. Feel free to have a crack at it.
Posted by: rbairos
|
August 20, 2010 5:00 PM
Steve LaBonne writes:
"Nope, because RAM is just a matter of what bit is stored at what address- the massive connectivity of the brain is completely lacking. The content of RAM can be addressed one bit at a time. There is not remotely any analogue of that in the brain. In a fundamental way, RAM is radically simpler than the brain ("radically" isn't even strong enough a word). So is even the most complex CPU."
I hate arguing analogies of analogies however..
Are you claiming the brain employs some type of non-deterministic heuristics or not?
Would you only claim to understand a 286 PC by examining each of its 2^640K possible states or not?
I cant show the fallacy of your argument more simply.
Posted by: Nerd of Redhead, OM
|
August 20, 2010 5:02 PM
The fanboi still doesn't get it. He thinks computing power will solve his problem. The science of the brain is the real problem, and that won't be solved by algorithms and computation, but rather good old fashioned lab work. Which requires a hell of a lot of time, and several breakthroughs in techniques before it can even become routine, much less sufficient data available to begin the task of approximating the working brain. And the brain interacts with the environment during its development, and during its working lifetime. It is not a static organ.
Posted by: Greylander
|
August 20, 2010 5:02 PM
@ #637 Feynmaniac:
No no no... Past tense "did" is completely inappropriate. If "exponential..." equals "God"... then "God *will* do it."
Get your insults right.
I should also point out that we predict nothing that is not possible *in principle* given present scientific understanding of fundamental physics, unless you invoke some mystical baloney about souls or the nature of consciousness. For example, no singularity fan (with any sense) is predicting that we will overcome the speed of light in a century or a millenia or ever... because there is no evidence that this should ever be possible in principle, and if it is possible it will depend upon completely new discoveries in physics which are entirely unpredictable. Thus there is only a meaningful disagreement on how soon these things will be achieved. The PZbots (as opposed to PZfans which is a superset of PZbots, of which I am an element) here clearly at best think centuries... I suspect most think millenia or "so far in the future there is absolutely no point speculating").
Posted by: CJO
|
August 20, 2010 5:02 PM
Neither side is calling it a hopeless mystery. We're correctly characterizing it as a very difficult puzzle (the difference between a puzzle and a mystery being that a puzzle has a solution in principle). Your "side" however is ignoring the staggering complexity of the puzzle in favor of waxing ecstatic about our ever-increasing computational power, digressing into irrelevant claims about machine vision systems that are at best crude emulators of human capabilities, not simulations, and blithely ignoring every argument for the position that the limits of computation aren't the barrier to solving the puzzle.
And I've never thought that anyone involved in the conversation is in any way opposed to "a continuous intensive anatomical study of the brain" but so what? What follows from that study that advances your argument? The outcomes of our research in that area continuously highlight the irrelevance of the information content of the genome to the question, which was PZ's point in the first place.
Posted by: rbairos
|
August 20, 2010 5:03 PM
"Nobody who has tried to make this excuse for Kurzweil has yet been able to explain just WHAT he was doing by babbling about the information content of the genome. Feel free to have a crack at it."
He's putting a lower limit on the amount of information that creates a self-learning, adapting human-like brain in a biological environment.
For example the amount of information that would be required to describe the exact relevant difference between a human genome and a carrot genome.
Posted by: Kristjan Wager
|
August 20, 2010 5:04 PM
Of course he did. E.g. he said the following
Here he clearly seem to think that the environment only has an effect on the brain by providing information, but he clearly doesn't understand the environment's impact on the development of the brain - the end design if you will. Dismissing this, and just looking at the genome, vastly misunderstands and underestimates the factors involved.
Posted by: Stephen Wells
|
August 20, 2010 5:06 PM
I think Kurzweil is effectively arguing that an upper bound on the amount of information in a cellular automaton is the size of the rule set for the automaton.
Which means the entire information content of anything done by a Turing-complete computational machine is apparently bounded by the rule set for Conway's Life.
Let's introduce Kurzweil to Wolfram and watch the fun!
Honestly, "brain prior to its interaction with the environment" indeed. Brains GROW, slowly, inside FETUSES, inside MOTHERS.There's some environment prior to brain!
Sheesh.
Posted by: rbairos
|
August 20, 2010 5:07 PM
"The fanboi still doesn't get it. He thinks computing power will solve his problem. The science of the brain is the real problem, and that won't be solved by algorithms and computation, but rather good old fashioned lab work"
I am drowning in a sea of strawmen...
Posted by: Kristjan Wager
|
August 20, 2010 5:10 PM
Then stop building them
Posted by: rbairos
|
August 20, 2010 5:15 PM
"The fanboi still doesn't get it. He thinks computing power will solve his problem. The science of the brain is the real problem, and that won't be solved by algorithms and computation, but rather good old fashioned lab work"
Kurzweil goes to great lengths showing how raw computational power is insufficient to create meaningful AI at any level.
Seriously, your view of his arguments are incomplete at best, a caricature at worst.
Posted by: rbairos
|
August 20, 2010 5:25 PM
Feymaniac writes:
"Honestly, "exponential achievement of technology" is simlpy Kurzweil and fans' version of God-did-it. It offers pretty much the same in way of explanation."
As opposed to your creationist-like argument against macro evolution, which seems to believe no amount of progress in information processing, no matter how far or exponential will ever bring about an understanding (or re-creation) of human intelligence.
Posted by: First Approximation, L'esprit de l'escalier
|
August 20, 2010 5:25 PM
I was comparing both groups lack of explanation. Past or future doesn't really matter (I actually think it's quite appropriate that the theists are looking to the past).
FWIW, I actually think much higher of Kurzweil and his fans than of fundamentalists (not that that says much).
Posted by: Stephen Wells
|
August 20, 2010 5:30 PM
To put it another way, this quote: "The amount of information in the genome (after lossless compression, which is feasible because of the massive redundancy in the genome) is about 50 million bytes (down from 800 million bytes in the uncompressed genome). It is true that the information in the genome goes through a complex route to create a brain, but the information in the genome constrains the amount of information in the brain prior to the brain’s interaction with its environment."...
is wrong, because "the information in the genome" does not even constrain the information in one brain cell, let alone one brain! The information in one water molecule does not constrain the information in one ocean!
Really, anybody capable of producing the line "the amount of information in the brain prior to the brain’s interaction with its environment" is so ignorant of biology they shouldn't try to pee without professional supervision.
Posted by: Kristjan Wager
|
August 20, 2010 5:34 PM
I thought I just told you stop building strawmen. No one here argues that we won't ever understand, or even re-create, human intelligence.
What people here are arguing, is that Kurzweil and his fans, don't understand the complexity of the task, and have a simplistic (and wrong) understanding of how the brain works.
Kurzweil's response to PZ only demonstrated this further.
Posted by: First Approximation, L'esprit de l'escalier
|
August 20, 2010 5:34 PM
I believe no such thing. I think progress in technology is necessary, but not sufficient, to understanding/re-creating human intelligence.
Posted by: CJO
|
August 20, 2010 5:35 PM
As opposed to your creationist-like argument against macro evolution, which seems to believe no amount of progress in information processing, no matter how far or exponential will ever bring about an understanding (or re-creation) of human intelligence.
You're conflating two distinct things there, which has been a common equivocation in this debate on the part of Kurzweil's defenders. We're talking about a robust simulation of a functioning, adult human brain, not AI in general.
And, anyway, how would fast, exponential "progress in information processing" by itself bring about an understanding of human intelligence? Seriously, you people attribute magical powers to computers that they just don't have.
Posted by: rbairos
|
August 20, 2010 5:37 PM
"is wrong, because "the information in the genome" does not even constrain the information in one brain cell, let alone one brain! The information in one water molecule does not constrain the information in one ocean!"
The information in one transistor does not constrain the information in one memory bit!
The information in one pen stroke does not constrain the information in one character!
Hallelujah..
Posted by: stevieinthecity#9dac9
|
August 20, 2010 5:37 PM
They clearly don't understand how far from understanding they are.
Posted by: rbairos
|
August 20, 2010 5:40 PM
Feymaniac writes: "I believe no such thing. I think progress in technology is necessary, but not sufficient, to understanding/re-creating human intelligence."
Are we just arguing semantics now?
Technology is a pretty open ended word.
For example, what more outside technology is necessary to understand/re-create an eye pupil?
Posted by: Kristjan Wager
|
August 20, 2010 5:41 PM
Yet you seem to believe that just by knowing one pen stroke, one can re-create the works lost in the fire of the Alexandria Library.
Posted by: Ichthyic
|
August 20, 2010 5:41 PM
shorter rbairos:
"If you're ignorant and you know it, clap your hands!"
Posted by: rbairos
|
August 20, 2010 5:42 PM
Kristjan Wager writes:
"Yet you seem to believe that just by knowing one pen stroke, one can re-create the works lost in the fire of the Alexandria Library."
Man, if you want to carry on this analogy, I believe the pertinent information lost in the fire of Alexandria is not dependent upon the chemical composition of the paper it was written on.
Posted by: Stephen Wells
|
August 20, 2010 5:46 PM
Or yet another way (this is fun!) : Kurzweil is conflating the information content of a genetic sequence written down (ACGTA...) with the information content of a complete description of what that gene does. We have one; we don't have the other.
Which is to conflate "One if by land, two if by sea" with the United States of America.
rbairos, I take it you concede that Kurzweil is utterly wrong about the quoted claim?
Posted by: Ichthyic
|
August 20, 2010 5:47 PM
create an eye pupil?
you mean a lens?
yeah.
oh, and a lens is SURELY the same thing as a brain.
keep clapping.
Posted by: rbairos
|
August 20, 2010 5:49 PM
"And, anyway, how would fast, exponential "progress in information processing" by itself bring about an understanding of human intelligence? Seriously, you people attribute magical powers to computers that they just don't have."
Its not an isolated component in the path to understanding human understanding...
However, I'll add the quest for understanding *human biological* intelligence is secondary to the goal of understanding/creating intelligent systems in general.
Posted by: helivoy
|
August 20, 2010 5:50 PM
Kurzweil et al are preaching a mix of dualism and genetic determinism. Very nineteenth century. Actually, more like ninth century, plus the "cutting-edge" word salad.
Ghost in the Shell: Why Our Brains Will Never Live in the Matrix
Miranda Wrongs: Reading Too Much into the Genome
Athena Andreadis, aka helivoy
Posted by: Kristjan Wager
|
August 20, 2010 5:50 PM
No, that's not a correct analogy. More like: you think that by just knowing what letters were used (perhaps even what numbers of each letter), will make it possible to recreate the works, ignoring the use of paper, the subjects of the works, and even the languages. Yes, it is theoretically possible, but highly unlikely until much more knowledge has been gained through further studies.
And no, better and faster pens (or even more writers) won't solve the problem of lack of knowledge.
Posted by: rbairos
|
August 20, 2010 5:51 PM
"create an eye pupil?
you mean a lens?
oh, and a lens is SURELY the same thing as a brain.
keep clapping."
What the heck are you babbling on about ?
Posted by: rbairos
|
August 20, 2010 5:53 PM
"No, that's not a correct analogy. More like: you think that by just knowing what letters were used (perhaps even what numbers of each letter), will make it possible to recreate the works, ignoring the use of paper, the subjects of the works, and even the languages. "
Uh. no I dont.
How many times does this have to be explicitly repeated?
Posted by: Nerd of Redhead, OM
|
August 20, 2010 5:53 PM
Good question, what the heck are you babbling about except for licking your masters boots by trying to defend his (and your) ignorance of scientific matters?Posted by: rbairos
|
August 20, 2010 5:55 PM
The insults are flying fast and furious!
When should I expect a brick through the window?
Posted by: Kristjan Wager
|
August 20, 2010 5:56 PM
I should perhaps mention that in knowledge management there are four levels of knowledge:
1) known knowns
2) unknown knowns
3) known unknowns
4) unknown unknowns
The objections people have mentioned here are generally within the sphere of known unknowns. We know that we don't know this. But as KG pointed out several hundred comments ago, it's very likely that there are still unknown unknowns regarding our brains. We first have to uncover these (or be pretty damn sure that there aren't any), before we can even begin to have an idea of what we need to re-create a brain.
Posted by: First Approximation, L'esprit de l'escalier
|
August 20, 2010 5:56 PM
Yes, looking back, I was a bit vague there.
Let me rephrase: progress in information processing is necessary, but not sufficient, to understanding/re-creating human intelligence.
Posted by: carnatus
|
August 20, 2010 5:57 PM
While I agree with your assessment of Mr Kurzweil, you're not doing yourself a favor.
Computers we're dealing with today are still Von Neumann machines, data and code are one and the same thing exactly - of which you're certainly well aware.
Classes of systems where a simple set of rules produces behaviors of enormous complexity are legion, from cellular automata on down the line - of which you're also likely well aware.
And I'm sure I don't need to even state that there's no need to understand the process of brain formation and growth in order to effectively simulate its behavior.
To sum it up, Kurzweil may be a kook alright, but your unnecessarily strident critique calls your own credibility in question.
Posted by: Stephen Wells
|
August 20, 2010 5:58 PM
Oh look, he's being persecuted! Just like Galileo! He must be right after all.
Posted by: rbairos
|
August 20, 2010 5:59 PM
"Let me rephrase: progress in information processing is necessary, but not sufficient, to understanding/re-creating human intelligence."
I agree.
Posted by: stevieinthecity#9dac9
|
August 20, 2010 6:00 PM
And another. Too strident? Seriously?
Concern troll.
Posted by: rbairos
|
August 20, 2010 6:02 PM
"Oh look, he's being persecuted! Just like Galileo! He must be right after all."
Who's being persecuted?
I could just do without the childish insults if its all the same..
Posted by: Kristjan Wager
|
August 20, 2010 6:02 PM
Unless you are repugnating Kurzweil's ignorant statement regarding the connection between the genome and the brain, then this is similar to what you're defending.
Posted by: Ichthyic
|
August 20, 2010 6:02 PM
When should I expect a brick through the window?
*yawn*
done clapping yet?
What the heck are you babbling on about ?
you equated a pupil with a lens. you don't *make* pupils, you can, of course artificially produce a lens for an eye. Even then, you think this is all about technology only? do you understand what the differences are between say, a hand lens and the lens in a human eye?
your ignorance of even basic biology will inevitably mean you will make even worse mistakes (and you have).
...like thinking your example of an eye "pupil" is somehow qualitatively similar to what would be required to reproduce an actual brain.
you clowns just like to imagine you know how this would work, but you don't.
it's quite pathetic, given how otherwise intelligent many of you appear to be.
Posted by: KG
|
August 20, 2010 6:06 PM
rbairosI'll treasure this one! Someone who doesn't know the basic anatomy of the eye instructing us all in cognitive neuroscience!
- Stephen WellsEven simpler than that! Wolfram's rule 110 has been proved computationally universal by Matthew Cook. I also presented work at the 1998 Santa Fe Institute workshop where Cook first presented the proof.
[/Kwok]
Wolfram is actually similar to Kurzweil in some ways: a bright guy with an enormous bee in his bonnet, completely immune to evidence and argument.
Posted by: rbairos
|
August 20, 2010 6:07 PM
"...like thinking your example of an eye "pupil" is somehow qualitatively similar to what would be required to reproduce an actual brain"
Man alive, pupil has more than one meaning.
Please forgive me for clarifying with the term "eye pupil".
Can you move on from my phrasing yet?
Secondly, understanding how a pupil works is *qualitatively* similar to how other organs work, in terms of it requiring applied technology and observation, unless the brain employs some non-deterministic mechanism.
Did you perhaps mean "quantitatively" ?
Posted by: rbairos
|
August 20, 2010 6:11 PM
"I'll treasure this one! Someone who doesn't know the basic anatomy of the eye instructing us all in cognitive neuroscience!"
What exactly did I see that was incorrect about the anatomy of the eye?
Posted by: Ichthyic
|
August 20, 2010 6:15 PM
Man alive, pupil has more than one meaning.
like "theory"?
phht.
biologically, no, it really doesn't.
Pupil
lens
done clapping for your own ignorance yet?
meh, probably not.
Did you perhaps mean "quantitatively" ?
No, I meant qualitatively. that you even think I meant that there is only a quantitative difference between being able to artificially construct an eye lens vs. a brain is exactly reflective of the very thing we are trying to pound into that thick skull of yours.
I can see we are wasting our time.
Posted by: rbairos
|
August 20, 2010 6:18 PM
Ichthyic, a pupil is a student, or a component of the eye. Get over it.
Secondly, the original example question was to clarify what was meant by 'technology' as it wasn't clear.
Your tangents are getting especially strained if not ridiculous.
Posted by: Ichthyic
|
August 20, 2010 6:22 PM
Ichthyic, a pupil is a student, or a component of the eye. Get over it.
wtf?
man, I really was wasting my time with you.
Posted by: Nerd of Redhead, OM
|
August 20, 2010 6:22 PM
Yawn, apologist is no scientist. His ignorance of basics is obvious to those of us who practice in the field, and his pretense of knowing something is totally laughable.
Sometimes 'tis better to keep your mouth shut and have folks think you are stoopid, than to open your mouth and prove them right. That is you apologist. Your smartest recourse is to quit. You don't have what is needed to convince us of anything other than your stoopidity. But you continue anyway. I'll check back when I need a good belly laugh. Which is all you are good for.
Posted by: stevieinthecity#9dac9
|
August 20, 2010 6:23 PM
It's ok. You can admit that you're having trouble keeping up.
Posted by: rbairos
|
August 20, 2010 6:32 PM
Okay then..well smart ass quotes, childish taunts, strawmen, obsession over my phrasing, mis-attributed positions aside, its clear dialog on Kurzweil's response will be difficult.
Please go back to patting yourselves on the back.
Thats not to say some of you weren't raising interesting objections that I'd probably agree with.
Do I think the amount of information in the genome misrepresents the amount of information needed to encode an adaptive brain like system?
Probably.
Posted by: rbairos
|
August 20, 2010 6:34 PM
Ichthyic, Im sorry I didnt roll over and beg for forgiveness by using the verbose term "eye pupil".
By your several posts on the topic, its clear you really do know how to have an insightful and relevant discussion on the topic at hand.
Posted by: John Morales
|
August 20, 2010 6:37 PM
I read Kurzweil's response, and this stood out:
Searching brings up articles about it from the 1980's. So... why are there no prosthetic cerebella in regular use by now? Interface problems?
Must be 10-20 years away. :)
Posted by: KG
|
August 20, 2010 6:39 PM
Posted by: KG
|
August 20, 2010 6:42 PM
rbairosThe pupil is not an organ, dolt. It's the hole (covered by a transparent membrane) which admits light to the eye.
Posted by: rbairos
|
August 20, 2010 6:43 PM
KG seriously? Can you let it go? This is not debate. Its childish pedantry. (I was going to type 'pendantics' but thought I'd look up the word and save you and Ichthyic 30 posts..)
Posted by: KG
|
August 20, 2010 6:45 PM
rbairos,
You are demonstrating repeatedly that trying to have a worthwhile debate with you would be like arguing philosophy with an armchair.
Posted by: rbairos
|
August 20, 2010 6:47 PM
"The pupil is not an organ, dolt. It's the hole (covered by a transparent membrane) which admits light to the eye."
Fine, the mechanism used to regulate the amount of light entering an area including its protective membrane. Really, move on.
Posted by: rbairos
|
August 20, 2010 6:50 PM
"Searching brings up articles about it from the 1980's. So... why are there no prosthetic cerebella in regular use by now? Interface problems?"
Likely, amongst other technical problems.
Whats your best estimation on prosthetic cerebella?
Posted by: John Morales
|
August 20, 2010 6:53 PM
rbairos:
1. You think this is a debate? :)
2. Punctiliousness has its place.
Posted by: rbairos
|
August 20, 2010 6:56 PM
"1. You think this is a debate? :)"
:)
"2. Punctiliousness has its place"
Good word. I had to look it up, like I did
the correct form of the word 'pedantics'
Posted by: KG
|
August 20, 2010 7:19 PM
After reviewing this debate, I've changed my mind: I think complete simulation of the Singulatarian brain is likely to be a fairly trivial problem. After all, reasoning ability and world knowledge are clearly not going to be needed. Indeed, maybe it's been solved already and we're just arguing with a set of bots. Certainly, rbairos's continual misuse of fairly simply words suggest it.
Posted by: John Morales
|
August 20, 2010 7:26 PM
rbairos, well, if nothing else you're expanding your lexicon.
Now, if you care to apply your new-found knowledge, you might wish to examine the claim that the cerebellum ("little brain") has been "modeled, simulated and tested", and then consider what said simulations of models actually accomplish.
Posted by: BrianX
|
August 20, 2010 7:50 PM
To sum it up, Kurzweil may be a kook alright, but your unnecessarily strident critique calls your own credibility in question.
Enter the concern troll.
Posted by: Greylander
|
August 20, 2010 8:01 PM
#642 Steve LaBonne
Steve, I have clearly stated things that contradict this.
But this does bring up an issue that may indicat part of the misunderstanding between the two sides of this argument may. Perhaps you think of those synaptic connection as "hardware"... and as such you would tend to equate the physical synaptic connections with "functional description" (to use my phrase) and you may tend to not to think of those connections as "data" or "software".
When we are talking about the "50 million bytes" of code we are talking about coding the equations that tell us things like how those synapses grow and change, how they increase/decrease the concentration of various receptors, how much of any given neurotransmitter they release in response to varying levels of other neurotransmitters, and so forth. You may still not accept that those equations can be encoded with a "mere" 50 million byte... but it is important that you understand what is actually claimed.
In any simulation of the brain, the growth and change of synapses would be *data* upon which the synapses operate. To simulate changes in the synapes, you change the data according to the *equations* that govern the system. For my own estimate, I consider more that the 10^14 synapses that exist, but I also consider that because new ones appear old ones disappear, we must will end up having to consider something like 10^15 "potential synapses". Each of those potential synapses could need a great deal of data to describe it's state (or, rather, the relevant aspects thereof) -- maybe 1KB, maybe 10KB, maybe more. Maybe less. This puts my estimate of the total size of the simulation at around 10^18 bytes.
What I contend is encoded in the genome is not the wiring (or at most very very little of it), but the "rules" or "equations" that govern how that wiring is created/adjusted over time in response to environmental stimulus.
You may still contend that the problem is far more complex than I give credit... but at least bring your objection on point.
Insisting that I and others in my camp do not understand that there are 10^14 synapses who's wiring is not hard-coded in the genome only makes those in your camp look stupid when we make ourselves very clear and you still go round and round on this non-issue.
Posted by: KG
|
August 20, 2010 8:05 PM
Kurzweil's reply is just more of the same drivel. No, the length of the genome does not tell you anything useful about the amount of information needed to specify the brain "before it interacts with its environment". First, because there isn't such a time. Second, because the genome produces a brain only because it is "interpreted" by a particular physical environment. You could use a different "interpreter" and get just a string of letters (CGTTATCG...) or musical tones, or moves on a 2-D grid to north, south, east and west, or whatever. That the genome produces a functioning body depends (among many other things) on the way transcription and translation occur, on the way strings of amino acids fold, on the way various chemicals diffuse through the growing embryo, and those depend on the precise values of physical constants, and on the chemical and structural environment in which these processes take place. The genome can "assume" that these factors are going to be what it "expects" - it gets this information "for free", but that doesn't mean it is not as essential to the formation of the embryo brain as the information in the genome. Unless you can take the genome and deduce what the brain is going to be like without looking at the physics and chemistry involved in development, the claim that the design of the brain is encoded in the genome is clearly bullshit.
Posted by: John Morales
|
August 20, 2010 8:05 PM
Greylander, any comment on Kurzweil's claim about the cerebellum?
Posted by: Nerd of Redhead, OM
|
August 20, 2010 8:10 PM
No, you look silly calling it a non-issue. Just like like all your other posts. Silly and trivializing the important science. End of story.Posted by: John Morales
|
August 20, 2010 8:22 PM
Nerd, Greylander never did respond to my #527, where I asked him about computational chemistry.
Greylander @214¹:
--
¹ Is this a quote-mine? :)
Posted by: Nerd of Redhead, OM
|
August 20, 2010 8:29 PM
Whack away at that toad if you want John. I'll gladly step aside. Just getting a little tired of non-scientists trying to trivialize the scientific problems to get to that K-fuckwits idiotic 10 years prediction...
Posted by: Greylander
|
August 20, 2010 8:47 PM
#618 stevieinthecity#9dac9
Bullshit. PZ may not have deliberately strawmanned Ray but he did nonetheless. PZ repeatedly claims the Ray was claiming we would reverse engineer the brain from looking at the genome. It is the main thrust of PZ post and the principlel thing PZ ridicules. Ray never made that claim. Not even close. Statements about information content are not statements about how that information is to be interpreted.
Even the gizmode article did not mis-represent Ray as having made the "get it from the genome" claim. PZ owns entirely the failure to comprehend the information theoretic argument.
This is just an incredibly stupid thing to say. The microscope was just a tool. Are you saying the microscopes have nothing to do with all our present understanding of biological processes at and below the cellular scale? Fundamentally, science is done by gathering data, finding patterns in the data, confirming the existence of those patterns by gathering more data, and so on. Those mere tools will let us gather more data or better quality data or both. Ergo science will be done faster. Oh, and computers are excellent assistants for spotting the patterns... and getting better all the time at spotting patterns without assistance.
So higher resolution imaging (as just a single example) has a great deal to do with increasing the rate of improvement in understanding brain function... and anything else that can be imaged.
Posted by: John Morales
|
August 20, 2010 8:57 PM
Greylander:
You may not have deliberately strawmanned PZ, but you did nonetheless.
Look at the red bit in the OP.
If the design of the
brainhouse is in thegenomeblueprint, then from thegenomeblueprint we should be able to build abrainhouse.Look at PZ's contention: We cannot derive the brain from the protein sequences underlying it.
Posted by: Nerd of Redhead, OM
|
August 20, 2010 9:12 PM
Greylander, you keep misrepresenting what was said. PZ is right in that reverse engineering from the genome to the finished brain is total and utter bullshit. Attempting to say or represent PZ as saying anything else is a lie. Period. Which makes you a guess what? There simply isn't enough science available, nor will there be, for K-fuckwits prediction to come true. Until you face that fact, you just seem like a delusional fool, same as any godbot with a presupposition for the existence of his imaginary deity.
Posted by: chaingangcharlie
|
August 20, 2010 10:20 PM
I think being a Deity must be like .... waking up one morning & realising that the only way to create consciousness 'ab initio' would be to create something similar to & not much less complex or time consuming than a whole new universe.
Thus, the answer to the question 'why is the universe so big & so complex & so old ?' might be 'just can't synthesise thought without all that crazy crap '
Wake me up when a computer independently reaches the same conclusion.
Posted by: Greylander
|
August 20, 2010 10:50 PM
#717 John Morales:
Patience, grasshopper. Only so much time... and I do have to sleep once in a while. :)
Although improvements in computational chemistry -- simulations of molecular interactions -- will certainly help, it is not strictly necessary to what you quoted. Automation and mass deployment of automated micro-scale lab experiments will be more important. Set aside a moment whether you think what I describe will be feasible in the next 2 decades, and just ask "what if...". What if you (or neurobiologists) could set up thousands of simultaneous experiemental trials on individual or small groups of neurons, in wherein each trial can have numerous parameters tweaked, exposing the neuron (or individual synapses thereof, even) to differing quantities of this or that neurotransmitter or other chemical, measuring responses. They can do thousands of these experiments in a day. Then by entering new instructions, then can tweak the experiements and do a thousand more. How much faster will they be able to determine relationships between things like concentration of neurotransmitter X with increases or decreases of receptor Y?
As to improvements in computational chemisty, two decades from now, we will have in the neighborhood of two million times cost-effective computign power. We already know the underlying physical laws. So whatever we can simulate now, we will then be able to do simulations involving roughly a million times more particles (atoms if the sim is done on the level of individual atoms). Or we can do the same simulations as today, but a million times faster. Or we can vastly improve the acuracy. Basically whatever tradeoff you want to make between speed, size of simulation (number of atoms), and accuracy, we'll have much much more of all three to play with.
Consider what this will mean for protein folding:
"Recent tests show that the ZA mechanism speeds up conformational searching sufficiently that physics-only models can now find approximately correct folds for chain lengths up to around
100 monomers"*http://www.mpikg-golm.mpg.de/th/people/weikl/pdfs/cosb07.pdf [from 2007]
Now imagine when we can solve for 100,0000 monomers, 1000 times faster than we can do 100 monomers today. Actually I'm not familiar in detail with the algorithms and the time-complexity may not scale linearly, but for good reasons I can be pretty sure they do not scale worse than N2, so even if in 20 years we can just do say, 10,000 monomers ten times as fast as 100 monomers today, that'll pretty damn useful. There are also good reason to expect better than N^2 scaling of the algorithms.
So yeah, computational chemistry and other molecular level simulations will be extremely important tools moving forward into the near future. When will it (for all practical purposes) replace doing physical chemistry experiments? I can't say. But obviously the more we can do in simulation, the less we will need to do physically -- or it might be like be a bit like the "paperless office" although today something like 99% of all information is stored only on computers, the remaining 1% on paper dwarfs the 100% of 50 years ago, mainly because computers enable us to generate so much more information. Likewise... in a few decade 99% of chemisty may be done in simulation, but the remaining 1% of experiments (in part to continually spot-check the simulations) done physically may outnumber all of today's experiments. Especially because many current types of experiments will be automated on a micro-scale.
I think you also question why we haven't made more progress in comp. chemisty up to now... it is still a question of computing power. And computational chemisty is non-trivial in terms of sheer number crunching required.
Not a quotemine, but it is mine quote.Posted by: Greylander
|
August 20, 2010 11:36 PM
@ #721 Nerd of Redhead:
I might almost stand slightly corrected. PZ never quite explicitly states in his OP that Kurzweil claims that he would reverse engineer the brain directly from the genome. But then why does PZ go on and on about how insanely wrong this is unless he thinks it is what Kurzweil claimed? And then why did PZ say the following later in the comments:
PZ obviously does believe and implies that Kurzweil claimed we would reverse engineer the brain by looking at the genome.
Yes another example of a non-argument. "You guys are all wrong. Everyone who disagrees with me is a delusional fool."
Look, if I am soooo obviously wrong, it should be easy for you to select a substantive point I made and make a direct challenge. I can find the meaningful weaknesses in my own arguments better than anyone here has come even close to attempting.
I've noticed that careful detailed arguments are mostly met by the PZbots with very general contentless objections. And summarized arguments are often pounced on for overlooking some detail as if the poster is a moron for not elaborating at length upon that detail, or must be ignorant of it's significance. Not everyone does this. But far too many.
Posted by: John Morales
|
August 20, 2010 11:42 PM
Greylander, interesting response.
I hope you're aware of the difference between monomers and polymers.
I also wonder for what value of 'approximately' is "find[ing] approximately correct folds" equivalent to "solving".
Unlike you, I'm not in the habit of prognosticating from insufficient information.
--
I will note, in passing, that a few tens of micrograms of LSD will have significant effects on perception and cognition in the human brain.
This suggests to me that approximations are going to have to be, well, not too approximate.
Posted by: John Morales
|
August 20, 2010 11:52 PM
Greylander:
In your opinion.
I think PZ believes (as I do) that Kurzweil brought up the genome to indicate that he considers that the information required to make a brain is encoded within it; that is the point of contention and the basis for this OP's title.
PZ, a developmental evolutionary biologist, disagrees with that proposition.
Inasmuch as I'm not either's class in their respective fields, I have to rely on their respective authority. In this case, the computer person is opining on the biologist's field, whilst you seem to think that it's the other way around.
Accordingly, we each have a different perspective.
Posted by: chasrmartin.myopenid.com
|
August 20, 2010 11:55 PM
PZ, you should really be embarrassed about this.
(1) You take issue with the notion that the design of the brain is encoded in the genome. Fine. Care to propose where the design of the brain comes from, then? The morphological field?
I've got my own issues with some of this, but the fact is there's nothing but the genome in which the design of the brain can be encoded -- oh, you could argue that epigenetic information isn't part of "the genome" but that's a quibble -- even if you double or quadruple the amount of information to account for epigenetics, mitochondrial DNA, and so on, that only adds 2-4 years to Kurtzweil's schedule.
Once you make that observation, then Kurtzweil's argument is a simple information-theoretic one: if there are only six gigabits of information contained in the genome, then it is possible to simulate the entire behavior of the genome with a program that is no larger than 6 gigabits plus a constant (albeit potentially a large constant.)
This is a theorem, a mathematical truth, proven by Greg Chaitin in 1975 and presented in "A theory of program size formally identical to information theory." I certainly understand why a biologist might not have run into it, but frankly your counter-argument primarily exposes your lack of mathematical knowledge and sophistication.
if we consider the human genome, then that simulation of the full behavior must necessarily, by definition, include whatever it is the brain does. If you completely simulate the development of the organism from zygote to adult, including providing appropriate simulated stimuli, feedback, and learning, you will necessarily end up with a simulation that simulates what it's like to be "a person".
Again, this is necessarily true, at least unless you want to posit something like, on one hand, a "soul" that is extraphysical, or on the other hand, something like Penrose's quantum nondeterminacy as inherently necessary for "consciousness."
So far, both approaches have proven curiously difficult to falsify.
(2) You then assert that you don't think Kurtzweil knows how the brain works. This instantly causes me to ask: and you do? Would you mind explaining it all to me? Certainly when I was doing my PhD 20 years ago at Duke Medical School, I didn't know anyone who thought they understood how the brain really worked -- some very small structures, certainly. Larger ones, not so much. Hell, we don't even have a satisfactory definition of some of the larger-scale behaviors of the brain, like "consciousness".
Failing that understanding, then, Kurtzweil's position is simply that of stating a hypothesis. You may not believe it -- I don't find it very plausible myself, in part because I don't think we can even state what it means to "work just like the brain" -- but disbelief is not disproof, and frankly derision is not argument.
Posted by: John Morales
|
August 21, 2010 12:06 AM
chasrmartin, you haven't read the comment thread, have you?
You make claims that have been previously made and responded to.
Posted by: John Morales
|
August 21, 2010 12:11 AM
chasrmartin (condensed version):
Apparently, unlike chasrmartin, PZ put forth no argument, though, like chasrmartin, he did engage in derision. Thus, chasrmartin is entitled to chide PZ in this manner.
Heh.
Posted by: Nerd of Redhead, OM
|
August 21, 2010 12:19 AM
chasrmartin There is a lot of developmental biology between the genome and the final brain. Most of that development is unknown at this time. You appear to be devoid of that knowledge. And give the fluidity of the human brain, and the changes with learning and experience, that trumps the genome on the final product. You are wrong.
Posted by: Ichthyic
|
August 21, 2010 12:24 AM
Care to propose where the design of the brain comes from, then? The morphological field?
did you read past the red marked line?
I'm guessing not.
cya.
Posted by: BrianX
|
August 21, 2010 1:33 AM
You then assert that you don't think Kurtzweil knows how the brain works. This instantly causes me to ask: and you do?
/me points to the addendum to the OP and notes that brainal stuff is what PZ does for a living.
Posted by: Greylander
|
August 21, 2010 1:45 AM
@ #721 Nerd of Redhead:
BTW, thanks for causing me to re-read PZ's OP. There is more stupid in there that needs walloping.
PZ:
This is utterly stupid. Programs are data. Of course, wh ile all programs are data, not all batches of data are programs, but PZ even mentions "conditional logic" and "reglatory parts of the genome" -- in other words: instuctions. The genome is a program, or a mix of program and non-programmatic data.
But the following is even more stupid:
Ontogeny -- the development or course of development especially of an individual organism
Development is something that organisms do. This is a bit like saying the text and images in your browser window are the browser program, and the executable binaries are the data.
There is no perfect analogy, but a far more correct one is this: The genome is the program. The machinery in the cytoplasm is the hardware. Effects of the external environment are input. The hardware follows the program to respond to the environment in various ways, including methylation of the DNA, and also to make copies of both itself and it's software (the genome) in the current state. Then you have to units of hardware each executing the same program, but experiencing slightly different environments, so they do not necessarily do the same thing, and their internal states (such as methylation) may diverge. Because each daughter cell is part of the environment of the other, the behavior of each affects the input to the other -- such as releasing signaling molecules like sonic hedgehog.
In point of fact concentrations and gradients of of concentrations of signaling molecules are important inputs for the cells to "know" where it is in relation to other cells. Based on these inputs, and the instructions in the genome (that conditional logic PZ mentioned), along with the current state of the genome (methylation and other epigenetic effects) and the interior state of the cell, additional epigenetic effects are applied to the chromosomes, allowing the cells to differentiate.
By the time you have a complete body, you have billions of cells (their machinery is the hardware running the program) all running identical programs, but executing different parts of those programs on different data. The epigenetic factors can be thought of as a kind of memory to keep track of where each cell is in the course of running the program.
THAT is an appropriate genome-program analogy. Calling the ontogeny a program is utterly laughable.
With crap analogies like that, PZ demonstrates that he is as out of his depth on matters computational as he claims Kurzweil (and presumably myself) are out of our depth in matters biological.
Furthermore. Most organisms do a significant amount of important development inside the near-isolation of egg or uterus, which means there is almost no appreciable information coming in from the environment. Almost all input to each cell comes only from the signaling of neighboring cells. And where did each of those cells get the instructions on what to signal and how to react to incoming signals? From the genome. Those signals to not add any new information to the system that was not there in the fertilized egg to start with. The developing embryo may get some information from the mother, probably mostly to do with immune system function, but considering it is even possible to grow an embryo to term in the uterus of another species, it seems unlikely that the input from the mother has much effect on the vast majority of structural and functional features of the end product.
Finally, once the nervous system begins to develop in the fetus, it does begins sensing it its environment (i.e. babies begin to familiarize with the sounds of their native language while still in the uterus). This means synapses that grow/change/fade in response to environment. As has been made clear quite often, these effects are considered part of the "data" upon which the underlying processes operate. Those underlying processes were coded for in the genome (albeit in a very indirect way, that relies on all those cell:cell communications of which PZ is so fond).
Ergo, the vast bulk of information describing how the brain works as a learning/thinking/deciding/adapting machine is in the genome.
Got it?
No, I'm sure you don't.
----
P.S. I swear the first person who replies saying something utterly stupid like "the genomes is not a series of binary codes executed sequentially", I will reach right through the intertubes and slap you completely silly. PZ just called 'ontology' a program (ha!)... remember that is how this epic post got started.
----
BTW, for any who wonder, I actually have a great deal of respect for PZ -- his knowledge, intelligence, writing skill, and public relations pinache are all to be commended (not, I am sure, that he cares whether I commend him or not). I'm just following his lead and calling out the stupid where I find it, opening it up to ridicule. It might not do much good here on his home turf, amongst all the PZbots, but... meh.
Posted by: jdennings
|
August 21, 2010 1:53 AM
What is it with some (past?) middle-aged academics that prompts them to engage in derisive diatribes worthy of adolescent pissing contexts?
Somewhere, somehow, Kurtzweil hit a nerve that Myers equated to some statement about his dick size. And the best Myers could do was hit ad hominem.
Pathetic. Myers didn't even bother to read/hear original claims. Talk about lousy scientific debate ...
Regardless of who is right/wrong, one thing stands out:
Class and elegance, Mr. Myers, are definitely not within your reach.
Posted by: Greylander
|
August 21, 2010 2:27 AM
#726 John Morales,
Oh please, PZ is clearly trying to make Kurzweil look foolish for supposed trying to reverse engineer the brain from the genome (or believing it will be done in the next 20 years).
The point you mention (below) is also made by PZ, but in support of the main thesis.
Most of the information *is* in the genome, but this in part depends on what is meant by "make a brain". For example, if you think of the way in which the brain wires and rewires itself as part of the process of learning and adapting as "making a brain" (i.e. it makes and remakes itself) then you will object, correctly by your definition of terms, that the design of the brain is in the genome. However that is not what I nor Kurzweil claim. We claim that the underlying rules according to which the brain wires and rewires itself are coded in the genome. You can still argue with that, but it becomes a very different ballgame.
I think PZ will be first to point out argument from authority is not a strong argument. BUT we all must rely on experts at some point. I joined this fray precisely to add some expertize and elaboration on the Kurzweil side of the debate. I am chagrined at myself for getting caught up in some puerile exchanges, but I am trying to get into the whole spirit of ridiculing that which is foolish and stupid. PZ gets no special immunity.
Since you bring up expertize/authority, I want you to consider the possibility that I have considerable expertize on both side of this divide. I am not making that claim, but I want you to consider the possibility, and ask yourself how it would color your opinion of the proceedings.
Posted by: BrianX
|
August 21, 2010 3:06 AM
Since you bring up expertize/authority, I want you to consider the possibility that I have considerable expertize on both side of this divide. I am not making that claim, but I want you to consider the possibility, and ask yourself how it would color your opinion of the proceedings.
I can't speak for John, but since I don't know who you are beyond your handle, I'd take it about as seriously as I would someone making the same assertion on Wikipedia -- not at all. With PZ I can at least look up his papers and the responses to them if I wish. But who the hell are you?
Posted by: Greylander
|
August 21, 2010 3:29 AM
@ #725 John Morales:
http://en.wikipedia.org/wiki/Monomer
When they mention 100 monomers they are talking about simulating the folding of a polymer (protein molecule) that is composed of 100 monomers. Basically it is a 100-particle simulation of the motion of the monomers according to the electromagnetic interaction forces between each pair (100^2 interactions to account for) plus some simplified model to account for forces due to surrounding water molecules. It is the N^2 interactions that may require such a simulation to scale proportionally to N^2 in terms of computational complexity (cpu time or FLOPS required). However since all but the nearest neighbors (in physical proximity, not necessarily sequence) will have negligible interaction, there are tricks such that the complexity will remain closer to proportional to N... hence as computers get faster the size of problem that can be solved will increase proportionately.
I don't know myself.
Posted by: Greylander
|
August 21, 2010 3:57 AM
@#736 BrianX,
Ah, Brian... haven't you figured it out? I'm your destiny and you are my future wife. We were made for each other.
For what it is worth, I didn't ask "What if I told you...". I asked to consider the possibility. I should have added "consider the possibility that you *know* (i.e. can verify, as with PZ) that I had solid credentials on both sides (neurobiology and computational science)", how would that color your view on the whole exchange?
And I brought it up because John mentioned relying on the expert.
Now, if you had to equally qualified (by your own standards) experts who disagree, then how do you choose between them? You still have to look at their arguments and decide for yourself who seems to make the most sense. But that is what you should always do regardless of supposed expertize on either side, unless you just don't have time to read their arguments. But anyone slogging through this thread obviously does have the time...
Put another way, what if instead of some anonymous bloke posting in the comments here, I was another respected scienceblogger, who happened to be a neurobiologist or maybe both a neurobiologist and a computer scientist, posting rebuttals on my own blog?
And just to be clear, I am not trying to hint or even "make you wonder" about my credentials. That's not the point. Make whatever assumptions you wish.
Posted by: Stephen Wells
|
August 21, 2010 4:13 AM
Greylander, producing this: "Most organisms do a significant amount of important development inside the near-isolation of egg or uterus, which means there is almost no appreciable information coming in from the environment." puts you in the same category of deeply shameful biological ignorance as Kurzweil with his "brain before it iteacts with the environment."
The uterus IS a complex environment! Hell, the EGG is a complex environment. You need not only the information in the genome (and I've pointed out that Kurzweil mistakes the information in a text string specifying the sequece of bases for "the information in the genome", which is a bad mistake), you also need a complete description of transcription, translation, protein function...
Do you see the problem yet?
Posted by: KG
|
August 21, 2010 4:36 AM
Posted by: KG
|
August 21, 2010 4:39 AM
Ugh - sorry about that: here it is in more reader-friendly form.
- GreylanderThis really is quite amazingly stupid. Of course there is information coming in from the environment. Let's take a bird's egg, since this looks much better for your case than a human embryo, which depends absolutely on a complex chemical flow in and out. Under almost all physically realisable conditions, the egg will not develop; it will die. Wrong temperature, wrong pressure, wrong atmosphere, too much ionising radiation, presence within the egg of tiny amounts of any one of untold number of poisons, a small hole in the shell. Any of these, at any time in the egg's development, and the egg will die. How much information is required to specify the range of conditions under which the egg will hatch, when those conditions have to be right every millisecond? The suggestion that such a quantity is even calculable is ridiculous. The basic problem here is that information theory, which is extremely useful where its application makes sense, simply cannot tell us anything useful here. The idea that the information in a string sets a limit on the complexity of its possible effects is complete nonsense. The complexity of the "interpreter" has to be considered, and here, the "interpreter" is the whole of the physical world within the relevant light-cone, because anything happening within that light-cone could halt or alter the course of development. So Kurzweil's invocation of information theory simply makes no sense at all, unless it is interpreted as PZ has done - he was being charitable if anything. As I noted in an earlier post, if "the design of the brain is in the genome", then you should be able to deduce the structure and functioning of the brain simply from the string of uninterpreted symbols CGATTCGT... Good luck with that.
Posted by: albatros183
|
August 21, 2010 5:06 AM
GAAAAAAA...
The genome is the instructions
The other various chemical makeup is the data set
Inst+Data = some large amount of resulting data that might or might not be like a human brain
You will also need a bunch of code to interpret the data that might or might not be one(1) million lines who gives a Flying $%$^ if it's 1 million or 100 million
please give us all a break and blow up the sun already so we can live out our ~billion years or not
This cyborg nonsense is just that
FacePalm
Posted by: Brain 2045
|
August 21, 2010 5:18 AM
.
Henry Markram says that it's possible to simulate the full human brain on a computer in 10 years:
.
----> ----> ---->
Check this 2 Astonishing lectures of Henry Markram:
(watch them in full screen)
1.
http://neuroinformatics2008.org/congress-movies/Henry%20Markram.flv/view
2.
http://ditwww.epfl.ch/cgi-perl/EPFLTV/home.pl?page=start_video&lang=2&connected=0&id=365&video_type=10&win_close=0
----> ----> ---->
Very interesting article from the ‘Seed’ magazine:
Part 1:
http://seedmagazine.com/content/article/out_of_the_blue
Part 2:
http://seedmagazine.com/content/article/out_of_the_blue/P2
----> ----> ---->
Few more interesting links about the project:
http://news.bbc.co.uk/2/hi/sci/tech/8012496.stm
http://www.youtube.com/watch?v=Bz5IUaRr8No
http://www.youtube.com/watch?v=RLCT3wU4fek
.
Posted by: Greylander
|
August 21, 2010 6:38 AM
@ #740 KG:
Wow... I just knew someone was going to come back with a lame counter along these lines. The stupid just keep on a'coming from you PZbots, eh?
Are you seriously contending that the *absence* of life-threatening conditions adds anything meaningful to the fundamental processes governing how the organism functions?
I am quite aware that the fact that a hundred ton boulder will squish you to a pulp if it falls on you is not coded into your genome. Although apparently the equivalent level of brain function *was* coded into your genome.
"My mother did not take LSD" when I was in the womb did not "add information" to my brain structure or function. If anything it avoided destroying information.
I have accounted for the cellular machinery and the environment and the laws of physics in everything that I have said.
The whole world within the light cone? Are you freaking kidding me? You seem to have an almost religious faith in the infinite complexity of living organisms. This is just absurd. In a trivial sense, everything in the past light cone of a given event is tanken together is a "cause" for that event. But does that event (or a small local region of events) contain the equivalent information of the entire retro light cone? No. Information is neither created nor destroyed and most of the information in the past light cone has gone to concurrent (space-like separated) events. As a practical matter, most events in the past light cone can be disregarded as contributing negligible information to the event(s) in question. You need to show some link between your source of information and where you claim it ends up.
Are you going to invoke quantum entanglements next? Jumpin' Jesus on a Pogo Stick!! Where do you come up with this crap? I hope you had a serious education in physics (I don't care if formal or self-taught, but grad-level required) if you are daring to bring light cones into this... but I'm sorry to say, if you did have such an education, it was deficient.
(I have to make an aside here... I'm perfectly capable of holding this conversation without all the sneering and "how stupid stupid you all are" kind of language... but PZ sets the tone, and y'all dish it out so well... I'm just trying to fit in... it is kinda fun. Now back to your regularly scheduled...)
"Absense of shit that kills you or fucks you up" does not contribute, in any worthwhile sense, information to you other than "the environment didn't kill me or fuck me up, so, like, cool, I'm not dead or fucked up! Yay me!"
Thought experiement: Take a million cloned pregnant rats, each impregnated with a litter of their own clones (suppose we've engineered a parthenogenetic breed line), and spread'em round to every remotely viable environment you can think of -- a *huge* variety of environments external to the expectant mothers. A few weeks later you will have ten million baby rats, all pretty much the same -- healthy normal baby rats. In some places, some conditions, like shortage of food, overcrowding, and so forth will filter in and have some effects on the development -- I did say "near isolation" not "perfect isolation". But even for those rat babies, you've got heart, lungs, kidneys, skin, fur... all of it still working pretty much the same way as if they had been gestated under healthier conditions. Maybe they are born a little underweight. Even their brains work pretty the same -- still made up of a network of neurons which still encode learned information by growing/changing/pruning synapses. The mother's stress/hunger might result in personality alterations in the rats, maybe more aggressive or fearful -- but these kinds of adjustments are responses programmed for in the genome (PZ's "conditional logic" see the OP). Their brains at birth are wired a bit differently, maybe there's more of some type of cell that release more of some type of hormone which maybe causes the rat to be more fearful (world is more dangerous, be more cautious) or more aggressive (food is scarce, fight for every morsel). But the equations that describe the time evolution of the neurons and synapses and so forth are going to be the same, you just have a somewhat different "initial" conditions. For example: a rule for one type of neuron that might say something like "in the presence of (and moreso with greater concentration of) hormone X, reach a dendrite off in the direction of any nearby neuron that always seems to fire around the same time I [me being a neuron] do" -- that kind of rule is the same in the all the rats, but maybe the famine or overcrowding babies have more hormone X than normal. What the rules actually are is not germane to the discussion. That these rules are invariant from rat to rat is what is relevant.
The whole point of an egg or uterus is to keep the developing embryo/infant isolated and protected from all the potentially hazardous variations in external environment. Under normal circumstance, the developing organism get a nice stable, *predictable* environment. Predictability is in a very literal sense the antithesis of information by the way. The fact that is such a wide variety of environments, all the babies come out essential the same (there are endless varieties of wildly different but perfectly healthy environments) tells us that the details of the environment play almost no role in development.
Actually, come to think of it, the mere fact of identical twins and clones thoroughly demolishes your position.
Details of a non-hostile environment contribute nothing to the basic structures and processes of the organism. Environmental information is incorporated at a more plastic level -- neurons/synapses, immune system -- but that plasticity is in effect planned for in the genome.
Hostile environments destroy or scramble the information governing the basic structures/processes. Healthy environments, rather than add information to fundamental structures/processes, simply fail to destroy it.
In other words.
YOU FAIL.
Next?
Posted by: Stephen Wells
|
August 21, 2010 6:45 AM
Greylander, are you simply pretending I don't exist now? The uterus is a biochemically complex environment. The interior of the cell is a biochemically complex environment. The amount of data in a text representation of the coding sequence simply does not measure this.
For pity's sake, the letter "A" does not contain any information about the physical/chemical characteristics of an adenine base. Don't you see this is a basic problem with Kurzweil's argument?
Posted by: Greylander
|
August 21, 2010 7:51 AM
@ #738 Stephen Wells
And the uterus provides a nice cozy amniotic sac, a placental barrier to keep mommy and baby bloodstreams from mixing... the thing is designed... er.. oops. the thing evolved to provide a cozy, safe, stable, predictable environment for baby to develop in. The uterus, functioning optimally is giving the fetus nutrients and space to grow essentially undisturbed. And lets see, is it up to 4-month premies that can survive nowadays with good chance of growing up normal and healthy? with the same fundamental structures and processes as anyone else? I guess if the uterus supplies any special instructions to guide development, it stops after about 5 months. And how about in-vitro fertilization? I guess the uterus does not supply any special instructions for the first few days of development either.
You are running out of places to hide your fundamental-process-altering information. Oh wait. I get it. You are hiding the information in the gaps!
You say my 'god' is the singularity. I say yours is this phantom source of information.
Again, to re-iterate. No one, especially not I, has claimed that information is not incorporated from the environment. If your legs get chopped off -- that is some serious information. My step-daughter has asthma because she was born preemie and she most likely would not have if full term. Those physiological differences that account for the asthma are the incorporation of "information from the environment" (but I would argue it is likely to just as much be the destruction of information by the environment). There may be differences in her personality that we could notice if we could compare to her hypothetical full term twin.
I suppose you have things like methylation and other epigenetics in mind? I covered that. Just for sake of argument I'll concede... but then the burden is on you to show good reason (a) why all or most of the epigenetic effects were not 'programmed' by the genome in the first place; and (b) the there is some order(s) of magnitude increase in total information. But if you suppose that every random twist, turn, wiggle, and kink constitutes pertinent information, then you have to show that all those wiggles and twists actually affect the functioning of the cell somehow.
And I accounted for all this in my first couple posts. Yes there *might* also be information in the cellular machinery of the initial fertilized egg which is not already in its DNA. But how much? The sequences of various proteins(see below also) are in the DNA already. Cells are suishy fluid things with lots of jiggly randomness, the information describing the details of position/orientation/motion of all the molecules is not preserved in development. The tehniques involved with cloning strongly suggest that the egg surround in genome provides very little information that is preserved in development. Where is your phantom information hiding?
This last one is a FAIL. Protein function is determine by its sequence plus laws of physics (a few KB worth of information at best in the laws of physics). Sequence is provided by the genome. No new information hiding in the details of protein function.
Posted by: BG Denis
|
August 21, 2010 8:20 AM
All through these arguments nobody seems to accept that what we know today is and was quite impossible to define and even extrapolate 50 years from now.
Another matter nobody seems to acknowledge is the population working on the problems. If we go back to Einstein's time, when he had the famous argument with Bohr, the population on earth was about 2 billion people, many illiterate, without any high power computer aid to calculate complex problems. Now we have at least 3 times as many people on earth, but universities have bloomed and computer power is exploding (not yet finished). Just give Einstein a modern computer to calculate the equations he was working on and we would be far more advanced. But the possibility that there are tens, maybe hundreds and even thousands of Einsteins living today, having machines to help them comprehend the complex and the possibilities of solving problems. All that compounded by the fact that many of today's scientist work TOGETHER (LHC as an example) and not against each others on solving those problems, plus the fact of ever faster computers analyzing the data and we have set the place for quite a few surprises ahead. Anybody not understanding this compounding effect is simply blind to facts.
By the way, we are all supposedly intelligent human beings, moved by desires. The desire to insult confers to the one insulting the other the complete lack of respect necessary for moving forward in a manageable world.
Posted by: Greylander
|
August 21, 2010 8:36 AM
@ #743 Stephen Wells:
Good grief, patience patience. I don't necessarily answer in numerical order... :)
If you look at it that way, you are always going to see infinite amounts of information everywhere. Hell, the spin orientation of every electron of every shell of every atom in the DNA is another bit of information if you want to go there. The question is what information is relevant. And that is where you say "we don't know, and neither do you and that's [part of] the whole point." But actually we do know quite a bit that lets us rule out most of the possible infinite quantities of information with considerable certainty. All those electron spins I mentioned... the differences in forces resulting from those spins are going to be extremely small and mostly cancel each other out... so unless you can propose some mechanism by which they affect the processes of the cell, we can leave them out. What about the structure of Adenine? First insignificant amount of information there compared to the MB we're talking about (unless you propose detailed configuration of electrons and subatomic particles as relevant information, but then the onus is on you to justify how those details matter at all), and you only get to count it once, not once for each copy. You might propose the spacial location of each copy Adenine is important information, but other than the sequential positions of Adenine on the DNA strand, what will you propose, the spacial x,y,z coordinate relative to some reference? Again then you have to justify why that is important.
What we have strong evidence for is that it is the sequence of bases that matters. This is because we know that the cellular machinery responds with different behaviors sequence. I've dealt with the epigenetic factors elsewhere -- where are the instructions (condition logic) that control how the epigenetic factors are applied? In the genome (and in response, yes, to environmental factors... but where do most of those environmental factors come from?... signals from neighboring cells.... and where the instructions for what signals to send under what conditions?... the genome!).
What I see is a great many claims of more all kinds of sources of more information... all of which suffer from (a) failure to understand information theory in the first place; or (b) failure to understand how to distinguish relevant from irrelevant information; or (c) lack of any evidence or reasoning to justify proposed sources of information (the entire light cone my ass! looking at you KG); or (d) lack of justification why any significant amount of environmental information is incorporated into the fundamental structures and processes of an organism rather than in the the innate plasticity/adaptability of the organism (said plasticity programmed for in the genome).
Posted by: Stephen Wells
|
August 21, 2010 9:26 AM
Greylander, you're effectively claiming that all the information present in the uterine environment is captured by the phrase "the uterine environment". It isn't. You need to specify.
Posted by: KG
|
August 21, 2010 1:18 PM
GreylanderGarbage. Of course it contributes. Note your weasel phrase "in any worthwhile sense". All that means is "I don't want to admit it".
I have: if the conditions for the egg are slightly wrong, it will die. Potentially, anything within the light cone can cause it to do so - for example, by altering the spatial trajectory of a predator or pathogen so it eats/infects the egg. Incidentally, the idea that all baby rats with the same genome are near-identical if they survive at all is laughable.
It's very telling that you constantly resort to this sort of dishonesty. I think this sentence includes the first time I have even used the word "quantum" in this argument.
There's that weaseling again: "normal circumstances". How many bits does it take to specify what constitutes "normal circumstances"? Because all that information has to be included. In fact, there is no way it can be calculated - because there is a practically unlimited way in which circumstances can differ, and they don't come nicely labelled "normal" and "abnormal". All your (and Kurzweil's) babbling about information theory is utterly irrelevant. It is simply unjustified to pick out one component of the causal network that results in a living baby and say that only the information in that component counts.
How odd, then, that prediction of protein folding from physical principles has still only been carried out for tiny proteins (most protein folding prediction is done starting from known results from other proteins). Ah, maybe this means that guesses about how much information is involved in some facet of a problem doesn't actually tell you how hard the problem is.
How are you doing on predicting the structure and functioning of the brain from the sequence of nucleotides in the genome, by the way? Or at least, outlining the algorithms for doing so. I'll allow you a small amount of additional information, but you have to tell me where you'd get it from.
Posted by: Mario3k
|
August 21, 2010 3:45 PM
Guys, the brain is not so complex. We are not that smart. Simple.
A person smokes a cigarette.
Someone says: "cigaretts kill"
That person lights another cigarette.
"It will kill you"
No effect.
How smart a machine really is the brain?
It learns things (not to stop smoking) how impossible should that be?
Computers don't learn and don't create things yet. That's all. Most people never learn, let alone create.
What have you created recently?
Kurzweil has been creating a lot of interesting things. Most people in this post, including the author is just trying to refute his ideas.
Want my advice?
Go create, if the brain is so smart.
Good Night and Good Luck.
Posted by: stevieinthecity#9dac9
|
August 21, 2010 3:54 PM
That's the dumbest arguments here yet. We're not so smart. But Ray is. He'll be able to download his superior brain one day. Phhhht.
Stick to video games kid.
Posted by: Mario3k
|
August 21, 2010 3:58 PM
PS.: Before you start mentioning all the things humans created, let me tell you this:
We didn't create airplanes and microchips. This was created by a superorganism called society. Each airplane involves millions of brains.
We are not that smart. The superorganism of which we are a part is smart. And "it" will probably create a smart machine in 20 years (the real time frame Kurzweil predicted, according to his response to this post).
Posted by: stevieinthecity#9dac9
|
August 21, 2010 4:00 PM
You're not helping.
Posted by: enoonsti
|
August 21, 2010 4:05 PM
Dear David Gerard,
Dr. Greg Fahy is a biogerontologist and one of the leading experts in organ cryopreservation.
Here's your expertise in Solaris: http://davidgerard.co.uk/fuckhead.html
Dr. Brian Wowk is a medical physicist and cryobiologist.
Here's your expertise in Solaris: http://davidgerard.co.uk/fuckhead.html
Here is a talk by Dr. Wowk: http://video.google.com/videoplay?docid=2157944955525659858#
Posted by: Mario3k
|
August 21, 2010 4:05 PM
Are you hurt stevieinthecity#9dac9 (#750) ?
So you are saying you are extremelly smart?
Isn't that a bit arrogant?
Prove it. Have you created anything other than blogpost comments?
If you list anything, please attach website adress to prove it.
Posted by: Kristjan Wager
|
August 21, 2010 4:11 PM
Hello, Wright Brothers?
What I think you might be trying to say, in your rather incompetent and incoherent way, is that all inventions build upon the research of others, which is of course true. But society is not a super-organism in any meaningful sense of the word. It is, at best, a herd.
Posted by: enoonsti
|
August 21, 2010 4:20 PM
Dear BrianX,
I understand this was not directed at me, but I'd still like to respond to that "But who the hell are you?" with a sound "I am no one." Yet I still know that you're a TV producer, and my post directed to David Gerard is relevant to yourself.
Posted by: gamingmedley
|
August 21, 2010 4:24 PM
I imagine Ray's problem is less about understanding the brain as it is about understanding computers. A computer bit doesn't work the same way as an amino acid. At all.
Posted by: stevieinthecity#9dac9
|
August 21, 2010 4:41 PM
Mario I'm saying your argument is pathetic and irrelevant. My intelligence and yours are unimportant to the problem of "reverse engineering the brain". I don't have anything to prove to a fan of Ray. But it's cute you think that I should.
Posted by: Greylander
|
August 21, 2010 4:55 PM
[just notice I had this in the editor here but hand't hit 'submit' anyway the conversation has moved on to the new Kurzweil thread. I won't be tracking both -- if anyone else is still looking at this one]
@ #747 Stephen Wells:
Greylander, you're effectively claiming that all the information present in the uterine environment is captured by the phrase "the uterine environment". It isn't. You need to specify.
What, like graffiti left behind by the previous occupant? You claim that there is all this information I am obviously overlooking which gets incorporate into the fundamental structures and processes of the organism rather than incorporated into the innate plasticity of the organism. Give me an example. Suggest a plausible mechanism. Don't just wave your hands and say the uterine environment is complex. Of course it is complex. But how does any of *that* complexity get pushed into the developing fetus?
Posted by: Greylander
|
August 21, 2010 5:42 PM
Posted by: KG
Weasel phrase? I cannot write a book for each post elaborating every detail you think is important. In this case, the onus is upon you to demonstrate what/how information is significantly modifying the fundamental structures and processes in a healthy individual. Modeling the brain does not mean including modeling every way in which the brain may never have survived.
Now explain how "absence of conditions that killed me in the past" *adds* any information to the fundamental structures and processes of a normal healthy organism. You are making the extraordinary claim here. Where is your evidence, what are your reasons?You have a very strange notion of what it means to be "nearly identical" then.
What dishonesty? That is a question, not an assertion. I was mocking you for brining up the red herring of the light cone.Posted by: Greylander
|
August 21, 2010 5:44 PM
crap fucked up quotes again... well i'm done in this thread... see y'all in next one.
Posted by: Jim
|
August 21, 2010 5:47 PM
The name calling doesn't add anything to your argument, but I agree. Remember that Kurzweil is desperate to live forever. It clouds his thinking. However, He may not live to see the brain reverse engineered. I wish him good luck.
Posted by: dewexdewex
|
August 21, 2010 6:37 PM
Making a brain from an orgasm is more fun than pressing a button in a factory.
We await the factory orgasm, naturally.
Posted by: Jean-Luc Matteo-Donné
|
August 21, 2010 8:29 PM
Bonjour,
Peut-etre faut-il comprendre que si nous retirons le cerveau du corps et que nous le nourrissons... C'est theoriquement possible. Quelques drogues contre les souffrances dues aux connexions du systeme nerveux au corps. Nous pouvons le nourrir, le cerveau avec du sang artificiel, des nutriments qu'il a besoin et autres medications... C'est théoriquement faisable scientifiquement.
En somme resterait DNA noyau des cellules du cerveau et DNA mitochondrial. Ces DNA, surtout celui du noyau, peut être diminue de 98%, probablement, assurant juste le fonctionnement des cellules neuronales.
Nous pouvons deja brancher une grande partie du cerveau au virtual machines I.A. Et communiquer avec ce cerveau, le reconnecter entierement au virtuel dans un premier temps, puis à un corps reel non fragile dans un deuxieme temps.
Tout cela est possible des maintenant ! Je ne donne pas de liens si vous etes assez intelligents et scientifiques, logiques, pour trouver ces liens.
Dans 20 ans il sera possible de chouchouter, choyer le cerveau, faire le menage dans les cellules sans toucher aux centaines de millions de connexions neurales.
Nous pouvons des maintenant etre "immortels, dans le changement bien sur"!
Et pouvoir vivre 100 ans a partir de maintenant signifie avoir 100 ans de recherches scientifiques et technologiques en plus donc vivre 10000 ans voire plus.
Nous controlerions tout ce qui est relie à notre cerveau et meme notre cerveau. Nous serions toujours heureux, jamais nous nous ennuierions.
Le vrai probleme n'est pas l'immortalite ou quasi-immortalite mais celui de la connaissance de ce qu'est exactement notre identite, car en derniere analyse la seule chose que nous ne connaissons pas c'est l'identite, meme si nous la vivons nous ne sommes pas certains d'etre si reels que cela.
L'identite est un plus grand defi que la conscience reflexive, cette derniere nous echappant deja totalement.
Amicalement et bonne journee USA !
FRANCE
P.S. La mort n'est pas une fatalite !
La logique n'est pas une théorie, mais une image réfléchie du monde.
Tractatus logico-philosophicus (1918)
Ludwig Wittgenstein
Posted by: Trevor Turton
|
August 22, 2010 1:03 AM
Fascinating insights and comparisons. Thank you. From a CS perspective it might be more fruitful to regard the genome as virtual machine code (VMC) such as Java and C# compilers produce, rather than base machine code. VMC invokes members of extensive class libraries that provide a rich run-time environment. These class libraries are not specified in the VMC, they are created independently. It looks like the genome depends on an extensive class library of preexistent protein behaviour, which is not encoded in the genome, nor can it be deduced from the genome. Sometimes the answers to our questions are not as simple as we would like them to be.
Posted by: Aquaria
|
August 22, 2010 4:57 AM
#766:
It was a joke, even in the 70s.
You know that--right?
Posted by: KonradZeusNeumann
|
August 22, 2010 6:15 AM
PZ Myers is looking more and more like a washed up
scientific worm. If any one has followed the progress of the blue brain project can say confidently we are making large strides in neural simulation.
I have started to slightly dislike Dawkins now I hate
PZ Myers if I every see this man in person I will make him question his existence in the most painful
way.
I corner him on his rejection of AI based on primitive existential fear (intellectuals have more to fear, they have more to lose), on creating sensationalism for attention and more.
Do not pay any attention to PZ Myers hes worse then a kook.
Posted by: John Morales
|
August 22, 2010 6:37 AM
KZN:
Well, at least he allows kooks to post, as you prove.
Posted by: 'Tis Himself, OM
|
August 22, 2010 10:47 AM
KonradZeusNeumann #768
This is a particularly nebulous statement which actually says nothing. Describe "large strides" in a quantitative way, i.e. furlongs per fortnight or nanopascals per hectare or some other actual, meaningful measurement.
Myers is not rejecting the development of AI. He is rejecting Kurzweil's claim to reverse engineer the human brain in a decade. If you're going to throw accusations then accuse the person of something they've said or done, not on some presumption which satisfies your prejudices.
This is a classic case of projection.
Posted by: David Marjanović
|
August 22, 2010 12:02 PM
Bonjour,
Je regrette de t'informer qu'on t'a chié au cerveau.
Théoriquement, oui. En pratique ? Pas dans les décennies qui viennent.
Resterait après quoi ?
Tu as bien raison que ce n'est pas beaucoup plus de 5 % de l'ADN nucléaire qui a une fonction... par contre, le cerveau, ce n'est pas que des neurons. Il y en a énormément de cellules glia, par exemple, qui sont complètement indispensable pour le fonctionnement du cerveau.
Faux. :-|
Vraiment, dis-nous – d'où as-tu pris cette blague ?
Cela on ne peut pas encore faire. As-tu jamais pensé à la complexité du travail de brancher chaque neuron qui connecte le cerveau à quelque chose d'autre à des câbles menants à un ordinateur ?!?
Non.
Non, non, non. C'est toi qui offres l'hypothèse, c'est donc toi qui offres les données. Nous, on ne va pas faire tes devoirs pour toi. Ne sois pas tellement paresseux.
Extrêmement difficile à imaginer.
Comment ? En nous connectant à un ordinateur ???
Cela me semble comme une religion de rédemption...
Pourquoi ? Moi, je le trouve facile, ce problème : chaque cerveau est unique, à cause de génétique et épigénétique, des coïncidences dans le développement, et des expériences.
Autriche.
Ô putain. Maintenant il croit qu'il est drôle. <facepalm>
Before you start making threats that don't even mean anything, learn how to write on a computer. For crying out loud, there are line breaks inside your paragraphs.
Your laughable strawmen have already been adressed.
Posted by: Sven DiMilo
|
August 22, 2010 12:18 PM
Maybe I'm coming in a little late on this? But this assertion--just made up? Or meaningful in info-theory?
Posted by: SpeedOfSound
|
August 22, 2010 1:46 PM
Wow. Lots of comments on this one. I deal with this issue of the what it means to say "we can explain the human brain" or consciousness in a thread I started on RatSkep about reductions intent.
There is much confusion over many bold statements about what we can a can't do with science. Kudos to you PZ for putting your finger on this particular bit of press fanaticism. My studies in neuroscience lead me to believe that we could model a galaxy of stars with fewer bytes than we could one human brain.
I'm of the opinion that we can't even explain a single flame in a single fire with completeness let alone a brain.
On the other hand the crowd that loves to woo is quick to show you equally misguided press clippings about what neuroscience can't do or can't do yet! These clips are used by the other side to prove that we must look beyond physicalism for a solution. Above all they seem to be trying hard to get us look anywhere BUT physicalism and biochemistry. For some reason.
Oddly neuroscience has given me a solution to all the crap. Understanding just a bit about how brains function leaves one with a certainty that words like 'explain' and 'know' are about as flimsy and fuzzy as the the meat proteins that thought them up.
Much to learn,we have, about epistemology.
Posted by: Furcas
|
August 22, 2010 2:02 PM
A question for the naysayers:
If in the next decade someone like Henry Markram successfully emulates a rat brain, puts it in a robotic body, and this body acts like an intelligent animal (i.e. attempts to seek food, moves around without bumping into walls, reacts to stimuli, and so forth), will you all admit you were wrong?
Posted by: Furcas
|
August 22, 2010 2:04 PM
Ah, to clarify, the sentence above should read, "... successfully emulates a rat brain, puts the emulation in a robotic body, ..."
Posted by: SpeedOfSound
|
August 22, 2010 2:17 PM
I say nay nor yay. Certainly not nay, I'm an optimist. But we need to understand the scope of the problem. They have a reasonable cortical column running on a supercomputer. Yay! Now ten thousand more and how to hook them up and train them and then we can move on to the thalamus for a few decades. But there's much much more after the thalamus.
But why? The pet store sells rats. What is the intention of modeling?
Posted by: PZ Myers
|
August 22, 2010 2:38 PM
Are you suggesting that a Roomba is a full brain simulator? Because that would meet the criteria you list above, but it is not anywhere close to emulating a brain.
Posted by: Furcas
|
August 22, 2010 4:58 PM
SpeedOfSound wrote:
Understanding the brain, obviously. It's a lot easier to understand something if you can run it as many times as you like, modify a small part of it and run it again to see how its behavior changes, run the emulation a thousand times more quickly than the 'real' thing, and so on and so forth.
PZ wrote:
No.
First, my criterion is "acts like an intelligent animal", the rest were just examples, so a Roomba wouldn't qualify. I say, "like an intelligent animal" and not, "like the rat whose brain was scanned" because I don't expect the robotic body to be very good and thus a number of differences are to be expected.
Second, a Roomba is designed to accomplish certain goals. I agree with you that if something is designed to act like something else, that it partially succeeds in doing so is only weak evidence that it's like the thing it's trying to imitate on a deep level. However, a brain emulation like the one Markram is trying to create wouldn't be designed to imitate the external behavior of a rat, it would be a functional copy (or a close approximation) of a rat brain. If the alleged emulation is a failed copy, i.e. isn't a good approximation of the brain, it would be a hell of a coincidence if it just happened to behave more or less like a mammal does. Therefore, if the emulation behaves like a mammal, it's very strong evidence that the emulation is a good approximation.
Posted by: Kristjan Wager
|
August 22, 2010 5:04 PM
Of course, Kurzweil talks about a simulation, not a emulation. These things are not quite the same
Posted by: theswede
|
August 22, 2010 6:04 PM
Furcas, considering such a "brain" you describe would act in a manner very similar to a Roomba, and you don't expect us to "admit we're wrong" over a Roomba ... why would anyone admit to being wrong? Or for that matter *be* wrong? An emulator of a brain is rather far from what this whole debacle is about, so why would building one of those have any relevance?
I'm actually ashamed to be a Computer Scientist right now. But then, I actually have to work against the real world, with control of physical systems. Maybe that's why I have a slightly better appreciation of just how brutal the collision between theories and their implementation in physical systems is.
Posted by: Furcas
|
August 22, 2010 7:07 PM
theswede,
Either you're stupid or you just don't know what I'm talking about. I'll assume the latter and clarify:
As I mentioned above, I'm not talking about a program designed to act like a rat, I'm talking about a very accurate emulation of a brain made by scanning the brain and recreating all of the observed patterns in a computer. I'm talking about a functional copy, like the Blue Brain Project is trying to make.
Posted by: Jean-Luc Matteo-Donné
|
August 22, 2010 7:46 PM
David Marjanović,
Il est exceptionnel que je fasse le moindre effort pour quelqu'un comme vous qui montre une violence verbale extraordinairement contenue de plus. C'était juste pour écrire que cela se remarquait, que l'incompétence va que très souvent de pair avec la violence verbale. Rien a ajouter pour vous, je vous laisse vous démerder avec votre problème la limite de l'angoisse pour certains, de l'indifférence pour moi. On se retrouve dans 10 ans, si vous n'avez pas eu une crise cardiaque.
Cordialement.
Posted by: Jean-Luc Matteo-Donné
|
August 22, 2010 8:12 PM
David Marjanović, 20 ans sait déjà tout sur tout ! Je m'excuses de vous avoir pris pour un adulte, vous n'êtes qu'un gamin au fond, un gamin !
Vétérinaire des souris né en Croatie d'où des jeux de mots douteux ayant des problèmes d'identité.
Veuillez m'excuser encore, je ne savais pas. Le mieux est d'attendre quand même 20 ans, vous avez tout à découvrir par vous-mêmes, des millions de choses, on fera le point dans 10 ans. Bonnes études l'ami !
Sur le coup j'avais pas compris cette violence gratuite et cette arrogance directe sans aucune subtilité sans humour... hum, je suis vraiment désolé !
Posted by: yogismusic
|
August 22, 2010 9:15 PM
The brain is a computer of sorts, and I'm in the camp that says there is problem in principle with replicating it artificially.
Hmmm...I'm always wary of the brain/computer analogy. I think it shows we understand a lot more about a computer than we do about our brains.
Back in the early stages of the industrial revolution, everything was compared to a clock. Ther unvierse, and yes, the human brain was declared to be a clockwork of sorts. Now, during the computer revolution we make the same claims about the brain being a computer.
The problem with these kinds of analogies, as you point out, is not that they are entirely wrong. Maybe the brain does have some computer-like activities it performs.
What I disagree with is the overly simplistic jump from there to the idea that you can build a brain if you can understand how your desktop computer works.
This is Kurzweill's big mistake.
Posted by: theswede
|
August 23, 2010 1:25 AM
Furcas: Either you're stupid or you just don't know what we're talking about. I'll assume the latter and clarify:
The discussion here is not about emulating a brain - we can do that now, to the extent we can scan its patterns (which isn't much), and this has nothing at all to do with anything either the OP or RK has been talking about.
No, what has been talked about, from start to finish, is building a simulation of a brain. More specifically, how building a simulation of a brain is magically no more difficult than coding the genome.
Thus, seeing a brain emulator in function is completely irrelevant. It won't show that we understand the brain in sufficient detail to build a simulator. Nor will it show we have solved how to get from genome to brain. Nor will it show anything at all related to the discussion at hand, actually.
Posted by: Daniel B
|
August 23, 2010 7:11 AM
KG
This is merely a statement of opinion. I think you underestimate the capacity to test, cheat, fudge and 'go back to step X' in the digital realm. Of course nobody is suggesting that most efficient path will be to grow an embryo in a virtual womb.
Your rudeness and stupidity are becoming tiresome. We would only have to get the "Upbringing" (BTW I like how you criticise my term without making any attempt to come up with an alternative) of a human brain right once to have reproducible artificial intelligence. Furthermore we can make backups all the way along. So the onus is really on you to describe how it isn't possible to find a single successful path of 'upbringing' for our hypothetical newborn equivalent AI - not the other way around. The first training/parenting attempts could easily involve 100's of experts, billions of dollars and possibly even more than 20 years. But the economic incentives for true AI would certainly justify this expense.
Posted by: Nerd of Redhead, OM
|
August 23, 2010 7:15 AM
No, it is a statement of fact. You fanbois keep failing the science needed for the the genome to become the simulation. Your concern about tone is noted and rejected. Those who complain about tone have usually have nothing valid to say.Posted by: John Morales
|
August 23, 2010 7:19 AM
Daniel:
This is the stuff of dystopic nightmares.
Posted by: Stephen Wells
|
August 23, 2010 7:53 AM
Greylander, the fact that you think successful development contains only the information "I didn't die", when in fact it contains the information "At this time in my development I was in an environment with these properties (cue extremely long list of biochemical factors) whereas at this other time I was....", tells us that you have no appreciation for the real complexity of biology.
And, as commented elsewhere, the genome doesn't even fully specify the cellular environment, let alone the uterine one, so no, you don't get to claim it was all in the genome all the time.
Posted by: KG
|
August 23, 2010 8:18 AM
Daniel B.,
My rudeness is simply a response to your impenetrable stupidity.
I design and use simulations (in the "digital realm") as a large part of my job. So I know that these things take a lot of time and computational resources; and more important, that you need to understand why something has gone wrong to have a decent chance of fixing it.
If I tried to tell a funding body that the onus was on them to show that my proposed approach would not work, my chance of getting funding would be what, do you think?
You can't even get to your supposed starting point, the "hypothetical newborn equivalent AI": it would have to be a robot that was able to continue growing its brain, and that people would react to as they would a person. The brain starts learning long before birth - from as soon as neural impulses can propagate - and carries on growing new bits and losing old ones throughout life. Learning, development, embodiment and social interaction are inextricably bound up with each other. That you don't grasp this is a good measure of the depths of your ignorance; that you refuse to do so when it is pointed out to you is a good measure of the depths of your stupidity.
An unsupported opinion is not evidence. We have a reasonable low-tech way of producing human-like intelligent agents. Computers are most useful for things we can't readily do in our heads.
Posted by: JeffreyD
|
August 23, 2010 8:51 AM
Jean-Luc Matteo-Donné #782 and #783.
Talk about a lack of subtly or a sense of humour, do you have anything to actually say about the topic or do you only plan to insult David Marjanović about his age and complain about his tone? Your post at #765 added little - there is small worth to saying, "I won't post links" - often that means you have none.
Yeah, many of us can read French. Not sure how "tone troll" can translate into French, but vous êtes un idiot ou un troll should be an acceptable translation.
Posted by: David Marjanović
|
August 23, 2010 2:13 PM
"Le moindre effort" ? Nous donner un scénario à peine cohérent et conclure en nous disant, avec un sourire arrogant, que nous ne sommes pas dignes d'évaluer vos données ?
Et si ma "violence verbale" est "extraordinairement contenue", cela la fait pire ? :-D
(Le moindre effort, en effet, cela aurait été d'écrire en anglais, pour que le reste du monde puisse vous comprendre. Si vous pouvez lire 781 commentaires en anglais, vous pouvez en écrire un ! Mais il y a bien des autres francophones ici.)
Si je dis "c'est stupide, parce que" et le suivis par une liste de raisons, je comprends que vous voudriez que ce soit un signe d'incompétence, mais... :-)
Vous semblez croyer que je sois enragé au-delà de toute mesure. Mais non. Je vous montre du doigt, et je ris !
Allons. Un commentaire sur un blog, ce n'est pas une lettre.
C'est bien vrai que je ne suis pas adulte, mais quand même je viens d'envoyer ma thèse doctorale à l'impression. J'ai 28 ans.
Il semble que vous avez cherché mon nom sur Google. Attention, on m'a dit qu'il y a sept gens ayant mon nom sur Facebook, et aucun d'eux n'est moi ! J'ai même un homonyme qui vit en France.
Me voilà. Pour me trouver, il faut scholar.google.fr plutôt que www.google.fr.
Attendez un peu. Voyons si j'ai bien compris ce que vous voulez dire:
Il y a des jeux de mots qui sont capables d'avoir des problèmes d'identité... ô pardon, c'est le vétérinaire qui les a, et vous avez oublié les parenthèses. C'est ça ?
Faire des jeux de mots est causé par une naissance en Croatie ? ~:-|
J'ai fait des jeux de mots (douteux ou non) ? Ce n'était pas mon intention, et je ne me souviens pas d'en faire. SVP expliquez-moi qu'est-ce que j'ai fait.
Viennent de se terminer, merci. Oui, je vais faire un postdoc (et puis des autres jusqu'à ce que je trouverai un poste), mais... :-)
Violence ? Vous savez qu'est-ce que cela veut dire ? Dire "c'est stupide, parce que", ce n'est pas de la violence. C'est comme les scientifiques parlent entre eux. (Pas forcément comme ils écrivent leur articles, mais comme ils parlent entre eux.) Sans ironie ni humour, je vous traite comme un collègue.
Posted by: Furcas
|
August 23, 2010 8:01 PM
theswede wrote:
First, regarding the correct usage of 'simulation' and 'emulation', I've seen these two words used interchangeably by experts (no, I'm not talking about Kurzweil) when referring to creating a functioning copy of a brain. For example, this article talks about IBM's brain simulation (the largest to date), while this one (by Anders Sandberg and Nick Bostrom) talks about a roadmap for whole brain emulation. If you bother to read these articles (which, of course, you won't), you'll see they're about the same kind of thing.
Second, this has a lot to do with what Kurzweil talked about. Actually, no, scratch that, it's exactly what he's talking about. It's not what the OP is about, because PZ foolishly or dishonestly misinterpreted a bad third hand article about Kurzweil's speech, rather than waiting for the speech to be online or even to read his book (which he says he's done, but it's obviously a lie, he's only read the part about exponential growth throughout history, which is only a third of the book). Kurzweil has not said that "building a simulation of a brain is magically no more difficult than coding the genome", or anything even remotely approaching such a foolish statement.
Third and finally, you really don't have a clue. You write about "understand[ing] the brain in sufficient detail to build a simulator", but that's the exact opposite of what Kurzweil, Markram, the IBM research group, Bostrom, Sandberg, Hanson, and many others have in mind. The point is to scan and emulate/simulate the brain in order to understand it, not the other way around.
Posted by: DANIL
|
August 23, 2010 11:10 PM
We live in an artificial environment created by ourselves, nature is mindless, its behavior is capricious,man did not invent man,man has a brain,nature does not, mans struggle to liberate himself from the constraints of nature,is a major chapter in the history of mankind. Humans are not born with knowledge, we form ideas from data we acquire, data that has been processed previously by others, some dead, some alive.We make reality real, by processing in our brain the data we obtain from our senses,and we have learned to create methods of identification, and devices to "make real" things we can not perceive through our senses. Dr.Jose Delgado showed 30 years ago how through implanting electrodes in certain areas of the brain, we could control in a subject, emotions,behavior,and other urges. Evolution is a scientific fact, man has a past, a present and a future. Time and space, are understood by the human brain in a unique way.
Will the artificial brain follow a slow learning process in order to acquire the consciousness of time? will the artificial brain,understand the artificiality of our environment as distinctly different from the natural in nature?
Posted by: Jean-Luc Matteo-Donné
|
August 24, 2010 1:42 AM
David Marjanović, j'ai la chance d'être assez riche pour vivre en marge de la société, je n'ai donc pas d'obligations de résultats débiles que subissent la plupart des gens. Je ne suis pas un collègle, ni un ami, ni un ennemi !
Maintenant je n'ai absolument pas pas à prouver quoique cela soit si je pense que je m'adresse à des personnes avec un haut niveau e connaissances. Je ne suis pas un enseignant et ne désire pas l'être! Si vous niez ce que je dis alors prouvez-le SI vous voulez ! Mais ne me demandez pas de prouver des dire qui sont d'une telle évidence pour moi et certains de mes amis transhumanistes français et intellectuels dans mes relations de chaque jour.
Si je vous dit que l'on peut couper la tête de quelqu'un, endormir son cerveau en parti par des drogues et le nourrir avec du sang artificiel et des nutriments nécessaires c'est que j'ai bien pesé ces possibilités depuis pas mal de temps. J'ai des arguments mais tout ne se dit pas avec des mots sinon je vais être obligé de perdre une journée pour vous seul, et je vous ai écrit que je ne ne me sens pas un enseignant, donc...
Also,
Je sais très bien écrire et lire l'anglais mais mais j'excelle en français, je préfère cette mangue et de loin; elle me semble plus subtile.
H+
P.S. : Jeffrey vous m'êtes très antipathique, d'ailleurs ceux qui crie au troll au moindre problème sont de vrais boulets. Rien d'autre à ajouter.
Posted by: Jean-Luc Matteo-Donné
|
August 24, 2010 1:51 AM
corrections du #795
*pas pas, supprimer un des deux pas.
*de connaissance et non d e connaissances
*si vous le voulez et non si vous voulez
*si je vous dis et non si je vous dit
*cette langue et non cette mangue.
*** Je suis réveillé depuis 10 mns, enfin***
Je trouve très lourd que l'on ne puisse pas se corriger, au moins dans les 5 ou 10 minutes, après avoir écrit un message, ici.
Posted by: Stephen Wells
|
August 24, 2010 6:40 AM
@793: Kurzweil has said, and emphasised in his attempt to rebut PZ's criticism, that "the design of the brain is in the genome". It's not, and his claim is dumb enough to justify all the flak he's getting.
Posted by: KG
|
August 24, 2010 7:30 AM
The Prophet Kurzweil (May He Be Uploaded) can't even keep his story straight for the course of a single talk. In this video, he predicts reverse engineering of the brain both by 2019 (at around 10:00), and by 2029 (somewhat later in the video). If he means what "reverse engineering" usually means (understanding how a device works to the extent that you can build something that performs the tasks that it performs in the way it performs them), both dates are absurd.
Even Markram, who unlike Kurzweil is a nueroscientist, says quite different things when he's boasting to the media:
From what he says when he's writing in peer-reviewed journals. Here's the abstract from the liked article:
Or even in the Blue Brain FAQ, where there is an explicit denial that Blue Brain is an artificial intelligence project.
Thus whatever terms are used (modelling, simulation, emulation) there is a systematic ambiguity about what is meant: build a machine that solves the problems the brain solves in the way it solves them (or that can learn to do so as a child does), or produce a simulation/model/emulation of the whole brain (or much of it) that is useful for neurophysiological research. The second is just about feasible in 10 years (but probably not one that works in real time), the first is a pipe-dream. Kurzweil undoubtedly requires the first, because he thinks uploading will be possible in the late 2030s(The Singularity is Near, p.324) - but if the second is done in around ten years, he will certainly claim it as a successful prediction.
Posted by: DANIL
|
August 24, 2010 10:48 AM
Liberation from the constraints of a mindless nature has been possible because of our brain. The human factor appeared during the process of constructing an artificial world.
It is our destiny to become more artificial, to harness all the forces of nature to assure energy and food for the sustainability of an over populated world.
Artificial intelligence is the essence of humanity.
The creation of a “simulated” or “emulated” brain, is an absolute priority, the survival of our species, will depend to a large extent on the success of this endeavor.
Posted by: https://www.google.com/accounts/o8/id?id=AItOawku8Fdhltvvaj8PR1_AP247K9q5xGzmViI
|
August 25, 2010 6:17 AM
I like your article, but feel it was unnecessary to include the personal attacks of Kurzweil in there. Just make your points based on your facts and analogies. There's no place for this kind of behavior in scientific debate. It's a failure of ethics and logic to think that personal attacks help to win an argument.
Respectfully,
Max
Posted by: KG
|
August 25, 2010 6:36 AM
DANIL,
That's insane. We know what needs to be done to achieve sustainabilty. The main obstacles are political. Scientific and technical research is needed on improving the efficiency with which energy, water and other resources are used, and on generating renewable energy. Results are needed in the near future. A "simulated" or "emulated" brain is completely irrelevant to these problems.
Max,
This is not a scientific debate. Kurzweil is a messianic cult leader.
Posted by: David Marjanović
|
August 25, 2010 8:47 AM
Ben, si. Vous proposez une hypothèse, c'est à vous de montrer au reste du monde les données qui soutiennent l'hypothèse. C'est ça comment ça fonctionne en science.
Mais évidemment vous désirez être un scientifique (ce qui veut dire quelqu'un qui fait de la science, pas forcément un professionnel) : vous proposez une hypothèse, et vous mentionnez qu'il y a des données qui la soutiennent...
Donc, je vous conseille de vous comporter comme un scientifique et nous mener à vos données. Sinon, nous allons continuer de rire à "ce mec aléatoire sur internet qui croît qu'il va tout bouleverser parce qu'il n'a rien compris".
Je répète : je n'ai pas besoin de prouver qu'il n'y a pas de théière qui tourne autour du soleil, entre la Terre et Mars. Ce qui est proposé sans données peut être nié sans données.
Pourquoi c'est "d'une telle évidence" pour vous tous ? C'est ça que vous devez expliquer si vous voulez que l'on vous prenne comme sérieux dans un environnement plein de scientifiques.
Vous n'avez pas remarqué ce que vous avez fait ? Vous êtes venu ici, disant "vous avez tous tort, et Kurzweil a raison, mais je ne peux pas expliquer pourquoi, adieu".
Vous avez proposé une idée, vous êtes déjà obligé de la soutenir.
Ce n'est pas moi seul qui s'intéresse à cette discussion ; vous avez bien vu que nous deux ne sommes pas seuls ici.
OK...
On pense toujours cela des langues que l'on connaît mieux. :-)
Là, je suis bien d'accord, mais les informaticiens de ScienceBlogs n'écoutent pas bien à leurs lecteurs... mais, la prochaine fois, cliquez tout simplement sur "Preview".
==========================
There is no such thing as destiny. The universe isn't a video tape that is played.
There is no such thing as an essence.
Why do you think so? ~:-|
You haven't watched a lot of scientific debates, have you?
I've watched one on my thesis topic where I genuinely feared I'd need to physically step in to prevent the two esteemed colleagues* from jumping at each other. They got louder and louder, and at least one of them got red in the face. (The situation defused when one of them brought up some evidence for his hypothesis.)
* This is not sarcasm.
What makes you think that PZ thinks they do???
The insults are added, they aren't part of the argument itself. They're part of the conclusion, not of the materials or the methods. I'm rather surprised you didn't notice.
Posted by: JeffreyD
|
August 25, 2010 10:43 AM
Jean-Luc Petard at #795 - vous êtes pour moi très antipathique parce que vous êtes un âne. Dans les mots d'Albert Camus, le cerveau il etait en option chez toi.
Posted by: Rev. BigDumbChimp
|
August 25, 2010 10:47 AM
I believe the blog you are looking for is here.
Posted by: Paul W., OM
|
August 25, 2010 10:55 AM
yogismusic:
(*whips out soapbox, stands on it, begins ranting*)
I'm always wary of people who think it's an analogy.
It's not, IMNSHO; it's a literal truth.
A big problem is that many people don't know what a computer is.
A computer is just a fairly complicated and flexible pattern transformer. That's all, and that's a lot.
A computer is anything that can take complicated patterns as inputs, flexibly transform them according to other complicated patterns, and optionally output appropriate transformed patterns. (Which may or may not be complicated, e.g., a big complex representation of something, a series of commands to a motor system, or a one-bit yes/no answer.)
Not all computers are binary or digital. Not all computers are serial. Not all computers are von Neumann machines, and not all are programmed by humans. Computers can be made out of pretty much anything, because what you need is just a way of making various differences in state---any aspect of state---affect other differences of state in certain patterned ways. (We just happen to use transistors because it's easy to copy a zillion tiny patterns in bits of silicon and use them as fast, reliable interconnected parts.)
Some computers are analog or hybrid digital/analog, and some are massively parallel and distributed, and asynchronous by default.
Zillions of them are made out of meat, and we call them brains.
A brain is a computer; it's just not your father's Intel 8080 chip, programmed in BASIC--- it's a way cool computer.
Saying that is very unlike comparing it to a clock or a steam engine. The brain is not about force or leverage or pressure or gear angles, or any particular physical aspect of any particular physical kind of thing. It is about information, represented by any set of physical differences that can be used to make the right differences in the mappings from the input(s) to the output(s).
I basically agree with that.
One problem, though, is that most people don't know how a desktop computer really works, under the hood. They don't realize that desktop computers do have a lot of "weird" features that brains do, in some reduced form, somewhere in there.
For example, most people don't even realize that "serial" and "digital" computers are built out of parallel digital parts, which are in turn built out of even more parallel analog parts---what makes them compute digital functions is not a local property of "digital" parts, but an emergent property of how they're connected together; that's not entirely unlike the way discrete representations emerge from analog neural networks, although it's usually hidden at a low level of implementation, and abstracted away from.
Likewise, what makes the parallel implementation of even a "serial" processor give the illusion of strictly serial execution---just faster---is an emergent property of nonlocal regularities in how the processing and storage units are connected in parallel. Those parallel units (e.g., arithmetic units) contain even more parallelism within them, too. That is apparently not entirely unlike the way the brain works, in certain basic respects---e.g., we have an illusion of serial conscious processing that emerges from rather parallel connection of various subsystems, which themselves contain a lot more parallelism---and there are some basic advantages to that kind of organization for both kinds of system, despite their many differences.
Of course, if you think a serial processor somehow just executes "serially," as per the von Neumann abstraction you learn in a beginning programming course, you'll never understand the real commonalities.
Likewise, many computer programs, such as operating systems, commonly do things that many people think computers don't do. (e.g., reflect on their own structure and state, and modify themselves, albeit in extremely rudimentary ways.)
Unfortunately, most people---even most computer weenies---don't know enough about computation in its full generality to see how the brain is a computer, or to make good use of that knowledge in understanding cognition and neuroscience.
Still, I don't think it's a good idea to say that the brain "isn't like a computer," or "isn't a computer." That's not just misleading but false, and it missing the main point of computers, or brains, or both.
What the brain isn't much like, in some important ways, is most people's naive concepts about what they usually call "computers," and their even more naive concepts of how they relates to computers generally.
Posted by: KG
|
August 25, 2010 11:04 AM
Come off it, Paul W.,
When people say the brain is not/is not like a computer, they are using "computer" in the sense that it's normally understood: a present day, digital computer, designed rather than evolved, and designed moreover with as clean a separation of levels of structure and functioning as possible. The brain is no more a computer than it's a telephone exchange - which was actually the immediately preceding metaphor.
Posted by: Paul W., OM
|
August 25, 2010 1:01 PM
KG:
I don't think that's all that's going on. I think a lot of people really, really don't get what a computer is or isn't, in a way that's seriously problematic for this kind of discussion.
Suppose somebody said that it was just a "bad metaphor" to talk about people being animals---people aren't animals, so animal models are misleading at best for the study of people.
That might have some merit to it---if what they mean is that humans are qualitatively different from nonhuman animals in certain relevant ways for a certain discussion.
But keep in mind that a lot of people use "animal" to mean something like quadruped, and don't think of fish and flatworms as animals, exactly---maybe not even birds. If you hear somebody flatly saying human's aren't animals, and that it's just a metaphor, like comparing humans to clockworks, you know that something is seriously wrong somewhere and you at least need to back up and straighten out the terminology.)
Anybody who really knows what the relevant differences between human and nonhuman animals are, and whether they're qualitative or quantitative, just isn't going to say that "humans aren't animals" or that it's just a naive metaphor. They understand that it's not just a metaphor, even if humans are not typical or prototypical animals.
Likewise, most people won't dismiss physicists' talk of "light waves" by flatly saying that "light isn't made of waves"---you know, like on the ocean, with moving water---and that it's just a naive metaphor to talk about light that way.
In either of those cases, an appropriately knowledgeable person would acknowledge the ambiguity in the term---whether you're talking about a general definition or just resemblance to a common prototype---and that there's more than a mere metaphor at stake. There are basic animal phenomena, but humans are literally a kind of animal; likewise, there are various basic wave phenomena such as transmission, reflection and refraction, and light is made of waves in a literal sense, even if it's not the prototypical sense of water-surface motion wave that most people would think of first.
I'll stop taking issue with statements like "the brain is not a computer" when biologists stop objecting to "humans are not animals" and physicists are happy with flat statements like "light isn't made of waves," coming from people who seem not to know what physicists mean by "wave."
I'm not going to hold my breath.
Scientists aren't supposed to let naive people define scientific terms in a discussion of scientific issues. They're supposed to tell people who use terms too narrowly that they're doing that, and that they don't seem to understand the relevant and important phenomena in their full generality, or in enough detail.
I might be happy to abandon the term "computation" to popular definition if there was a better term for what I'm talking about---computation---or if it wasn't an important concept, or maybe if it just wasn't relevant to the discussion.
But computation is tremendously important to discussions of the brain and mind. If we're going to discuss the relationship between what the brain is doing and computation at all, it's pretty important that people understand that what the brain doing is entirely or almost entirely computational. If people don't know what that means because of preconceptions about "computers," they need it explained to them, so that they have a basic understanding of the subject and the proper terminology.
---
One thing that makes all this very salient to me is that I recently went to a meeting of a philosophy book discussion group, where the subject was the mind/body problem.
Half the people there thought that the brain is literally not a computer, even in the general sense of "computer" I'm talking about.
These are educated people, all of them with at least a college degree, most of them atheists, and all of them interested in philosophy. And half of them think that the brain isn't basically a computer, even in my general sense, and that it's doing something fundamentally mysterious that scientists don't even have a basic conceptual handle on. They think that there's some profound link between matter and consciousness that will require a non-computational account of the mind and brain, and even revisions of quantum theory to explain how matter can be "aware," and stuff like that.
They really have no idea what computation actually is, and what even von Neumann computers can do in principle. They've heard too many people say that the brain is not a computer, and that computers can't do what brains do, and they think there's a profound qualitative difference between mysterious squishy brain stuff and boring, limited, dry computation stuff. Not just the kind of difference you and I understand between meat computers and silicon computers, but something basic, deep, and profoundly mysterious.
They do not that the brain is literally a massively parallel hybrid digital/analog computer; they are not just saying that it doesn't closely resemble a Z80.
Half the people at that book group dismissed me an another knowledgeable guy as kooky everything-looks-like-a-nail computer nerds for thinking that a computational account of the mind is even plausible.
I think that sort of naive, smug, dismissive attitude about the generality of computation is quite common. I see it all the time---even here on Pharyngula, to some extent.
I think it's an important point that biology is mostly computational, even at the cellular level---the genome literally is mostly an unfamiliar kind of program---and that of course intelligence and awareness are too.
Computation is a tremendously scientifically important natural phenomenon, and if people don't understand how that can be true because they equate computers with silicon chips, they really need it explained to them.
---
And that's bullshit. A brain is obviously much more like even a desktop PC, in certain crucial qualitative ways, than it is like a phone exchange. (Especially an old-style pre-"computer" phone exchange from when that metaphor was popular.) A PC can prove theorems, parse complicated syntaxes using a declarative grammar representation, do image analysis, carry on a rudimentary conversation about a limited subject, etc., etc., etc. And in principle it can do much more, when we figure out the programming.
An old-style phone exchange just can't. It isn't a universal computer, and hardly does any computation at all---it's basically a modestly flexible information routing system, not a flexible information processing system. It can't be programmed to reason or to learn in any complicated way.
(Notice that if you include the information processors connected by the routing system in your metaphor---i.e., the people connected by the phone system---its significance changes drastically. Now you can do some serious computation. Even if you assume that they're dumb people but capable of following simple instructions, you're onto something, because now you've got yourself a parallel computer. You can only fix the phone system analogy by putting the computation part in, because what the brain is doing is computation; that's what all the routing is in support of.)
Sure, the brain and a silicon computer are very different in some obvious ways that you and I know about---e.g., how software is or isn't cleanly delineated from hardware, etc.
Still, I'm pretty sure that most people who latch onto that and say "the brain is not a computer" and that that's "a naive metaphor, like comparing the brain to a phone exchange" don't understand the issues the way you and I do, and are substantially wrong in how they mean those statements.
You and I know that even in prototypical (silicon) "computers," software (informational differences) is always implemented in hardware (physical differences) in an important sense; we know that the line between what we call hardware and software is not always clear. (Hence terms like "firmware," and microcode, and PROMS, and programmable logic arrays, and so on ad nauseam.)
We know that functions can generally be implemented in either hardware (e.g., a parallel hardware voxel processor) or software (e.g., an emulator for a foreign instruction set), and that it's usually just a boring engineering decision for this kind of discussion, not something of deep philosophical significance, or a big qualitative difference between brains and computers, or even between brains and prototypical "computers."
Most people don't know what you and I know, including a lot of people who try to discuss the kind of stuff we're discussing, and they get hung up on shit that just doesn't matter much, thinking it's a great big deal; they dismiss talk of brains as computers largely because they just don't understand computers, not because they do.
When they stop doing that, and saying that the brain "isn't a computer," I'll stop pointing out that "it is too!"
Posted by: Paul W., OM
|
August 25, 2010 1:22 PM
By the way, I'm extremely skeptical that anybody smart and knowledgeable ever actually thought that a telephone system was a good "metaphor for the brain," because they already got the general idea of computation, more or less.
I'm pretty sure that everybody with a clue already understood that such a brain would be utterly useless without something to process the information being routed around, and that the telephone system analogy always did include something very like computational elements processing information connected by a routing system, somewhat like a telephone system connecting people.
They assumed that local nodes did the kind of thing computers do---that idea had been around since Leibniz and Babbage, and had gained currency---although whether a local node was a programmable or universal computer was up for grabs.
And given that, it's still not a terrible metaphor, if you don't push it too far. The brain really is a bunch of much-smaller information processing elements connected by an extensive communication network. (It's not a general automatic any-node-to-any-node routing system---but in those days, telephone systems weren't quite that, either---many people didn't have phones, some used their neighbors', some had party lines, connections between local phone systems might require manual intervention, and long distance was different, too.)
And even in those days, nobody thought that the telephone system was more intelligent than its subsidiary processors (i.e., the people). They realized that the local computation and communication were coordinated in ways that made the distributed system much smarter than its local parts.
Those people weren't so naive to compare the wiring of the brain to a telephone system---the things they were surest of were right---and we're not repeating some major everything-looks-like-a-nail mistake they naively made.
Posted by: windy
|
August 25, 2010 2:22 PM
Stephen Wells:
KG:
You and KG may be throwing out a baby with the Kurzweilian bathwater here. Yes, development requires many environmental and cellular preconditions that are not specified in the genome. But the environment doesn't care if an egg hatches or if you have five fingers or a functioning brain. Development can't depend on a near-infinite number of environmental inputs being "just right" by chance. The calculation on how to achieve good enough environmental conditions often enough to produce something that's close enough to particular end result has already been done by natural selection! (It's another question whether it's feasible to 'reverse engineer' that)
Posted by: Stephen Wells
|
August 25, 2010 3:48 PM
@windy: it's not the external environment I'm mostly worried about. The thing is that if a genome is put anywhere other than in a functional cell, it won't do the things that maintain a functioning cell, and reading the genome doesn't tell you that; greylander can't or won't see this.
As a side effect, he was also ridiculously dismissive of the developmental environment, but that is not the biggest issue.
Posted by: windy
|
August 25, 2010 5:29 PM
Yes, but I don't know if this is quite as strong an objection as it seems. If we were trying to reverse engineer an alien von Neumann probe that contained all the necessary instructions to maintain and replicate itself, how much would it matter that the instructions don't tell you how to build one from scratch?
Posted by: theswede
|
August 25, 2010 5:48 PM
Furcas,
First, regarding the correct usage of 'simulation' and 'emulation', I've seen these two words used interchangeably by experts
So have I. Even experts are wrong sometimes. The words are not interchangeable just because they have been misused by experts. For one, the purpose of emulation and simulation are different, and therefore there will be subtle (and sometimes not so subtle) differences in practical results.
Kurzweil has not said that "building a simulation of a brain is magically no more difficult than coding the genome", or anything even remotely approaching such a foolish statement.
That is precisely what he has said. And yes, it is a foolish statement.
You write about "understand[ing] the brain in sufficient detail to build a simulator", but that's the exact opposite of what Kurzweil, Markram, the IBM research group, Bostrom, Sandberg, Hanson, and many others have in mind.
That may very well be, but it isn't the opposite of what they're talking about.
The point is to scan and emulate/simulate the brain in order to understand it, not the other way around.
Emulate or simulate? The two are very different, and will lead to very different results. An emulation will let us run the thoughts running in scanned brains, while a simulator will allow us to tweak the thoughts and see how that changes the output. And yes, I am aware of the overlap between them, but for purposes of study a simulator is not only vastly superior, but the only real way forward. An emulator can only get us so far, unless we reach perfect fidelity of course - but we're so far away from even a rough outline that speculating what that would entail is at best a waste of time.
I see a lot of noise claiming "But RK can't have said what he evidently said, because that's foolish!", and your arguments are no different. What's with that approach? RK is wrong, so what? Big deal, he still made some of the keenest electric pianos ever.
Posted by: KG
|
August 25, 2010 6:36 PM
Paul W.,
No, the brain is not entirely or mostly doing computation in anything like the computational theory sense, and no, the genome is not a program in anything close to the computational theory sense. The brain is constructing and reconstructing itself all the time, which a computer is not, and is also the seat of emotions, and is able to generate its own top-level goals. The genome is not substrate-neutral, and it is not a string of symbols from a finite alphabet. Both are parts of larger self-maintaining dissipative systems.
Notice that exactly the same is true of all current computers. Embodiment and social interaction are not minor twiddles, but fundamental to intelligence.
The fact that these layers are cleanly separated, however, while they are not separated in the brain, is of considerable philosophical significance. As is the fact that brains have evolved by natural selection, while computers have been designed. These distinctions may well begin to melt away in the next century, but they have not done so yet.
windy,
Nor does the genome.
It doesn't need to, because often enough, the genome finds itself in an environment which gives it the inputs that cause development to occur - but what this environment is, is not specified in the genome.
Posted by: John Morales
|
August 25, 2010 9:40 PM
KG,
The womb is also "coded" in the genome, so it's recursive instantiation iteration going back to first life (the bootstrap). And the womb needs a life-support system.
Posted by: Paul W., OM
|
August 26, 2010 2:41 PM
KG:
I'm not sure what you mean by this. I agree that the brain, like any organ, is doing a lot of stuff that isn't its function, but is necessary for its function, given the overall design. I don't think that affects my claim that the brain is a computer. The heart is a pump, but of course it's also an organ that grows and self-repairs to a considerable extent as well. That doesn't make it not a pump. Likewise for the humerus being a lever, the kidney being a filter, and so on. The fact that something is a self-regulating and self-repairing X doesn't mean that it's not an X; it just means it's a fancy one.
Even my PC does that sort of thing, in rudimentary ways. It includes a thermostat and a fan, and little rubber feet that damp vibrations, for example, so even my PC isn't literally only doing computation. It's self-regulating and even self-repairing to a certain extent---it turns its fan up or even its clock speed down when it gets too warm; it detects bad blocks on the magnetic disk and flash RAM disk, remaps them to spare blocks, etc. Most of the mass and structure and volume of my PC isn't interestingly doing computation, but that doesn't mean that it's better described as a paperweight or a monitor support than as a computer.
Certainly brain cells and brains are doing a lot more of that sort of subsidiary support stuff, but I don't think that means that the brain isn't a computer, or that it isn't best described as a computer. IMHO it is precisely a computer in the sense that my PC is, despite the fact that I also use it to support a monitor, and indirectly, the little whiteboard and dry-erase pen I have attached to the monitor. Its primary and dominant function by any reasonable standard is to compute, even if it happens to have a spandrely auxiliary function as a structural support for a whiteboard as well. So you're right, the brain isn't just a computer, but then neither is my PC. So what?
I'm also not clear what you mean by "the computational theory sense"---I suspect that your idea of "computational theory" is much, much narrower than mine. For example, analog signal detection theory is computational theory in my general sense. Computational theory in the general sense isn't just limited to traditional automata theory or to algorithmic analysis for idealized von Neumann machines. It also includes things like theory of robust nondeterministic distributed computation in networks made of unreliable parts, emergence of distributed representations in neural networks, etc., etc. That's all theory of computation in any sense general enough to be relevant to whether something is a computer.
Again I suspect your concept of "computational theory" is way too narrow. (I also suspect that some things are closer than you realize to things you'd recognize as computational.)
As far as I know, the bulk of the genome is a forward-chaining, nondeterministic production system. Genes function as discrete rules with discrete (but sometimes analog valued) conditions and consequents, that fire asynchronously and in parallel by default. That is mainly what most of the genome does, and it is the main reason that it works. It works because it is a kind of evolvable program for controlling the cell. (And indirectly, the whole organism.)
I've built forward-chaining nondeterministic asynchronous production systems myself, and none of my computer science colleagues ever imagined for a moment that what I was doing was somehow not computing, or that the rule set did not constitute a program. Of course it's a program that computes; it's just a different style of program and computation than most people know about. The fact that most people are woefully ignorant of most styles of programs and computation doesn't mean they're not precisely programs and computation.
In the most important sense, I think that's false. A typical computer is constantly reconfiguring itself, e.g. restructuring its memory by moving abstract units of memory between different physical units of memory of the same sort, of different sorts, or both, multiplexing abstract processors across fewer processors, and so on. Many times a second, even.
The fact that human-constructed computers mostly do this the easy way---e.g., by changing voltages that open and close gates that implement the reconfiguration---doesn't mean it isn't happening, isn't real, and isn't basically the same thing. The fact that the brain does it the hard way, e.g., by growing new dendrites, doesn't mean that it's all that fundamentally different. It's certainly an interesting and important area of study, but I don't think it's the kind of thing that changes whether something is a computer, or whether what it's doing is implementing a computation.
Are you suggesting that emotions and top-level goal setting are not computational? If so, I beg to differ. Emotions are basically the regulatory states of a kind of real-time computation, pervasively affecting how the computation proceeds in certain basic ways. Goals are implemented by representations that have a certain relationship to planning computations.
I don't think emotions and goals have anything to do with not being a computer. In fact, I think that it turns out that to have emotions and goals, it is necessary to be a computational system. Nothing but a computer can represent things in the way that constitutes their being goals, or operate on those representations and other representations in a way that would count as "choosing" a goal.
I don't think that there's anything non-computational about top-level goals either, or that we can somehow non-computationally choose them. To the degree that our general goal schemas are evolved-in and relatively fixed, they're preprogrammed, and that programming is done indirectly by the genome. (E.g., if you develop normally, you'll generally have a sex drive, an aversion to pain, a desire for control of your life, a need for socializing, and so on.) In general, we can't choose arbitrary top-level goals. We're wired to have a strong tendency to want certain kinds of things and want to avoid others. Sure, we can choose among good things, and reconceptualize things and see how to want things we didn't want before, and sacrifice some for others, etc., but so far as I know there's nothing non-computational about how that happens. It's all computation.
What exactly do you mean by that? In a general sense, no software is substrate-neutral. Any software only runs directly on a small range of possible kinds of hardware. (Or subsidiary software such as operating systems that implement a kind of virtual machine, more or less.) That's why we have to build emulators to run some x86 PC programs on an x86 Mac, or PowerPC Mac programs on an x86 Mac, or whatever. And it's why many programs only work with certain versions of OS's and libraries. Full substrate neutrality is mostly a myth---its a goal that we try to approximate, rather than a default state.
Likewise, the computers we build don't really implement the abstractions of traditional computational theory, narrowly conceived. Nobody can actually even build a universal computer, for starters, because nobody has infinite memory. And we don't generally even try build a machines that actually have constant-time memory access, such that traditional algorithmic analyses would directly apply, because they'd have to be be horribly slow. As in any other science, we have certain simplistic abstractions, and we have reality, and when it's clear that the abstractions are way too simplistic---e.g., an algorithm running thousands of times slower because it's paging, or distributed systems where nodes fail unpredictably, we come up with more complex and realistic abstractions.
and it is not a string of symbols from a finite alphabet.
I'm not sure what you mean by that, or why you find it so all-fired significant.
As I understand it, most of the genome is mainly a set of rules using a finite alphabet of symbols. (Although that alphabet can change during evolution.) The shapes of repressor and promoter binding sites act as discrete symbols---a molecule usually either matches or it doesn't---which either do or don't correspond to shapes of parts of other molecules.
Schematically, most genes are regulatory genes, whose products control other genes, that are something roughly like
IF A and B and C and NOT D and NOT E
THEN F and G and H and I.
(This is not an IF/THEN control-flow statement like in a sequential programming language. It's a rule that may fire or not fire---producing its consequent---asynchronously and in parallel with other rules, when the gene is transcribed.)
A, B, C are shapes of promoter binding sites on the left side of the rule (gene), which enable the rule to fire if bound to, D and E are repressor binding sites, which disable the rule if bound to, and F, G, H, and I are shapes of exposed parts of the resulting gene product (usually a protein, after folding), which may bind to other genes' repressor and promoter binding sites.
A regulatory genes act as a kind of analog function unit in a hybrid digital/analog computer. They take concentrations of chemicals as their inputs, and produce concentrations of chemicals as their outputs. The concentrations of chemicals act like "wires" connecting analog electrical function units. (If the concentrations of promoters are high, and the concentrations are low, the rule will fire often because there will often be molecules docked to the promoter sites and not the repressor sites.)
As I understand it, most genes are regulatory, and most of them are binary---the way the rules are written and interact, they tend to fire at a fairly high rate or a negligibly low rate, which counts as on or off for the purposes of controlling other genes. That's done with feedbacks and rate limitations similar to the ones used in any digital computer to use analog parts as digital ones. (That's what makes the genome a hybrid computer program---some of its rules act to compute analog quantities, others to compute binary digital ones.)
Well, yes, but at a rudimentary level, so is my PC. It, the tech support person, and I are part of a larger self-maintaining system that's certainly dissipative. So? How does a thing being embedded in a complicated system that isn't a perpetual motion machine make it not a computer?
I suspect one of us missed the other's point; I'm not sure which.
When I mentioned the inclusion of people talking on the phone in the brain-as-telephone-network metaphor, I explicitly said it'd make it way more interesting even if the people were dumb and only following rote instructions. The point of that was that what makes it interesting in the way I'm talking about doesn't depend on it being a social network of intelligent systems. It only depends on there being some computation---even a little---at the nodes connected by the network. Then you have a massively distributed computer, which was always the real point of an analogy to a communications network, and the analogy doesn't seem so naive after all. As I said, I'm pretty sure nobody smart ever meant the analogy in any other way----it's been obvious all along that the brain isn't just an information routing system, but an interconnected set of processing units.
The point of explaining that is that it's a bad example of the supposed tendency of people to see the brain as like whatever new technology they're familiar with. If that's the kind of "naive analogy" mistake I'm being accused of repeating, it's not particularly embarrassing.
That analogy was actually basically right, and is still basically right---for reasons that should be obvious even to people who call me naive for thinking the brain is like a computer. :-) To think that's a good example of naive everything-looks-like-a-nail thinking is itself naive every-computational-hypothesis-looks-naive thinking.
As for that last quoted bit, above, I have to wonder if you think I'm arguing for something that I'm not actually arguing for. I think the brain is a computer, but I certainly acknowledge that it's an embedded, embodied, situated, and social computer. I don't see a problem with that, for my position.
(I am not defending Kurzweil's predictions or his apparent assumptions that it's easy to emulate a brain, that physical and social development aren't a big hurdle, etc. I do think some of the problems in modeling the genome and brain will be easier than some people here think---I think there's a useful sense in which the design of the brain is mostly in the genome, but there is also a substantial amount that isn't, and even just figuring out its "basic principles of operation" in light of the genetic program will require some difficult reverse engineering that will take decades longer than Kurzweil says. How many decades, I wouldn't dare to guess.)
I guess I'm unclear on the kind of philosophical significance you see here. Certainly it's philosophically significant that the mind/brain evolved, rather than being designed. No doubt about that, and it's one reason I find it interesting.
It's also philosophically interesting, if I'm right, that the genome is a program that computationally controls the construction of the brain, which is itself a computer that computes the mind. For example, it means that much of biology (if done right) is really a kind of computer science (if that's done right). The fact that computer science as an academic discipline doesn't currently emphasize that important kind of computation is a problem, and that's part of my point.
The fact that the brain doesn't have the same hardware/software distinction as a von Neumann machine doesn't strike me as nearly as big an issue. It's very scientifically important, certainly, but I don't think it has much significance for the question of whether the brain is a computer. We draw the hardware/software distinction at a very low level in computers we build, and make it as clean as is practical, but that still mainly seems like an engineering choice, not a difference of great philosophical significance. We do it that way because it's easy for us to do that, given our available manufacturing technology and design processes. If those things were to change, we might go in a very different direction.
Posted by: CJO
|
August 26, 2010 3:50 PM
We draw the hardware/software distinction at a very low level in computers we build, and make it as clean as is practical, but that still mainly seems like an engineering choice, not a difference of great philosophical significance. We do it that way because it's easy for us to do that, given our available manufacturing technology and design processes. If those things were to change, we might go in a very different direction.
I think some of the significance of the brain being an evolved computer (and I'm not uncomfortable calling the interesting stuff the brain does computation) and a designed computer manufactured to specs with a clear hardware/software heirarchy is what I'll call the chaotic economy of nature. In all biological systems, we see a sort of principle of least action in effect such that any time a biological process can "call on" a deep regularity in physics or chemistry, it does so, and there's no "subroutine" in the genome to indicate that the "program" is leaning on external regularities for its execution. (As somebody in this interminable debate pointed out: where does the genome specify the properties of the iron ion bound into hemoglobin? The protein "just knows" what to do with it.)
Development generally is like this start to finish and the development of something as complex as the human CNS is going to have so many, and such dynamic, "subroutines" of this sort, that are not well-understood and are nowhere found specified or even implied in the genome, that it does become a difference of significance in terms of our prospects for reverse engineering the process. Greylander chalks all of this up to "the laws of chemistry and physics" and seems to take it as a leg up on simulation that said laws are "simply stated" that is, not highly complex from an information theoretical standpoint. And I think this is just dead wrong as the example of protein folding that we were batting around for a while shows. The devil is in the details of just how exactly the genome has evolved to take maximum advantage of all the enfolded and subtle complexity arising out of the iteration of "simply stated" laws it can grab "for free."
I mean the whole idea of digital computation is to as much as possible eliminate the noisy and the fuzzy aspects of fine-grained nature, while a biological computer like the brain positively feeds on noise and chaos, and the clear hardware/software distinction we need to maintain to make sense of the various states we want our digital computers to be in to run a program (not just an engineering/manufacturing consideration but a programming one too) is being violated in the brain at all levels, all the time, and in non-obvious ways that will have to be teased out and understood before they can be "reverse engineered".
Posted by: Adam
|
August 26, 2010 4:13 PM
I think Ray could be right- or reasonably close. But I suspect he is wrong in his timelines.
The idea of the Law of accelerated returns may be good. But he probably is missing a lot of the mountains that must be climbed along the way to fully reverse engineering the human brain. There may be so much that must be figured out that he is not aware of yet. So his theory could be right. But it may take a lot longer than he thinks.
Posted by: CJO
|
August 26, 2010 5:01 PM
Grr. first sentence, after parenthetical: "...and not a designed computer..."
Posted by: windy
|
August 26, 2010 5:07 PM
KG:
Of course it doesn't "care" in the literal sense, but the genome is invested in making sure that the environmental conditions are right, in a way that physical environments and light cones aren't.
Even if a bird egg needs a carefully regulated environment to develop, the same environment that successfully hatches a songbird egg will also hatch a cuckoo or a bird flea. The difference between a bird brain and a flea brain can't be due to some incomprehensible flow of information from the environment.
A lot of the relevant environmental inputs are provided by the genome indirectly - as John pointed out in #814. Also the fact that the genome finds itself in the right environment "often enough" is not random either.
Posted by: David Marjanović
|
August 26, 2010 6:00 PM
Wow, Paul W., that was an eleven-screener...
(On a laptop. But still.)
No, the iron ion reacts with the hemoglobin in the only way the laws of physics will let it; and natural selection has preserved those genomes that code for hemoglobin molecules which will react with iron(II) that way. No however metaphorical knowing involved – stuff just happens the only way it can.
Posted by: scramton
|
August 27, 2010 1:35 PM
Kurzweil thinks that computers will be so much smarter than our puny brains but we need to program these computers to model the human brain so this model can develop and be smarter than us? Does this guy know anything about computer programming?
Posted by: adrenogg
|
August 27, 2010 1:46 PM
I once rented Kurzweil's book on the singularity from the library. God damn it was boring.
I'd still like to be able to see the day when we can upload a human mind into a machine, but I'm not holding my breath. Maybe I'll settle for the slightly less grandiose task of waiting for cell senescence to be removed from the human genome.
Kurzweil always seemed hyper optimistic to me. Maybe the reason he's making such dumb claims is because he's reaching his twilight years and desperately hopes that saying it out loud will speed its progress so he can download himself on his deathbed.
Posted by: Paul W., OM
|
August 27, 2010 1:59 PM
Yes, Kurzweil knows some stuff about computer programming---enough to build some pretty awesome state-of-the-art software systems. (That doesn't mean that he knows what he needs to know to support the very optimistic claims he makes.)
And on this basic point---modeling the brain, understanding it, and then surpassing it---I think he may well be right. We can learn a lot from the brain, and once we do, we may well eventually be able to make something smarter. Evolution has obviously discovered some pretty cool stuff, but presumably missed some pretty cool stuff too.
I don't think it's going to be as easy to model or understand the genome or the brain as he makes it out---not by decades in each case---but I don't think he's entirely wrong on the above point, or the point that the genome contains a lot of information that's likely to be very useful in understanding brain development and structure, and (indirectly) the basic principles of its functioning.
(And that, in turn, is a far cry from emulating a particular brain so precisely as to embody a particular human's mind in it, if that makes any sense at all anyhow.)
Posted by: Paul W., OM
|
August 28, 2010 10:57 AM
(Uh oh, here's another long one. Sorry, David.)
CJO (et al.),
I'm not nearly as optimistic about this stuff as Greylander---reverse engineering of complicated stuff is never as easy as Kurzweil seems to imply, and most of it can't be done just by looking at the low-level code and magically understanding why it works.
On the other hand, I'm apparently more optimistic than some people here, because the genome just doesn't seem that weird to me. I've seen analog and hybrid circuits before, and asynchronous parallel and distributed systems, and it all looks basically familiar to me, so far as I currently understand it. If we map out all the connections between the rules---the matches between repressor and promoter binding site shapes and the pieces of transcription products that dock to them---that will be an enormous step forward. It'll give us a rough draft of the genetic program as a recognizable program, rather than as a string of uninterpreted symbols like AGCCTGAG..., or even whole molecule shapes. (I'm optimistic that this can mostly be done semi-automatically, without actually solving the computational protein folding problem, and the fact that there are only 20,000 coding genes is grounds for optimism that the "disassembled" code will not be too horrendously complicated to mostly understand within a few decades.)
One reason I'm optimistic is that the genetic regulatory networks we have figured out don't seem particularly bizarre and incomprehensible. (That might be an artifact of them being the ones we have figured out, but I don't think it's just that. I think it's mainly that evolution doesn't generally program the GRNs in astonishing incomprehensible ways that computer scientists can't recognize.)
Hox genes, for example, don't seem particularly surprising, in programming terms, given what we knew about bauplans, the basic idea of regulatory genes as rules in a forward-chaining production system, and the fact that evolution often kludges things together by cutting, pasting, tweaking, rather than doing them elegantly.
Hox genes implement a serial overall process within a system that's parallel and asynchronous by default, using very crude and ugly synchronization---a set of global one-bit flag variables, and kludged-up rules that roughly approximate a sane phase- and region-numbering system.
Some people seem to look at that and say "wow! that's nothing like a program!"
My reaction is exactly the opposite---what the heck did you expect? It looks exactly like a program written by a crummy programmer that just manages to get the job done. I've seen plenty of code like that.
The use of global variables to as synchronization variables to make a serial program is exactly what everybody does when they program asynchronous production systems to do things in a series of phases, and can afford the serial bottlenecks it introduces. Its the easy, dead obvious way to do it. The kludging with more and more flags is what you get if you're adding features incrementally, under deadline, and are resistant to reprogramming anything that already works. It's the kind of kludgey crud that beginning programmers tend to do---and even good programmers do if they're just being paid to add a feature, and don't have time to do it right, or don't have a long-term interest in the project. (Sometimes with a comment like "this really ought to be redone the obvious general way, but I don't have time, so I'll just special-case this for now.")
Hox genes are the kind of code I'd expect beginning programmers to write if we started them out programing parallel production systems, before they learned about indexed loops and recursion. (Or if those things were ruled out by design constraints, e.g., if we're going to translate the program into a combinational logic circuit in hardware.)
If most of the code in the genome is like that, it should be annoying and tedious to figure out, but not pose big deep problems because of an "alien" computing model. To me, at least, it looks familiar overall. (Sometimes painfully so.)
There is clearly some stuff going on with developmental programs that is much more elegant and spiffy, but my impression is that it's mostly the kind of stuff I'd recognize from computer science, not something really mysterious, weird, and alien.
For example, in biology you sometimes get recursive-fractal like structures without any recursive "subroutine calls," by taking advantage of the fact that you've got a big distributed system---each cell is running its own copy of the program and they're arrayed in three-dimensional space. (Or two-dimensional sheets in 3-space.) Sometimes that's done top-down, but it's often done bottom-up, with structure emerging at a fine grain first, then lumping together at larger and larger scales.
In light of computer science, that's not at all surprising or unfamiliar. If you program a massively parallel mesh-connected computer, you use similar techniques because they're simple and effective. You can farm tasks and/or data out to a bunch of processors in a hierarchical, top-down way by tossing a starting set across the mesh, maybe randomly, having those nodes toss subsidiary data/processes around less far, and so on, iteratively, until you hit a fine scale and stop because all your processors have work to do.
(Sometimes you do that sort of thing randomly, on purpose, to probabilistically ensure that no matter what patterns are in the data you're processing, they're unlikely to tweak the worst case of a rigid pattern of your data processing. We often randomize algorithms, and take advantage of the randomness, to make them more robust---it's not only biology that does that. Randomized algorithms are especially common in distributed systems, e.g. the basic Ethernet protocol uses randomized delays to resolve packet collisions---when two computers try to "speak" at the same time on the same wire, and garble each other, they both notice, wait a short random time, and try again. The randomness makes it very unlikely that they'll try to "speak" at the same time next time around, and that simple trick scales to many processors.)
You can do the same sort of hierarchy-building bottom-up, nondeterministically and probabilistically, by having each processor generate a pseudo-random number, and then applying a bottom-up parallel clustering algorithm---each node with the randomly right kind of value connects itself to some random nearby nodes to make a little cluster; the others don't. Then you iterate, connecting larger and larger clusters, until everything you want connected is connected in a randomized fractal-like way.
For that kind of algorithm, there typically isn't any explicitly recursive code. The recursive structure emerges from some simple local conditionals and the fact that the processors are connected together in a space-like connection topology.
That's another kind of thing that many people see in biology and think "wow, that's nothing like a computer," but somebody who's familiar with analogous situations in computer science would react oppositely to---"what did you expect? How else would you expect to hierarchically distribute processes or data among spatially arrayed processes?"
That is a basic way in which computers and biology are similar for deeply similar reasons. Parallel computers have to be arrayed in space somehow, and communications with nearby processors are generally cheaper and faster than communications with more distant processors, for physical, geometric reasons.
The fact that you're dealing with space imposes constraints on reasonable implementation strategies for distributed hierarchical structures, and also gives you a lot of "structure for free"---e.g., getting recursion from the interaction of nonrecursive local code in a spatially arrayed and connected network. It's not just biology that takes advantage of such things---we have to, too.
The fact that one computer is made of silicon and wires operating digitally and the other is made out of cell communicating with chemicals in a hybrid fashion is less important than the fact that they're both solving deeply similar problems within deeply similar constraints---e.g., a bunch of processing elements arrayed in a low-dimensonal space, which have to be coordinated. There are only a few basic ways to do it, top-down or bottom-up, and a few basic variants of those. (E.g., randomized or not, synchronous or not.)
In both cases, there's usually a simplicity constraint as well. In digital computer design and programming, we often have to keep things simple because simple hardware is faster---even if we could design something in a more complicated way and get it to work, the simple solution is often best, e.g., simple local routing done by screamingly fast hardware making local decisions, vs. fancier distributed algorithms that cost more in CPU time or hardware or communication costs than they save.
Even the physical growth of the computer just doesn't seem that weird to me, and it seems to me that people who think it's fundamentally weird are missing a central point of computer science.
Many people who think about, say, a mesh-connected parallel processor will only notice that it's a fixed array of fixed hardware connected in fixed ways in space. In contrast, biological cells reproduce and develop, growing tissues and organs that fill space in a richly structured way.
But whole the point of that fixed array of processors and the clean hardware/software distinction is that we can do fundamentally the same sort of thing, virtually, in a computer.
A subtler point is that the way we do that is ultimately limited by physics. Many of the same physical constraints on physical systems do show up, indirectly but clearly, in the virtual systems we build in computers.
Where biology might expand a network of processors (cells) until it fills a certain physical space, we would expand a virtual processor network until it fills the processor space. Sure, we're doing it the easy way, but the basic structures of problems and solutions is often fundamentally similar---the ways that we've made it easy usually don't profoundly change the things that make it scientifically interesting.
So, for example, an organism may grow a network of cells in actual space until it gets to be "the right size," where that size is ultimately constrained by what it can afford given some fitness issue. (E.g., the size of a brain that's smart enough, but not too expensive metabolically, and not too slow because nerve impulses can't cross it fast enough.)
I may grow a computational entity in a similar way, spawning and distributing subprocesses and data structures within the constraints of my spatial array of CPUs, until it reaches an analogous fitness constraint---e.g., how many processors I can afford, or how many I can pack closely without them being too close and overheating, or the network being physically too big and the communication delays through the wires getting too long.
The fact that one computer does stuff in hardware that the other does in software doesn't change the essential fact that they're using very similar engineering solutions to deeply similar engineering problems---e.g., distributing computations through the right size volume of space, minimizing expensive long-distance communications, developing a hierarchical structure and a mix of distributed and hierarchical control mechanisms, etc., etc., etc.
Almost every time somebody gives an example of how the genome is supposedly unlike a program, and biology is so different from computer science, I have that kind of reaction: computer scientists solve basically similar problems in basically similar ways, for basically similar reasons, every day. We may not have encountered the same exact combinations of weird features, at the same levels of abstraction, but we've encountered most of the "weird" features in various weird combinations, at various levels, so it's just not all that weird, really. It's basically familiar engineering.
Posted by: Paul W., OM
|
August 28, 2010 11:54 AM
I have some biotech questions relevant to how optimistic I should be about mapping out genetic regulatory networks mostly-automatically.
The basic idea I have in mind is to figure out which genes' transcription products match which other genes repressor and promoter binding sites.
I would think that you could mostly do that by brute force, given a big research project.
If you can identify most of the genes and most of their repressors and promoters and transcription products---how hard is that to do mostly automatically?---it seems to me that you should be able to do most of this by brute force, given a fairly big budget.
What you would do is make a bunch of types of chips resembling gene chips, where each type of chip has, say, 1000 little areas with 1000 different repressors and promoter sites. By bathing the chip in a solution of a given transcription product, you could find out which of a thousand things it can bind to. That would let you map out the basic rules in the genetic production system.
If we have 20,000 genes with, say, an average of 20 repressors or promoters each---is that a reasonable guess?---that's 400,000 sites to test each transcription product against. If each chip can test against 1000 sites, we need 400 kinds of chip, and we need as many chips as we have transcription products. By making arrays of one each of the 400 kinds of chips, we could test a given protein solution against all the binding sites in a single run, and read the results off the chips.
I'm assuming that there would be a variety of limitations on this, and we'd only be building a rough draft of the overall genetic regulatory network, at first, to be refined with other techniques.
What I would hope you could mainly do is avoid having to solve the protein folding problem by mostly using folded proteins---e.g., extracting a slew of proteins from a bunch of actual tissue of sacrificed and blenderized animals, separating them out, and using the resulting bunch of solutions.
The first gotcha I see there is whether you can mostly extract folded proteins mostly intact, with mostly the right binding site "keys" exposed, in sufficient quantities for the experiments. I wouldn't think you'd need a lot for any particular run---a very weak solution would work with the chips if you let it sit long enough that enough of the molecules bounce around to the binding sites and bind.
Of course this would get pretty hairy---e.g., you're not going to find proteins that aren't expressed in the animals you sacrifice, and separating them out and identifying might be a big, very tricky project in itself. You'd have to sacrifice animals in various states of development, health, etc., to find most of the more unusual ones.
A side benefit of doing that, if it's feasible to do at all, is that along the way you could also find out a lot about which proteins are commonly expressed in which tissues at which rough stages of development---e.g., separate livers, hearts, brains, etc. from early fetuses, late fetuses, young animals, old animals, etc.
One basic question I have is whether most signaling proteins would more or less maintain their shapes during extraction, if you're gentle with them and keep the solutions at reasonable temperature, PH, salinity, etc.
Another is whether I'm right that you could mostly identify which genes those proteins were generated by automatically, with some quick and dirty shotgun sequencing.
Any thoughts? Is this something biologists are thinking about doing? (It seems like the logical successor to the Human Genome Project.)
Is it a stupid idea that's been decisively shot down already?
Posted by: Daniel B
|
August 28, 2010 12:34 PM
KG
Despite the my best judgement, I'm going to assume that you aren't just trolling and that you earnestly believe what you are writing. Since you clearly don't understand what you are criticising I'll endeavour to to provide you with something of a 'singularity for dummies' which in this case will be literally accurate.
The fundamental assumption in Ray Kurweil's notion of the singularity is that computing power will continue to grow exponentially as it continues to do now. He believes it is doubling every year while others may suggest 18 months or even 2 years as the doubling period. So in his version, after 10 years you have 2 ^ 10 (about 1000 times) more computing power available. In 20 years you've got 1 million times, 30 years a billion times an so on. If you adopted a doubling period of 2 years then you'd have to double those time periods.
So, following from above, in 20-40 years time, a simulation that now takes 24 hours would be run in less than a tenth of a second. I hope that even you can imagine this would speed progress up considerably as well as motivating a significant change in the way you did things. Where previously you might have laboured over the working out the best values for particular parameters in your model, now you could just have the computer re-run the simulation over a large range of possibilities. In many cases it would be more efficient to use this 'brute force' method than trying to work out 'why'.
Of course just getting the computer to randomly try variable combinations, while powerful, is still sub-optimal. If you have a measure of 'goodness of fit' then it would be possible to utilise some sort of goal seeking algorithm (AI) in the variable selection process. Even now we have such methods. The more we utilise and rely on these AI algorithms (initially to take care of the 'mundane details') the more this would be expected to spur development of improved algorithms. In addition to this, the extra computer power should allow us to use algorithms that previously would have been too slow. Finally we will almost certainly be using the previous generation of AI to help us develop the next one which should allow the process to 'bootstrap' itself forward.
Ahh. A government funded worker are we? No wonder you find it so hard and disagreeable to think outside of the narrow constraints of your domain. I'd expect development in the area to be funded privately, for profit, by one of those gullible computer science guys like Bill Gates or Sergey Brin.
Hang on. This hypothetical scenario was a precondition for most what we've been arguing about. By seeking to negate it you are essentially conceding the point. I don't actually think our first 'human like' AI will come via this 'artificial embryo' route. It would be a lot harder than copying a brain in my view. My point was that providing artificial input to an artificial brain is far easier than both of these. We are almost able to provide artificial sight and hearing to real brains.
It wouldn't have to 'be a robot'. Children can and do grow mentally with a wide range of incapacitating disabilities. Even if you chose to use an artificial body there would be no necessity for the actual 'brain' to run within it. But the most efficient solution would be an artificial body in a simulated environment. We can already make 3d environments with reasonable physics and in 20-50 years this would improve dramatically.
Sadly many children grow up severely neglected and in some cases abused. The fact that some of them can make it to become functionally capable adults is a testament to the resilience of the algorithms that underlie our human intelligence. It is difficult to imagine that a team of experts and surrogate parents with the unlimited capacity to back up and modify the environment would not be able to do a far superior job.
Again this is outside the scope of our original discussion. But do you really think the precise details of the world outside the mother's womb matter enough make difference to make the difference between development working or not? If it was really needed you could always implant a sensory device inside a surrogate mother and use that data to provide sensory input to the developing artificial embryo.
Ideally yes. Fortunately all of these could be provided via artificial sensory input.
You just don't get how this combines with the doubling of computer power do you? Let's say the first human like AI can run in real time on computing hardware costing 100 million dollars (after consuming possibly billions in development effort). Let's take a look at what happens in the following decades after than
10-20 years
- Can run one AI instance at 1000 times normal speed on $100M hardware. They could complete 3 years equivalent work in just over a day. (Although they'd be pretty short on external input from the slow outside world)
- Can run a team of 100 AIs at 10x speed on $100M hardware.
- Can run 1 AI in real time on 10K computer. That already undercuts pretty much every first world worker.
20-40 years
- Team of 1000 AIs can operate 1000x speed on $100M hardware.
- 'Normal time' team of 100 AI's working on $1K computer hardware.
30-60 years
- 1 billion 'normal time' AIs on $100M hardware
- Team of 1000 AIs running at 100x on your $1K computer.
The difference between these AIs and real people is that once an expert is produced in an area they could be copied indefinitely.
Have no doubt that the creation of artificial intelligence coupled with increasing computing power would be the single biggest and most important development in our history if it could be achieved.
Posted by: Praedor
|
August 28, 2010 5:06 PM
One need not mimic, down to the ion level or protein level, the human brain (or any other brain for that matter) to simulate a human brain. All one needs is enough understanding of how the overall brain works, the circuits, the patterns. Now even detail that goes that far is not soon to be at hand but it is certainly closer than actually understanding the minute details of HOW it is accomplished.
Be that as it may, I will borrow from Fermi's Paradox to squelch any hope of self-assembler/universal assembler nanobots saving the day for living to eternity: if such replicators were in fact possible we would already have encountered them as this would be the easiest/fastest method of colonizing a galaxy. All it takes is one advanced civilization coming up with such replicators to start a very rapid colonization of the entire galaxy. They would arrive somewhere suitable and simply start assembling appropriate shelters and then start assembling alien explorers or even robotic alien explorers. Hasn't happened thus it cannot happen - possibly because everyone who devises the tech necessary quickly succumbs to grey goo catastrophe. One mistake or one miscreant is all it takes to set off a grey goo apocalypse (quite aside from the point that anyone/everyone could quickly build themselves nukes...with ensuing hilarity as a result).
The other issue I have with the whole idea of "uploading"...so what? I live inside my body. My consciousness resides within my body. I CARE what happens to my body. Simply transferring a copy, no matter how good, to a machine platform doesn't change the fact that I don't want to die. It's not as if you will simply fade out of your current body and fade into your new artificial body and continue all happy. Your original, your biological body person is still there shaking it's fist at the universe for its mortality. It doesn't do me jack squat good for my identical twin self to go galavanting about the galaxy to have fun for perpetuity...what about ME? I'm still here!
Posted by: amphiox
|
September 3, 2010 4:52 AM
There's a lesson here, somewhere, methinks.
Posted by: KG
|
September 3, 2010 11:04 AM
Daniel B.,
You're a moron, as this:
shows clearly enough. Whichever sector you consider, anyone proposing that large amounts be spent on a research project will need to show that their approach is likely to work, and "prove it won't" will not cut it. Moreover, only a moron thinks you can judge someone's intellectual range by whether they work in the public or private sector. FYI, I specialise in interdisciplinary research. I've worked in departments of engineering, computer science, transport studies, zoology and land use systems. I've published in journals of AI, theoretical biology, cybernetics, artificial life, philosophy, land use science, complex systems, economics, geoinformatics, and human ecology. I currently head an international project involving computer simulation, psychology, sociology, economics, and political science. What was that about the "narrow constraints" of my domain?
You again show you simply don't know what you are talking about. Sure, it's nice to be able to cover a search space of parameters quickly. But once you have your n simulation outputs, you need to understand why you got the pattern of results you did before you can intelligently decide the next parameter space to look at. (It's the need for intelligence you are failing to see, for rather obvious reasons.) Potentially, that will involve comparing arbitrarily many sets of subsets of your set of simulation outputs. Even you should be able to grasp that the size of the task could scale as a double exponential in n, as there are 2n subsets of outputs, and 22n sets of these subsets.
Posted by: seth edenbaum
|
September 4, 2010 10:07 PM
Create an indecisive computer, torn between the imperatives of logical calculation and reflex (conditioned response) and you'll have a man made simulation of animal "intelligence."
Consciousness is not complex calculation, it's anxiety. Only life forms are afraid. It amazes me how science fantasists ignore the obvious.
Posted by: alpha19
|
September 6, 2010 10:38 PM
PZ, you are a pompous and negative soul. You should be embarrassed about how you try to discredit Ray through insults as he is clearly smarter and more accomplished then you are and does not resort to your immature name callings. Hope you learned your lesson on using misinformation and not even having the courtesy to check your facts before you try to make a name for yourself by discrediting someone elses work. Contribute instead of destroy. Namaste
Posted by: steve.nordquist
|
September 6, 2010 11:03 PM
Whiner! Does Sejnowski set your schedule autocratically when this stuff happens (Gnosis in 10 years or you're culinary staff! Bring the deliverables schedule!), or do you hate enthusiasm and first-comers with tools canny enough that the claims are not far out?
This sort of thing is useful for racking up what should be done with brainbows and other things that kind of took 8 years to turn the crank on (and which throw your awful treadmill, of having protein deathmatches of all sorts to the 800k-mer limit, in the empirical bin.)
If you want a different curve, you get to expose your own Zeno limit; environmental constraints forcing it to be 4 times as long as it originally took for eukaryotes to evolve to intentionally brew Trappist Ales, say.
He never claimed it would be literate code (who wants the notes on the nodes responsible for poorly directed attention, why, and the use cases?), that it would be cranked out monotonically at 100,000 lines a year, or that you'd get perfect numeric proteome results (down to cellphone I/F, say.)
Posted by: john_frink
|
September 8, 2010 7:58 PM
The question is:
Does one have to know every detail about protein interactions and the whole bio-molecular machinery to figure out how genome sequences correspond to brain structure?
The answer would definitely be YES IF you were a space alien who has no information at all about life on earth and you were asked what kind of an organism can be created from a genome.
But we are not space aliens, we know what final product looks like. We can cheat.
For example, if I see some cool design trick on this page, do I have to know how transistors dispatch electrons inside my computer? No, that would be total waste of time. Simply looking into HTML code would give me an answer.
Posted by: somedude
|
September 11, 2010 9:20 PM
From Kurzweil's response:
I mentioned the genome in a completely different context. I presented a number of arguments as to why the design of the brain is not as complex as some theorists have advocated. This is to respond to the notion that it would require trillions of lines of code to create a comparable system. The argument from the amount of information in the genome is one of several such arguments.
If this is indeed what Kurzweil (a person I have never heard of before today, same as the author of this article, just in case anyone wants to accuse me of being on any side) said then it seems the only claim that he made is that it is possible to represent the brain in less information than the total of the "finished brain". I see nowherethat this representation is all that is required.
Posted by: Daniel B
|
September 13, 2010 9:04 AM
KG,
Actually I wasn't making a statement about all public sector workers. I was merely declaring it as one of likely explanations for your limited intellectual view. The notion that the truth value of an argument should be judged by the chances of it being accepted for government funding show a distinctly myopic perspective.
After essentially conceding defeat by abandoning any semblance of the original argument I would have thought you might be ready to start blaming this whole thing on a teenage son (or a pet) who was using your account without permission. Oh well at least your honesty is refreshing. I'm just a humble web programmer/manager/owner who still remembers something about information theory and algorithms from uni days.
You are equating 'understand' with 'intelligently decide'. Depending on one's definitions of the words that's either a tautology or statement without any evidence. Without a metric to measure the accuracy of the model everything seems pretty subjective (unless, as I suspect, your main metric is how to convince the funding body to continue supporting you).
Of god. It's worse than that. What if you had to consider all the possible subsets of the subsets or what if there where multiple independent grouping of subsets that had to be considered? I think the true compexity may be closer to 2∞n !!! I think we managed to identify an entire class of complexity previously unknown to the information theory experts.
...Seriously though, this process of looking at subsets is almost certainly an algorithm you are using to reduce the complexity of the process and to extract pieces that your limited human brain can reason with (in this case I'll withhold the obvious insult). I'd also wager that if you were dealing with 'n' being in the thousands or millions you'd be forced to use some sort of statistical or pattern matching to identify these subsets at all. For very complex problems with large 'n' the only viable use of human intelligence is to decide on which algorithm to use, develop a new algorithm or to change the scope of the original problem.
Posted by: David Marjanović
|
September 13, 2010 9:35 AM
This is an argument from astounding ignorance. You don't even know how research grant proposals get accepted or rejected by not-for-profit organizations (whether funded by a country, like the NSF or the Humboldt Foundation, or not), do you? It's by peer review. The proposal is sent to scientists who work in the field in question, and they are asked to write assessments (for free, BTW) on whether it should be funded and why.
Have you ever written a manuscript for a scientific journal, or a grant proposal?
Posted by: David Marjanović
|
September 13, 2010 9:38 AM
...but not necessarily in the same country. Many agencies even require that the reviewers work abroad. The idea is to avoid conflicts of interest over funding.
Posted by: homunq
|
October 17, 2010 2:23 PM
The fact that the genome is size X is, in fact, evidence that the smallest possible program which, if left running long enough with enough memory, would end up simulating a human brain, is not too much larger than X. However, the complete laws of physics are probably much smaller than X, so we already knew that.
Is it possible to abstract away the messy squishy parts, and get a program of that size which would not only work eventually, but also not be googolplexes off from Greylander's memory and speed requirement estimates? Not proven, but possible.
Is it possible for humans to figure out how to do this? Certainly not by a bunch of differential equations. Unless you understand why it has to work the way it works, you are going to get too many things wrong to ever fix.
But lets say that yes, all these problems are solved. Amazing breakthroughs, and they're solved in 20 years. And then those brilliant scientists of the future look back over those breakthroughs, and they see Kurzweil predicting it all.
And they say, "Sheesh, what a dweeb." Even they would think that, with all the unforeseeable hurdles that they had overcome, Kurzweil and Greylander's confident predictions and use of data are statistically indistinguishable from wild-ass-guesses; and that even assuming (which I don't) that K+G have an eternal exponential Moore's law on their side, their level of confidence is totally unjustified.
Separately:
Greylander: So, I build up my super-neural-simulator with my best estimates of all the varied synapses and cells of a brain and... it doesn't work. But I'm pretty sure I have all the coupling constants correct to within +/- 2%! What do I do next? Which constants should I recheck?
In other words, how is the debugging step not equivalent to the original task of UNDERSTANDING a human brain? The super-neural-simulator would be an amazing tool for doing science - but there would still be decades of science left to do before you'd be talking to it.
This is not an argument that reductionism doesn't work on a brain. It's just saying that the people who are actually doing that reductionism are the ones I trust to say how hard it is. And it's not a 10-year task, or a 20-year task.
Posted by: KonradZeusNeumann
|
October 21, 2010 8:48 AM
Good luck PZ Myers with your political move.
I wouldn't be saying anything if my intuition wasn't telling me PZ Myers's motives are rotten.
Truly rotten.
Truly good luck with your low brow science .. blog opinion. We know what you are.
There are many thoughts I could express in argument of Ray Kurzweil's belief of AI but you PZ Myers have said none of them. You reference nothing of statistical fuzzy logic described by Von Neumann or the human nervous system. You know nothing of the study of arithmetic and mind.
Nothing can stop the growth of knowledge now.
Nothing can stop technology.
Evolution is the rendering of strength.
The future is for the brave.
"not in wilderness .. but wildness is the preservation of the world."
Posted by: mick.long
|
November 4, 2010 4:39 PM
@ KonradZeusNeumann
"Nothing can stop the growth of knowledge now.
Nothing can stop technology.
Evolution is the rendering of strength.
The future is for the brave.
"not in wilderness .. but wildness is the preservation of the world."
You forgot "The Rapture is coming!" :D
Posted by: David Marjanović
|
November 4, 2010 5:12 PM
Er, no, because you forgot to simulate the food & water and the oxygen and the egg cell and the point where the sperm cell enters and so on.
Evolution is descent with heritable modification.
Neither Kurzweil nor science support that kind of ideology.
Posted by: Silič O'Nopolitanopoulos, Färschdbischuf Beesknees aus Ulm und Klein Elguth, Elector Pharynguline.
|
November 4, 2010 5:27 PM
ORLY?Posted by: Dimi
|
December 4, 2010 4:08 AM
I don't see a problem. Maybe to understand how the brain works you should simply stop using the brain while trying to understand it. What we are is not just a physical body containing a single component called brain.
Posted by: lordsetar
|
December 4, 2010 4:46 AM
Dimi #843:
That's slightly impossible considering what the brain's function is. Well, the body does have other components than the brain, yes. We call those "systems" and "organs" and "tissues". However, if you're trying to say that there is some sort of nonphysical component to the body then I say [citation needed].
Posted by: Zupecki
|
December 27, 2010 2:52 AM
I'm no neuro scientist, although I am someone very interested in all facets of life.
I have two issues with this article;
1) Subjectively laced comments like 'kook'. I think someone who claims to have your stature should be able to remain slightly more composed and reasoned with his descriptions of, what you see to be, another persons intellectual fallabilities.
2) Ray Kurzweil said we will have reached the 'Computational Capacity' of the Human Brain by around 2020 - not that we will have reversed engineered it by then. He mentions 2029 as a staple date for this prediction.
So it's 19 years on his watch, not 10.
Just sayin'.
Thanks for the article, though, despite some of the terminology being outside my apparent frame of reference.
I believe Ray's optimism comes from the exponential nature of our progress - using technology to create technology on ever shrinking time scales (x in y=1/x).
I mean, just look at the time frame of the Human Genome Project - it was hugely overestimated and the intial, seemingly linear progression wasn't looking promising. But it broke that trend soon after.
I don't pretend to know whether Ray is wrong, or right, I just wanted to point out the above statements.
Thank you.
Posted by: John Morales
|
December 27, 2010 3:04 AM
Your (subjective) opinion is noted.
I also note you don't use a single term to summarise your opinion of PZ's opinions, but merely think that someone you imagine has claimed a certain "stature"¹ should "remain slightly more composed and reasoned"².
--
BTW, have you actually read the comments?
--
¹ Can you provide a citation of where PZ has claimed this purported "certain stature"?
² What is it in the OP that you consider discomposed or unreasoned?
Posted by: Zupecki
|
December 27, 2010 3:46 AM
No need to use a self righteous tone, friend :)
Opinion's are subjective, so no need to resort to sarcasm by tautologically stating in the parentheses.
Also, last time I checked, it's not a requirement of a forum to use a 'single term' to summarise my opinion of the subject matter. It is for comments, which has a loose definition.
I am not going to give my opinion, since I already stipulated that sections of the subject matter are absent from my frame of reference - and opinion does nothing to further any credible solution.
--
¹ Can you provide a citation of where PZ has claimed this purported "certain stature"?
Yes I can - "I'm a developmental neuroscientist."
Please do not take my use of the word 'stature' as an interpretation of my claiming he is arrogant. Stature merely means - "Importance or reputation gained by ability or achievement."
Since he is scientifically minded, I would just appreciate a little more respect portayed by use of language when denouncing another person - name calling never did anything constructive.
AND
"I'll make one more prediction. The media will not end their infatuation with this pseudo-scientific dingbat, Kurzweil, no matter how uninformed and ridiculous his claims get."
Present information in rebuttle, sure, but this isn't constructive. That is biased, subjective and unreasoned. Reasoning would be along the lines of using coherent points to deduce why Ray may think the way he does - PZ does a good job of presenting counter information... except where Ray's own mind is concerned.
That's all and I have read the bulk of the comments - but I cannot retain an ocean since the computational capacity of my brain is not wired to do so.
Posted by: John Morales
|
December 27, 2010 4:14 AM
Zupecki, why do you conflate opinionated with self-righteous? :)
We are both expressing our opinions, no more.
Too late. You already have.
For example, I know you would appreciate a little more respect (as you understand such things).
Hm. Well, since you're not a developmental neuroscientist, how do you know that PZ is not being sufficiently composed and reasoned in his adumbration of Kurzweil's basis for prognostication as it relates to neural development?
Specifically, how do you know Kurzweil's claims regarding the brain's development and operation aren't kooky?
Being mealy-mouthed is hardly constructive, either. Whatever field of expertise you possess, if someone makes a kooky claim in relation to that, do you shy away from so stating?
If so, why?
But this is not the first post PZ has made in relation to Kurzweil. Why should PZ recapitulate all his earlier contentions and rebuttals?
So you've at least seen the discussion relating to the timeframe, right?
As to your own brain's limitations, I don't disbelieve you.
Posted by: Zupecki
|
December 27, 2010 4:50 AM
Self righteousness and opinionated natures are somewhat synonymous.
Please read where I have placed certain points contextually - you told me I have not used a single statement that summarizes my opinion of PZ's opinions.
So I said I am not going to give my opinion (on that, specifically - as you already stated you felt I had not in your previous question) as I am not qualified to make a meaningful rebuttle at this point in time.
Surely you can respect the notion of my not wanting to dive into something that I am not wholly educated within.
--
"Hm. Well, since you're not a developmental neuroscientist, how do you know that PZ is not being sufficiently composed and reasoned in his adumbration of Kurzweil's basis for prognostication as it relates to neural development?"
I was referring to his treatment of Ray's character - I already stated that PZ does a good job of presenting counter arguments, where neural development is concerned. So I'm not questioning PZ's accuracy at all.
--
"Specifically, how do you know Kurzweil's claims regarding the brain's development and operation aren't kooky?"
I don't know - but that's not what this is about. This about the use of language when writing an informative article. If is important to not let personal opinion's creep in, this is good practice... or it can detract from the credibility in the readers mind.
--
"Being mealy-mouthed is hardly constructive, either. Whatever field of expertise you possess, if someone makes a kooky claim in relation to that, do you shy away from so stating?
If so, why?"
I don't shy away from stating that they're a kook - I don't state that at all. I would attempt to deconstruct their views/opinions/information in a logical fashion while also attempting to wield their own understanding back onto themselves in a strategic fashion that may allow them to see their fallability.
I don't call people dingbats, kooks or any other manner of slang names that result in the emotional tarnishing of the subject at hand.
The issue is that PZ believes that Ray is either not completely informed, isn't understanding, or underestimating the seemingly insurmountable task - that does not require scolding of any kind towards his character. It requires personal analysis of Ray's brain, himself, and of his motives and understandings to deduce whether he is genuine in his thoughts - of which, if he is incorrect, he is a victim of his own mind. Or whether he is a con artist trying to seek fame as another "Deepak Chopra" - without providing a reasoned argument for his opinion on Ray's character, it is no more than opinion - and opinions are meaningless.
The information presented, that counters Ray's predictions, is good - I have stated that.
I have seen the discussion relating to the time frame - the article says "within a decade", though, and I was making the point that is when he predicts hardware will reach that potential, not the software to mimic emergent learning as complex as human intelligence, or unintelligence if you look at how we communicate and deal with one another.
And my own brain's limitations are that of your own - I am not malformed in any way... so aside from neuron count, we can only gauge comparative efficiency in terms of the net we have nurtered - which all depends one the lives we live.
We are all limited by our biology and our language is extremely fallible.
I am not refuting anything PZ has said, nor supporting it as I don't have the required information to do either - but I will, of course, take what I can in consideration.
It's just the tone and references to Ray's character that I feel detract from the issue at hand.
Let's not continue debating over something that will seemingly continue until the Universe freezes over.
Posted by: John Morales
|
December 27, 2010 5:02 AM
Zupecki, I draw your attention to the HTML commands listed beneath the comment box.
And, furthermore, you decry those who do, though you have no genuine basis for informed adjudication of the merit of their rationale.
Let us not. Amen.
--
PS Welcome to Pharyngula!
Really.
Posted by: Zupecki
|
December 27, 2010 5:13 AM
Sorry, I was being lazy.
There is no true rationale that would result in 'putting someone down' - that is an emotional action and, as you well know, emotions are generally illogical.
And I do not need to know the subject matter in order to adjudicate bias.
Just saying there is no reason for calling him a dingbat... or do you think that is an appropriate title?
If you support putting people down, then good luck with that. I am not trying to offend PZ, and you seem defensive of him from every angle - he is human, and he isn't perfect.
Thanks for the welcome!
Posted by: John Morales
|
December 27, 2010 5:41 AM
Zupecki, no worries, and may I say you might like it here, if you can overcome the tone. :)
Note that this is PZ's soapbox; please don't get confused by the domain name, he's just expressing his opinion, and in return you're welcome (as are all of us commenters) to express yours.
A free-speech bastion, this is.
--
PS Kurzweil posts here.
Posted by: Zupecki
|
December 27, 2010 5:50 AM
Thanks man - glad we reached this point :)
Posted by: David Marjanović
|
December 27, 2010 6:14 AM
Because he is a scientist, PZ uses "kook", "dingbat", and especially "pseudoscientific" as technical terms with definitions. He notes that Kurzweil fulfills these definitions, that is all.
And since when has respect had anything to do with science? Science is about falsification and parsimony, not about respect. It's about what is said, not about how it's said.
Posted by: bender174
|
October 14, 2011 4:11 PM
Haters gon hate
http://27.media.tumblr.com/tumblr_lqikp0JnFF1qz9dz7o1_400.png