Oy, singularitarians. Chris Hallquist has a post up about the brain uploading problem — every time I see this kind of discussion, I cringe at the simple-minded naivete that's always on display. Here's all we have to do to upload a brain, for instance:
The version of the uploading idea: take a preserved dead brain, slice it into very thin slices, scan the slices, and build a computer simulation of the entire brain.
If this process manages to give you a sufficiently accurate simulation
It won't. It can't.
I read the paper he recommended: it's by a couple of philosophers. All we have to do is slice a brain up thin and "scan" it with sufficient resolution, and then we can just build a model of the brain.
I've worked with tiny little zebrafish brains, things a few hundred microns long on one axis, and I've done lots of EM work on them. You can't fix them into a state resembling life very accurately: even with chemical perfusion with strong aldehyedes of small tissue specimens that takes hundreds of milliseconds, you get degenerative changes. There's a technique where you slam the specimen into a block cooled to liquid helium temperatures -- even there you get variation in preservation, it still takes 0.1ms to cryofix the tissue, and what they're interested in preserving is cell states in a single cell layer, not whole multi-layered tissues. With the most elaborate and careful procedures, they report excellent fixation within 5 microns of the surface, and disruption of the tissue by ice crystal formation within 20 microns. So even with the best techniques available now, we could possibly preserve the thinnest, outermost, single cell layer of your brain…but all the fine axons and dendrites that penetrate deeper? Forget those.
We don't have a method to lock down the state of a 3kg brain. What you're going to be recording is the dying brain, with cells spewing and collapsing and triggering apoptotic activity everywhere.
And that's another thing: what the heck is going to be recorded? You need to measure the epigenetic state of every nucleus, the distribution of highly specific, low copy number molecules in every dendritic spine, the state of molecules in flux along transport pathways, and the precise concentration of all ions in every single compartment. Does anyone have a fixation method that preserves the chemical state of the tissue? All the ones I know of involve chemically modifying the cells and proteins and fluid environment. Does anyone have a scanning technique that records a complete chemical breakdown of every complex component present?
I think they're grossly underestimating the magnitude of the problem. We can't even record the complete state of a single cell; we can't model a nematode with a grand total of 959 cells. We can't even start on this problem, and here are philosophers and computer scientists blithely turning an immense and physically intractable problem into an assumption.
And then going on to make more ludicrous statements…
Axons carry spike signals at 75 meters per second or less (Kandel et al. 2000). That speed is a fixed consequence of our physiology. In contrast, software minds could be ported to faster hardware, and could therefore process information more rapidly
You're just going to increase the speed of the computations — how are you going to do that without disrupting the interactions between all of the subunits? You've assumed you've got this gigantic database of every cell and synapse in the brain, and you're going to just tweak the clock speed…how? You've got varying length constants in different axons, different kinds of processing, different kinds of synaptic outputs and receptor responses, and you're just going to wave your hand and say, "Make them go faster!" Jebus. As if timing and hysteresis and fatigue and timing-based potentiation don't play any role in brain function; as if sensory processing wasn't dependent on timing. We've got cells that respond to phase differences in the activity of inputs, and oh, yeah, we just have a dial that we'll turn up to 11 to make it go faster.
I'm not anti-AI; I think we are going to make great advances in the future, and we're going to learn all kinds of interesting things. But reverse-engineering something that is the product of almost 4 billion years of evolution, that has been tweaked and finessed in complex and incomprehensible ways, and that is dependent on activity at a sub-cellular level, by hacking it apart and taking pictures of it? Total bollocks.
If singularitarians were 19th century engineers, they'd be the ones talking about our glorious future of transportation by proposing to hack up horses and replace their muscles with hydraulics. Yes, that's the future: steam-powered robot horses. And if we shovel more coal into their bellies, they'll go faster!
The main author of that piece, Anders Sandberg, is trained as a neuroscientist, and it summarizes the results of a workshop of mainly neuroscientists on the topic.
"You’re just going to increase the speed of the computations — how are you going to do that without disrupting the interactions between all of the subunits?... We’ve got cells that respond to phase differences in the activity of inputs, and oh, yeah, we just have a dial that we’ll turn up to 11 to make it go faster."
This seems to assume, contrary to the authors, running a brain model at increased speeds while connected to real-time inputs. For a brain model connected to inputs from a virtual environment, the model and the environment can be sped up by the same factor: running the exact same programs (brain model and environment) on a faster (serial speed) computer gets the same results faster. While real-time interaction with the outside would not be practicable at such speedup, the accelerated models could still exchange text, audio, and video files (and view them at high speed-up) with slower minds.
And the brain looks embarrassingly parallel (signal speeds and spiking rates are orders of magnitude slower than electronic computer components) so that using more hardware should give speedups, as more cores let each core model fewer compartments faster..
The paper discusses a range of possible levels of detail about a brain that might be required to effectively model it, including very fine-grained chemical state. This post seems to suggest otherwise.
"we can’t model a nematode with a grand total of 959 cells"
The paper also emphasizes the weakness of current computational models and the need for massive progress in that area. They are not claiming that there will be robust brain emulations in 10 years, or that we have the ability to accurately model small brains and just need to scale that up. If we could create reliable emulations of C. Elegans' nervous system that could control robotic worm-bodies indistinguishably from the real thing, we would have solved the biggest problems already.
There is something very comforting, and juvenile, about taking a complex problem, stuffing it into a jar, slapping a neatly written label on it and sliding it onto a mental shelf called "understood". The ability of religion to answer questions is largely based upon this mental trick.
Aside - IMHO it pays to spend some time studying primitive religions and shamanism. Primitive religion is centered on a set of concepts that pretty well describe how the human brain works conceptually. In this case it is the power of naming. Naming, the ability and power to 'take control of' ideas by stuffing ideas into mental jars and affixing a label on them. Christianity has a jar labeled "miracles" and another labeled "Mystery of God" and these allow them to safely and conveniently handle a multitude of problematic issues. Naming, labeling, is a very human method of dealing with potential messes.
The uploading problem can be smoothed over by simply stuffing it into a mental jar and talking about it only as a feat that will soon be, in the course of time, inevitably, mastered and so become so trivial as to be beneath mentioning or worrying about. Stuff the troublesome idea that this will be a very long time coming, if it is possible at all, into a handy container and use the container as a placeholder. Use this placeholder so often that pretty much everyone, outside a certain biologist with an annoying habit of looking entirely too closely at the contents of an otherwise perfectly nice jars, skates over the rough spot without noticing.
The brain, and eventually consciousnesses, seems start, presumably, as soon as there are two brain cells to work with, and to stay active until you die. Capturing or modeling that activity is going to be like filming a horse running and then, in another application of primitive magic, assuming you capture the 'spirit' in a series of pictures.
Early anatomists dug into corpses, sometime living people, trying to find the life essence. We seem to be still digging but still avoiding the obvious that life, in this case brain function, is not captured in the physicality of the brain so much as the process of the brain. An individual process that is impossible to capture because it only exists as an active conversation between neurons.
It's an implicit circular argument. They assume that the brain is basically a modern computer...and then conclude that they could simulate it in a computer.
Now, if they could fully simulate it in a computer, then they probably could do things like run the simulation at faster speed -- the relative ease of that sort of thing is the nature of computation. There are layers upon layers of simulation and flexibility in those layers. But that initial assumption is a doozy.
> . . .even with chemical perfusion with strong aldehyedes of small tissue
> specimens that takes hundreds of milliseconds, you get degenerative changes. . .
Eh, those boring scientific details are so **irrelevant**, man!
Dontcha know that with the acceleration of accelerating, uh,
acceleration ([TM] Ray Kurzweil), all we have to do is wait a
while and all those niggling obstacles will disappear? (And if
those pessimistic human science drudges can't figure it out,
then shiny superintelligent AIs will!)
It's like riding an escalator. All you have to do is stand there,
and you're automatically delivered to the next floor of the evolutionary
I'm amused see a biologist mocking the technique of hacking things apart to understand them. It makes me smell the formaldehyde again. Such 19th-century engineers would have gotten the substance of contemporary transportation correct (minus the glorious part) if not the mechanism. Your position resembles that of friendy-AI folks, who often question the reverse-engineering model (albeit with bubbly enthusiasm rather than sober skepticism).
Let me sneak in here before the singularitarians show up in droves to beat you up, and say THANK YOU for pointing out just how clueless that whole "pull out the neurons one at a time and replace them with simulations until you've replaced the whole brain" fantasy really is. I'm a retired computer engineer (hardware for a few years, then software for a lot more) who started out supporting a neuroscience lab that was mapping the cat visual system, so I know both sides of the argument. I keep trying to explain to people who seem to think that there's no basic difference between a nervous system and a computer ("All computing devices are equivalent!" "So what makes you think the brain is a computing device?" "But thinking is computing!" "And you know this, how?") that not only is the brain not a digital computer, but we in fact don't know how it works at most of the levels of organization in it, but we can be reasonably sure that its function is dependent on its complexity, which is orders of magnitude greater than any human artifact. The usual response is "Oh, that's just a detail", at which point I usually give up in disgust.
The funny part is that the people I often argue with don't know much about how real computers work either, but they won't take my word for that either, even though I've worked as a hardware engineer for several microprocessor manufacturing companies. For instance, your remarks about timing and hysteresis are true for silicon circuits, too: I've had to debug digital systems that failed at certain clock speeds because their timing was determined by propagation times from one point to another. I guess the lure of the fantasy is just too great to allow people to hear the truth.
It always makes me cringe a little when philosophers and their ilk try and formulate how the singularity will begin--who better to speculate on how we'll transition from flesh to graphene than people who don't know how either one works? They're more closely related than I think PZ gives them credit for, but that includes all the annoying hardware issues, slapped-together code, and half a dozen components from different manufacturers all trying to create something that resembles a unified system if you tilt your head and squint a little. As we get better with AI, it'll obviously grow to meet us in the middle--but even if you did somehow divine the exact state of a brain despite all PZ just pointed out, that wouldn't as much be swapping out a body as making a new person. You can't just "move" a file from one system to another--you copy it, then delete the original. To migrate properly, you'd have to replace your storage medium one part at a time--it'd literally be a lifetime of risky surgery and adjustment. Just replacing your eyes or ears with the crude prosthetics we have now takes weeks of work and years of acclimation to the implants--imagine swapping out your goddamn hippocampus!
You can’t fix them into a state resembling life very accurately: even with chemical perfusion with strong aldehyedes of small tissue specimens that takes hundreds of milliseconds, you get degenerative changes.
How big or complex is the state needed to accurately capture a neuron? Is it actually known what, in principle, needs to be captured? (I have a vague idea that it's more than "this neuron is connected to these neurons", but I don't know what else there is.)
You’re just going to increase the speed of the computations — how are you going to do that without disrupting the interactions between all of the subunits?
I think you may be missing something here--nobody is saying that they'll tweak some part of the simulation, analogous to making ion pumps run really efficiently in meat brains or something like that; the idea is to run the whole simulation faster; everything scales together.
Wow! Visions of steam and coal-smoke spewing Klanks a la Foglios' "Girl Genius" are dancing in my head!
The man who says something is impossible should not interrupt the person doing it.
Unlike Kurzweil's maniacally optimistic predictions, I concede that even I (aged 25) will probably not see human mind uploading in my lifetime. The technological hurdles are great, and many scientists conveniently forget the statefulness of a cell, as you rightfully point out (e.g. its epigenetic code--don't get me started on this "you are your connectome" fallacy).
However, does this mean it will never be possible to take a living human, and destructively scan their brain so accurately that not even philosophers will be able to quibble about whether the upload is still the same person? I think not. There is tremendous pressure to accomplish this--not just pressure from various walks of life, but pressure from life itself--evolutionary pressure.
We're not trying to build a teleporter to the Moon here; we're applying an arsenal of (mostly existing) tools to a 1.5 kg blob of jelly. And in some respects, we're making pretty decent headway; see e.g. http://homes.mpimf-heidelberg.mpg.de/~smikula/blockface/
Of course neither of us has, at this point, any persuasive arguments to convince the other party of his being right. What I do hope you will acknowledge, is that this project is at least plausible within a reasonable timescale (where reasonable means human, not galactic orders of magnitude).
Now, I would like to tackle one more point that you discuss. You say that the statement about the speedup of a simulation, with respect to real time, is ludicrous. This is not so; in fact, it's entirely trivial. The key is not to speed up just the axons, or just one part of the brain--everything will start going haywire real quick without a doubt. But once we have the full brain in a simulation, this simulation does not need to interact with the real world to, say, read a book. A simulation speedup of the entire simulated environment with respect to the real world should go without a hitch--provided there's no communication between the two during this period. The upload would just be living in an accelerated universe. Whether this constitutes actual faster cognition, of course, is disputable.
There are plenty of misconceptions about uploading. Just the other day, I heard Titus Nachbauer say, in his feature-length documentary Transhuman, that we would no longer need to dream as uploads. This statement clearly displays a complete lack of understanding of how the brain works. However, even misconceptions like these hardly seem grounds for dismissing the entire concept of mind uploading.
What happened to all the comments?
Yikes, all the comments in all the threads are gone.
"You’re just going to increase the speed of the computations — how are you going to do that without disrupting the interactions between all of the subunits? You’ve assumed you’ve got this gigantic database of every cell and synapse in the brain, and you’re going to just tweak the clock speed…how? You’ve got varying length constants in different axons, different kinds of processing, different kinds of synaptic outputs and receptor responses, and you’re just going to wave your hand and say, “Make them go faster!” Jebus. As if timing and hysteresis and fatigue and timing-based potentiation don’t play any role in brain function; as if sensory processing wasn’t dependent on timing. We’ve got cells that respond to phase differences in the activity of inputs, and oh, yeah, we just have a dial that we’ll turn up to 11 to make it go faster."
This isn't really a valid criticism. *Assuming* you can simulate a brain, or a universe, or whatever, it is necessarily possible to vary the speed of the entire simulation. All parts would remain consistent with one another, no matter how finicky, because from the point of view of inside the simulation (as long as you don't look at any outside input :P) nothing has changed.
I love your comment about 4 billion years of evolution. Biological systems seem hopelessly complex, and lacking any sort of planning. If there are 100 ways for molecular task to get done, you can be guaranteed that nature will use all 100 different ways, even though using just one way would be more organized :-)
Well, regardless of how it happens, I for one, welcome our robot overlords. There, someone had to say it.
What they also forget is that you'd also have to simulate a human body as well. Why? Have you ever talked to someone who has lost a limb? How about someone that is completely paralysed from the neck down? Now imagine you have NO sensory information. NONE. Not even the blood flow of a body. Any sane person would be begging for death. Not to mention the fact that your brain, prior to being scanned, would be calibrated to take all that information. Suddenly depriving it of vital interactions seems like a great way to cause a sudden collaspe of anything resembling sanity.
I agree entirely, people are thinking way too far ahead, we don't even know where to start on this! Although I disagree about increasing the speed of computation. I'm fairly sure that if you're simulation is contained entirely within a computer, simply increasing the speed of computation is going to increase the speed of thought. Even if we assume that the processing is distributed, and not single threaded, then increasing the speed of the sub units is still going to increase the speed of the simulation, provided the connecting network can deal with the increased traffic. To say otherwise is to say that a computational simulation of the physical universe has a fundamental speed limit. Say I am modelling fluid flow, if I perform the calculations at a faster rate, I will see that the simulation runs more quickly. If they did not then some very strange things must be happening.
Yes, those silly 19th century engineers who thought they could design and build a better mode of transportation than a horse. Evolution worked for billions of years to give us a horse, you think mere humans could do better than that in only a few decades? This is why, here in the 21st century, most people still travel via horseback. Duh.
You're making two assumptions, which are laugh out loud wrong when stated blatantly. First, that there is only one possible solution to the problem, and if that doesn't work then the whole problem is insoluable. Coal is a bad energy source for vehicles? Cars aren't possible then- never mind using distilled naptha (aka gasoline) as an alternative solution, I've just disproved cars. Freezing and slicing doesn't work? OK, how about some super fine resolution MRI scanning (which has the added advantage of being able to be done without killing the subject)? Doesn't matter- the first proposed solution to the problem doesn't work, so the problem is insoluable.
The second error is much more obvious in your (hopefully unvoiced) response to the above paragraph. And that is the assumption that technology doesn't advance. What we have now is pretty much as good as things will ever get. So the response to the above paragraph was "where are these super high resolution MRI scanners", wasn't it? If we don't have them today, we'll never get them, because technology doesn't advance.
A more reasonable set of assumptions to make, given history, is that technology is going to continue to improve, and if one solution to a problem doesn't work, we'll come up with other solutions. At which point, unless you wish to argue that there is some bit of magic woo required for intelligence- call it spirit, or soul, or what have you- which will forever be beyond the ability of science to measure, then it's just a question of when, not if, we'll be able to produce a map of the brain of sufficient detail, and throw enough computation power at the simulator, to be able to upload a thinking mind.
I love this article!
I suppose one could argue that you also make quite a few assumptions that are probably not well supported, and where those of us (see http://carboncopies.org) who are not just philosophizing, but actually developing tools would answer "but... you wouldn't do that, you would do this".
Still, I wish every proponent of whole brain emulation thought about the issues with as much clarity as you, the skeptic, do!
Thank you for adding some realism to the discussion!
Assuming you have indeed managed to create a digital emulation of the human brain (not just its state but also its function), no matter how utopic with our current technology, it would indeed be possible to increase its working speed, since the physiological constraints of a physical brain wouldn't apply. It wouldn't create a more intelligent brain (it wouldn't be able to solve any new problems), just a faster brain (solve the same problems faster).
I think you assume that such a simulated brain would need to simulate the complete anatomy of the brain, which need not be the case as long as the result is the same. It's as if you are objecting to a simulated digital library being able to manage thousands of transactions per millisecond, because the simulated librarian wouldn't be able to run fast enough from the one simulated bookshelf to the other. The question is only whether such a perfect digital emulation is possible. As long as it is (which is a huge assumption nevertheless), then yes, faster computation speed would result in a faster thinking brain.
Except that's not what he was saying. What he was commenting on was that singularitarians usually grossly underestimate the complexity of the interactions that make up the human brain, and so they often believe that speeding up even one facility will lead to a speed increase in the brain. The 75 m/s figure that was given was only one figure which has been calibrated by evolution to properly synchronize with the speed of everything else in the brain (most of which are moving very fast already).
Love this post!
You misunderstand. These discussions of brain uploading aren't supposed to describe how it's done. They're presenting a philosophical argument: The brain is physical, and consciousness is a matter of computation; therefore, the brain can be copied into different hardware, and its operational characteristics can be changed.
Picking on them for not currently knowing how to do it, or because there may be more efficient ways of getting superior results, is missing the point.
"by a couple of philosophers".
Anders Sandberg is a neuroscientist. Nick Bostrom is indeed a philosopher, though he has an MSC in computational neuroscience.
"You’re just going to increase the speed of the computations — how are you going to do that without disrupting the interactions between all of the subunits?"
You know how on a really fast starship, time slows down, and yet human brains don't fall apart due to all the phase problems you just described? Yeah. That in reverse.
You've just made a 'rockets won't work in a vacuum' statement. Once you have a sim, you can run it twice as fast if your computer is good enough. To such a mind, the ourter world would appear to run at half speed. To the outer world, the mind would have reactions twice as fast, but also get tired twice as fast.
I'll go read the article, and I'm sure there are problems, but I lack 'trust' in people who make silly errors like this.
While your points are interesting and fair enough the title of the Hallquist article does say "in the next century or two." Don't you think techniques might advance sufficiently in that sort of time frame? I mean the KEMS machine can already kind of scan brains at 0.3um resolution, admittedly just microscope images, but things might move on? http://www.wired.co.uk/news/archive/2012-05/02/brain-slicing-and-mapping
I agree completely that nobody is going to scan and simulate a brain anytime in the forseeable future.
But, _if_ you had a computer simulation of a brain, there is nothing intrinsically difficult in making the entire brain run faster. You wouldn't change the physiology, as you seem to assume. You'd simulate the same physiology, but just change the relationship between the computer's clock and the real world's clock. It's just like simulating the motions of the solar system more rapidly: rather than fiddling with gravity within the simulations, you just put it on a faster computer and turn up the clock.
Greg, how would you get real-time, accurate scans of the precise chemical state of the entire brain? That's the whole point of what PZ's saying--the minute you take it apart enough to observe it closely, it's already starting to degrade. Even flash-freezing it would still cause such massive degradation that reconstructing your identity from what's left is about as plausible as the folks on CSI producing court-admissible photos with an enhancing tool. We just don't know enough about how the whole machine works to replicate it yet.
Besides, consciousness isn't just about processing power--IIRC, we already have prototype computers that are theoretically more powerful than a small rodent brain, but that doesn't mean we can build realistic mouse androids (musoids?) any more than we can get an ant colony to run Fedora.
I enjoy your rebuttal of Kurzwail and the singularitarians, but I see one person missing from the kook category - Aubrey De Grey.
Please don't argue with strawmen; we need your help to deal with this unusual and dangerous situation. OK, so you know some specific facts about extraction of information from brains-- that's useful to us, if you will put that information to use in helping us to predict WHEN certain transitions will take place. It seems like from what you're saying that you don't think we'll have full uploading until we have robots small enough to enter a live brain and take readings directly, yes? That much of a postponement in full uploading could have important consequences for how the Immortalists will act in the coming decades-- those are the questions we need to think about. It's not productive to debate whether the future is even going to occur at all, unless you have some thought-through notion of a particular steady-state rejectionist society for us to consider.
>It’s not productive to debate whether the future is even going to occur at all, unless you have some thought-through notion of a particular steady-state rejectionist society for us to consider.
MJ, do you have any real-world experience in research science or engineering?
I’m a physicist: for nearly fifty years I have been reading about research in “controlled fusion.” People were more optimistic about the prospects for practical controlled fusion fifty years ago than they are today.
“The future” has gotten more and more distant with each passing decade.
The devil is in the details. And, the details that PZ has laid out are absolutely overwhelming to anyone with serious technical training.
Sure, something may come up. Someday. And, maybe it will turn out that G. W. Bush was really a space alien.
Personally, I’d bet on seeing pigs fly (genetically modified pigs, to be sure – let PZ tell you how hard that would be, given the embryological development of pigs vs. birds!) before we see the kind of brain scan that PZ is debunking.
Dave Miller in Sacramento
i cant get how you are able to share like this wonderful posts admin very much thanks