10 Important Differences Between Brains and Computers : Developing Intelligence

Profile

Chris Chatham is a grad student at the University of Colorado, Boulder.

Search

Currently Reading

Brain Games

« Interference from the Past: Developmental Trends | Main | Blogging on the Brain: 3/29 »

10 Important Differences Between Brains and Computers

[ Artificial Intelligence, Cognitive Neuroscience, Computational Modeling ]
Posted on: March 27, 2007 12:38 PM, by Chris Chatham

"A good metaphor is something even the police should keep an eye on." - G.C. Lichtenberg

Although the brain-computer metaphor has served cognitive psychology well, research in cognitive neuroscience has revealed many important differences between brains and computers. Appreciating these differences may be crucial to understanding the mechanisms of neural information processing, and ultimately for the creation of artificial intelligence. Below, I review the most important of these differences (and the consequences to cognitive psychology of failing to recognize them): similar ground is covered in this excellent (though lengthy) lecture.

Difference # 1: Brains are analogue; computers are digital

It's easy to think that neurons are essentially binary, given that they fire an action potential if they reach a certain threshold, and otherwise do not fire. This superficial similarity to digital "1's and 0's" belies a wide variety of continuous and non-linear processes that directly influence neuronal processing.

For example, one of the primary mechanisms of information transmission appears to be the rate at which neurons fire - an essentially continuous variable. Similarly, networks of neurons can fire in relative synchrony or in relative disarray; this coherence affects the strength of the signals received by downstream neurons. Finally, inside each and every neuron is a leaky integrator circuit, composed of a variety of ion channels and continuously fluctuating membrane potentials.

Failure to recognize these important subtleties may have contributed to Minksy & Papert's infamous mischaracterization of perceptrons, a neural network without an intermediate layer between input and output. In linear networks, any function computed by a 3-layer network can also be computed by a suitably rearranged 2-layer network. In other words, combinations of multiple linear functions can be modeled precisely by just a single linear function. Since their simple 2-layer networks could not solve many important problems, Minksy & Papert reasoned that that larger networks also could not. In contrast, the computations performed by more realistic (i.e., nonlinear) networks are highly dependent on the number of layers - thus, "perceptrons" grossly underestimate the computational power of neural networks.

Difference # 2: The brain uses content-addressable memory

In computers, information in memory is accessed by polling its precise memory address. This is known as byte-addressable memory. In contrast, the brain uses content-addressable memory, such that information can be accessed in memory through "spreading activation" from closely related concepts. For example, thinking of the word "fox" may automatically spread activation to memories related to other clever animals, fox-hunting horseback riders, or attractive members of the opposite sex.

The end result is that your brain has a kind of "built-in Google," in which just a few cues (key words) are enough to cause a full memory to be retrieved. Of course, similar things can be done in computers, mostly by building massive indices of stored data, which then also need to be stored and searched through for the relevant information (incidentally, this is pretty much what Google does, with a few twists).

Although this may seem like a rather minor difference between computers and brains, it has profound effects on neural computation. For example, a lasting debate in cognitive psychology concerned whether information is lost from memory because of simply decay or because of interference from other information. In retrospect, this debate is partially based on the false asssumption that these two possibilities are dissociable, as they can be in computers. Many are now realizing that this debate represents a false dichotomy.

Difference # 3: The brain is a massively parallel machine; computers are modular and serial

An unfortunate legacy of the brain-computer metaphor is the tendency for cognitive psychologists to seek out modularity in the brain. For example, the idea that computers require memory has lead some to seek for the "memory area," when in fact these distinctions are far more messy. One consequence of this over-simplification is that we are only now learning that "memory" regions (such as the hippocampus) are also important for imagination, the representation of novel goals, spatial navigation, and other diverse functions.

Similarly, one could imagine there being a "language module" in the brain, as there might be in computers with natural language processing programs. Cognitive psychologists even claimed to have found this module, based on patients with damage to a region of the brain known as Broca's area. More recent evidence has shown that language too is computed by widely distributed and domain-general neural circuits, and Broca's area may also be involved in other computations (see here for more on this).

Difference # 4: Processing speed is not fixed in the brain; there is no system clock

The speed of neural information processing is subject to a variety of constraints, including the time for electrochemical signals to traverse axons and dendrites, axonal myelination, the diffusion time of neurotransmitters across the synaptic cleft, differences in synaptic efficacy, the coherence of neural firing, the current availability of neurotransmitters, and the prior history of neuronal firing. Although there are individual differences in something psychometricians call "processing speed," this does not reflect a monolithic or unitary construct, and certainly nothing as concrete as the speed of a microprocessor. Instead, psychometric "processing speed" probably indexes a heterogenous combination of all the speed constraints mentioned above.

Similarly, there does not appear to be any central clock in the brain, and there is debate as to how clock-like the brain's time-keeping devices actually are. To use just one example, the cerebellum is often thought to calculate information involving precise timing, as required for delicate motor movements; however, recent evidence suggests that time-keeping in the brain bears more similarity to ripples on a pond than to a standard digital clock.

Difference # 5 - Short-term memory is not like RAM

Although the apparent similarities between RAM and short-term or "working" memory emboldened many early cognitive psychologists, a closer examination reveals strikingly important differences. Although RAM and short-term memory both seem to require power (sustained neuronal firing in the case of short-term memory, and electricity in the case of RAM), short-term memory seems to hold only "pointers" to long term memory whereas RAM holds data that is isomorphic to that being held on the hard disk. (See here for more about "attentional pointers" in short term memory).

Unlike RAM, the capacity limit of short-term memory is not fixed; the capacity of short-term memory seems to fluctuate with differences in "processing speed" (see Difference #4) as well as with expertise and familiarity.

Difference # 6: No hardware/software distinction can be made with respect to the brain or mind

For years it was tempting to imagine that the brain was the hardware on which a "mind program" or "mind software" is executing. This gave rise to a variety of abstract program-like models of cognition, in which the details of how the brain actually executed those programs was considered irrelevant, in the same way that a Java program can accomplish the same function as a C++ program.

Unfortunately, this appealing hardware/software distinction obscures an important fact: the mind emerges directly from the brain, and changes in the mind are always accompanied by changes in the brain. Any abstract information processing account of cognition will always need to specify how neuronal architecture can implement those processes - otherwise, cognitive modeling is grossly underconstrained. Some blame this misunderstanding for the infamous failure of "symbolic AI."

Difference # 7: Synapses are far more complex than electrical logic gates

Another pernicious feature of the brain-computer metaphor is that it seems to suggest that brains might also operate on the basis of electrical signals (action potentials) traveling along individual logical gates. Unfortunately, this is only half true. The signals which are propagated along axons are actually electrochemical in nature, meaning that they travel much more slowly than electrical signals in a computer, and that they can be modulated in myriad ways. For example, signal transmission is dependent not only on the putative "logical gates" of synaptic architecture but also by the presence of a variety of chemicals in the synaptic cleft, the relative distance between synapse and dendrites, and many other factors. This adds to the complexity of the processing taking place at each synapse - and it is therefore profoundly wrong to think that neurons function merely as transistors.

Difference #8: Unlike computers, processing and memory are performed by the same components in the brain

Computers process information from memory using CPUs, and then write the results of that processing back to memory. No such distinction exists in the brain. As neurons process information they are also modifying their synapses - which are themselves the substrate of memory. As a result, retrieval from memory always slightly alters those memories (usually making them stronger, but sometimes making them less accurate - see here for more on this).

Difference # 9: The brain is a self-organizing system

This point follows naturally from the previous point - experience profoundly and directly shapes the nature of neural information processing in a way that simply does not happen in traditional microprocessors. For example, the brain is a self-repairing circuit - something known as "trauma-induced plasticity" kicks in after injury. This can lead to a variety of interesting changes, including some that seem to unlock unused potential in the brain (known as acquired savantism), and others that can result in profound cognitive dysfunction (as is unfortunately far more typical in traumatic brain injury and developmental disorders).

One consequence of failing to recognize this difference has been in the field of neuropsychology, where the cognitive performance of brain-damaged patients is examined to determine the computational function of the damaged region. Unfortunately, because of the poorly-understood nature of trauma-induced plasticity, the logic cannot be so straightforward. Similar problems underlie work on developmental disorders and the emerging field of "cognitive genetics", in which the consequences of neural self-organization are frequently neglected .

Difference # 10: Brains have bodies

This is not as trivial as it might seem: it turns out that the brain takes surprising advantage of the fact that it has a body at its disposal. For example, despite your intuitive feeling that you could close your eyes and know the locations of objects around you, a series of experiments in the field of change blindness has shown that our visual memories are actually quite sparse. In this case, the brain is "offloading" its memory requirements to the environment in which it exists: why bother remembering the location of objects when a quick glance will suffice? A surprising set of experiments by Jeremy Wolfe has shown that even after being asked hundreds of times which simple geometrical shapes are displayed on a computer screen, human subjects continue to answer those questions by gaze rather than rote memory. A wide variety of evidence from other domains suggests that we are only beginning to understand the importance of embodiment in information processing.

Bonus Difference: The brain is much, much bigger than any [current] computer

Accurate biological models of the brain would have to include some 225,000,000,000,000,000 (225 million billion) interactions between cell types, neurotransmitters, neuromodulators, axonal branches and dendritic spines, and that doesn't include the influences of dendritic geometry, or the approximately 1 trillion glial cells which may or may not be important for neural information processing. Because the brain is nonlinear, and because it is so much larger than all current computers, it seems likely that it functions in a completely different fashion. (See here for more on this.) The brain-computer metaphor obscures this important, though perhaps obvious, difference in raw computational power.

Share this: Facebook Twitter Stumbleupon Reddit Email + More

TrackBacks

TrackBack URL for this entry: http://scienceblogs.com/mt/pings/35137

Comments

Fantastic overview, something I think I'll refer people to if they start talking about computers being able to go beyond the human brain in the next 10 years

Posted by: Adam Roberts | March 27, 2007 1:54 PM

Excellent article and very informative.

--Brett

Posted by: Brett | March 27, 2007 2:11 PM

I have to disagree with you on a number of these points (namely, the first half). While they are all generally true regarding the computers we work with during our daily blogging activities, I don't think they are entirely true when taking into account a more sophisticated understanding of what a "computer" is.

Difference # 1: Brains are analogue; computers are digital

Calling something a computer doesn't necessarily mean that it must have digital circuits, though analog circuits clearly are rare. So this is good to point out, but there is interest in analog circuitsespecially in building artificial neural networks,.

Also, I fail to see how rate of firing is significant in digital versus analog. If the importance of the continuity of firing is the average signal strength over a given time period then a signal sent digitally in intervals should be equivalent in its ability to cause the next set of neurons to fire. There might be something I'm missing here, but as you stated it I don't see how this necessitates analog circuitry.

Also, I think the Minsky & Papert reference is off the mark. It certainly was a sad day for AI, but again I don't see what this has to do with analog versus digital. What you mentioned demonstrates their fallacious thinking about how one can structure neural networks and what they can compute, but what does that have to do with whether they are analog or digitial?

Alan Turing firmly believed it was possible to simulate the continuous nature of the brain on a digital machine; however I think this is still an open question. At any rate, I think the more important question is, given that one can build a machine using analog circuits and call it a computer, whether the entire brain can be described using math equations.

Difference # 2: The brain uses content-addressable memory

I think this is really only a superficial distinction. If you implement a neural network on a computer, then all of the inner workings of the CPU and its methods of memory allocation become irrelevant. If a neural network can be simulated digitally, then where or how it is implemented is a non-issue, as any implementation will be equivalent. (So the question is then analog versus digital.)

On the other hand, I can see how this architectural difference can be important to point out to people who have no idea how the brain or computers are constructed.

Difference # 3: The brain is a massively parallel machine; computers are modular and serial

In the same way that implementation wasn't an issue before, it isn't here either. Parallel processing can be implemented equivalently on a serial machine. Also, this isn't even true anymore, as most if not all supercomputers used for research have dozens, hundreds, or even thousands of processors. And even consumer level machines are becoming parallel, with dual-core CPUs coming out in the past few years.

The modularity issue I'm intrigued by. Clearly the areas of the brain are not as discrete as those in our computers, but don't we still refer to experience occurring in the neocortex? Although I really don't know enough about this (and I want to know more!) there must be some level of modularity occurring in the brain. My gut instinct is telling me here that a brain based completely on spaghetti wiring just wouldn't work very well...

Difference # 4: Processing speed is not fixed in the brain; there is no system clock

You might call me out for nitpicking here, but CPUs don't require system clocks. Asynchronous processors are being actively researched and produced.

A key advantage is that clockless CPUs don't consume energy when they aren't active. Machines based on synchronous processors, on the other hand, constantly have the "pulse" of the clock traveling through the system (and the frequency at which it "beats" determines the speed of the CPU). The pulse of the clock continuously coursing through all of the circuits also results in "wasted cycles," meaning power is being used when the CPU isn't doing anything, and heat is being dissipated for no reason.

Difference # 5 - Short-term memory is not like RAM

Again, this seems to be a superficial architectural difference. I think that if your intent is to simulate the brain using artificial neural networks, then the how the RAM or hard drive works is inconsequential.

I'll admit that it is something worth pointing out to someone who does take the brain/computer analogy too far (which is I guess exactly who you're targeting here) or doesn't know much about computers or brains.

Difference # 6: No hardware/software distinction can be made with respect to the brain or mind

This one I completely agree with. I always get the feeling when I read philosophy of AI papers that some of the philosophers take the sentiment "the mind is the program being executed on the machine that is the brain" too far. Consequently, and I feel this is actually a central problem with philosophy of AI, they pay too little attention to how the brain actually operates and try to think about how to implement consciousness on a computer without considering how the activities of the brain relate to the mind.

...

Anyway, I think it would be fair to describe the brain as an asynchronous, analog, and massively parallel computer where the hardware itself is inherently mutable and self-organizing.

Posted by: Jonathan | March 27, 2007 4:33 PM

Jonathan - thanks for these astute comments!

Minksy & Papert's conclusions were based on the faulty assumption that neural networks are computationally linear; linearity is a very unusual characteristic for analogue systems. Had M&P fully appreciated the analogue nature of the brain, they likely would not have made this faulty assumption.

A consistent thread in your comment is that some differences are merely "implementational" or "architectural" details, and thus are actually unimportant or otherwise superficial. IMO, that attitude is scientifically dangerous (how can you know for sure?) and *very* premature (when we have an artificially intelligent digital computer, I'll be convinced).

It is also the same attitude that pervaded both classic cognitive psychology and GOFAI (good old-fashioned AI). I don't think the track record of either is very good: 20th century advances in statistical theory may be responsible for the few successes in both disciplines (just don't tell Chris at Mixing Memory I said that... ;)

Posted by: Chris Chatham | March 27, 2007 5:03 PM

Just adding something to Jonathan's answer to #6: A computer can run entirely in hardware (actually it's a very strange affirmation, but you know what I mean). And more: if that hardware is a FPGA, it can mutate and self-organize.

I think Chris' arguments would target "today personal computers" and not computers in general, since "computer" is a very wide term. However, I think it was the true target of the article, just with some differences/mistakes in the use of the terms.

Posted by: Rafael C.P. | March 27, 2007 6:00 PM

Just wanted to chime in on what a great article this is. Not too complex or technical, and gives a great overview of the significant differences.

Posted by: Dan Lurie | March 27, 2007 8:13 PM

Rafael, when you say "I think Chris' arguments would target 'today personal computers' and not computers in general, since 'computer' is a very wide term", you have a good point, but that is exactly the computer model on which analogies were (and are) based. Chris's arguments are about the analogies were drawn, and not not about computers, and not about analogies can be drawn on the basis of future or experimental computer architectures.

Posted by: incze | March 27, 2007 8:32 PM

Thanks incze - that is exactly where I was coming from. That said, I am now considering a second post entitled "10 important similarities between brains and Turing machines" haha.

Posted by: Chris Chatham | March 27, 2007 8:57 PM

You compare high and low, fundamental things about the brain against implementation details of a specific microprocessor. But Jonathan already said all those things much better than I can.

Rafael, your point about targeting "todays computers" is questionable. There are much more obvious differences between a brain and todays computers: The microprocessor is square, the brain is roughly round. The computer run on electricity, the brain runs on icky chemical stuff. The computer has a hard disk, the brain do not. The computer stores data sequentially, the brain do not. Etc.

Posted by: Jonas | March 28, 2007 3:53 AM

Actually, all of these points are wrong: sometimes trivially, sometimes in a way that completely invalidates the conclusions that accompany them.

But the biggest problem isn't a mistake, but an unspoken assumption: it's being argued (incorrectly, as it happens) that brains aren't like computers, whenn the arguments being made are actually about the idea that brains aren't like one particular type of computational devices. Computers aren't inherently digital, for example, so the analog/digital distinction doesn't mean anything.

Posted by: Caledonian | March 28, 2007 9:46 AM

I forget where I was reading that before computers, people compared the workings of the brain to the steam engine, and before that, to watches. It seems whatever the most complex and intricate man-made device of the time was, that would be compared to the brain. It sort of makes sense because in some ways the brain is like any complex machine, and also it's interesting to compare the best man-made devices with what we may perceive to be the most important natural "device". I do think the brain has tons of things more in common with a computer than a watch (even a complex one), but maybe that is partly due to the current cultural importance of computers and my relative lack of amazement (and lack of understanding the specifics) of watches.

Posted by: Lisa | March 28, 2007 11:47 AM

Great post and comments.

To add something to taking down the idealized picture of computers:

Difference # 4 is perhaps the most superficial. Synchronous electronics is made to simplify architecture, but as noted it is wasteful and also difficult in large "nets" where the clock no longer look ideal. I haven't looked at all on todays common dual and quad processors, but it would surprise me if not one of the advantages is just decoupling of each processor from each others clock synchronity.

It is anyway rather difficult (read: impossible) to run the same software with the same timing on a more complex machine.

Difference # 6 and # 9: As noted, there are some special systems which allows reconfiguring hardware vs software to suit the task, and they may become more popular because they also save energy. (And in a small manner, that is also what happens when power saving slow down or turn off parts of some modern systems.) Ideally, this plasticity could also work around failing components in future aging systems to increase reliability and save repairs. Who knows, maybe it will become reality.

Difference # 7: Another superficial difference, since signal handling in VLSI components is complex and highly non-linear. Threshold (and local, connected) devices are made to simplify architecture, but are again wasteful and difficult in larger applications. The difference to #4 is that the alternatives are not much developed, and may never be used.

But it is rather impressive that the Blue Brain project apparently need to use one processor just to emulate one neuron...

Posted by: Torbj�rn Larsson | March 28, 2007 1:31 PM

Ok, agreed with you all =). I got to an extreme to explain what I thought, but it was wrong. I just think there are some restrictions above the arguments, but Jonathan already said it in details.

Posted by: Rafael C.P. | March 28, 2007 2:57 PM

This is a great overview and vey educational. Thanks for putting this together. I do however caution people from making assumptions as to what the nature/design of hardware and software systems may be ten years from now. That's adequate time for a whole new technology/discoveries to come to being and alter our definition of a human brain or a computer is.

Forgive me if this seems high-level and uninformed, but entirely new circuit materials and design *might* be right around the corner. Just the devil's advocate in me.

Posted by: THomas Tolleson | March 28, 2007 4:01 PM

Chris, I agree with all your points, but I can't help thinking 'straw man' as I read this. I mean, that may not be quite the right word for it, but isn't most of this already pretty well assimilated into common thinking about the brain? You have to go back pretty far in time for Minksy & Papert. With the exception of #7 (I do think a lot of writers equate neurons with big wet transistors), I don't think I've read anything that compares brain functions with computer circuitry in any kind of a literal sense. (Of course, I'm excluding certain philosophers when I say that - there's just no accounting for what some of them will argue...)

Now, I will admit that I'm not really that well-read on the subject, so maybe I've just been lucky so far.

Posted by: Kurt | March 28, 2007 11:16 PM

As someone whose specialty happens to be computer science, I would have to say that I agree overall with your overview, except for a few points.

Difference # 3: The brain is a massively parallel machine; computers are modular and serial

[pedantic computer science junky mode]

Well, no. You're generalizing a specific architecture (the serial von Neumann machine) as a computer. Parallelism, concurrency and distribution are huge areas of research now of days, primarily due to the fact that the hardware industry has reached a plateau with the serial arch.

You could grant that computers are probably not as massively parallel as human brain architecture, but that's really a question of scale and not essence. And as well, there is a good reason that parallelism hasn't been a hugely popular field up until now: parallel machines are notoriously difficult to program. (Even with the comparatively minor levels of multi-threading and task distribution being used with new multi-core processors on the market, software dev schedules are being doubled and sometimes tripled to assimilate the requirements.)

[/pedantic computer science junky mode]

Other than that, I don't have many complains. But when it comes to fields like A.I., I personally find their value from a purely CS centric perspective questionable. As technology, A.I. and adaptive computing have been beneficial in many ways, but I don't see them as being especially valuable to actually researching more "natural" computing devices (like the brain).

In the end, I see as somewhat akin to human flight. Our air traversing machines are certainly technically different than those produced by nature, but mostly because nature's ad hoc, Rube Goldberg designs didn't prove very useful. Computing is the same way, IMO. The technical value of A.I. should be able to stand on it's own.

Posted by: Tyler DiPietro | March 29, 2007 12:48 AM

This post is a healthy and much-needed rebuttal to the weird idea that in 20 or so years machine intelligence may exceed human intelligence (the so-called Singularity). The proponents of this idea seem to be basing their belief largely on extrapolating on Moore's Law.

But if Moore's law holds up for the next 20 years computers will be only about 4,000 times more capable. (And yes, I'm using the bastardized version of Moore's law that presumes increasing component density [what Moore was really talking about] correlates in a 1:1 way with increased processing power.) But if artificial intelligence requires only (or chiefly) on increased hardware power than it would already exist. It would just take 4,000 times longer than we would deem practical.

Before reading this article I would have argued is primarily as software problem. Now I have to agree that it is also a hardware problem. But who's to say we can't simulate (if not develop) hardware that will work sufficiently similar to our organic hardware? Then it will still come down to software problem. And we just don't have a good model of how human "brain software" works. And I don't think we will for a very long time.

Posted by: Tom Wrona | March 29, 2007 3:36 AM

Chris, thanks for the post: a very concise presentation of the arguments circling in my own field. I have to say though that the discussion is just as good :-) If I might add a few points...

Concerning point 10: I think this is something which has been remarkably overlooked, and one which, if you were listing your 10 points in order of importance, I would place near the top. The fact that brains (and therefore from point #6, the mind) are embodied (have a body which is localised in the real world) is, I believe, the most important constraint that is on this system. I would agree with your point that 'computers' lack this, however, I should point out that the rapidly growing field of cognitive robotics (of which I see myself as a part) is growing in importance precisely because of this. The view that brains (be they biological or otherwise) need bodies (again, biological or otherwise) is thus not one which has been forgotten.

Concerning point 3: I agree with your point, and I have to say that slightly disagree with Jonathans response that "Parallel processing can be implemented equivalently on a serial machine". I believe by definition, the best you can hope for in terms of parallel processing is a simulation of parallel processing. There are clever computational algorithms capable of getting very close in certain circumstances (threading, etc), but at the end of the day (and if using a single processor) only one computational instruction can be executed at any given time step, thus imposing a degree of serial processing on what would ideally be parallel. Also, on the point of multiprocessor systems, this is in all likelyhood adequate for relatively simple systems, however, I get the feeling (I hasten to add that this is based on limited experience of multiprocessor systems) that for larger systems (if one were to simulate many thousands of neurons an a processor per neuron), the problem would not be one of raw processing power, but one of communication limitations between the processors (bandwidth, speed etc) - although I have come across work which is attempting precisely this (in Switzerland perhaps?). Having said all this, I do believe that pseudo-parallel computation is more than adequate for most modelling purposes, and that the shortfall may be compesentated for to a certain extent.

Concerning point 9: if you are referring to hardware, then I completely agree - but not if you were also including software in that. There are many adaptive and self-organising algorithms/computational techniques capable of organisation given constraints and inputs.

More generally, the points you have raised are very important ones - particularly your assertion that the brain is nothing like a standard desktop - maybe your idea of writing another one on the similarities is a good one (on the functional nature of the two rather than the structural nature)! Thanks again though!

Posted by: Paul | March 29, 2007 5:20 AM

The things you say are all roughly correct. To make them more so would bog the article down in details that important to the majority of readers. If all you mean to do is tell people with a rough understanding of how their computer works that their brain doesn't work the same way, then all of your points are valid.

If you want to get into cutting-edge, high-end, or low-market-share technology then the argument requires more support, but is far from invalidated.

Also, the argument that the brain is not like a computer does not mean that the computer can't simulate some aspects of the brain - it means that they don't inherently work the same way.

Before I annoy anyone here's something for Jonathan:
As far as modularity goes, there is some. You can predict what kind of deficits a person will have based on where an injury occurs. The problem comes from assuming that this is where the processing of that particular thing is done. To use computers as an analogy - I can't help it - if we cut the power cord on a compter it stops adding numbers together. Thus addition takes place in the power cord.

To save reading all of this lengthy post, my other responses to Jonathan are summarized:
#2, #3 and #5 We can't accurately describe these phenomena, let alone model them. The distinction between the actual brain and the computer model only disappears when the model is completely accurate. Also, Chris has massively understated the differences between RAM and working memory. The main thing is that you COULD make enough RAM act like working memory, but that is the same as saying you could make a fly look like a raisin. It would be cruel and severely diminish the capability of your fly. For instance, to make capacity vary you could some of your RAM unavailable sometimes. Why would you do that?

To assume that current simulations of the spreading activation of a neuron works with a lookup table - the current method - is accurate is to assume that we know all about how this addressing works. I assure you that we don't. Some of my current work examines the effect of working memory load on inhibition (Following from http://www.sciencemag.org/cgi/content/abstract/sci;291/5509/1803, but I should add that I'm not affiliated with them in any way shape or form). Are you trying to tell me that the amount of RAM available will affect how we traverse a neural network lookup table? Because then the difference between working memory (which we don't really understand either) and RAM becomes extremely important.

Thus when Jonathan says "implement a neural network" does he mean a current neural network, in which case it isn't really very much like the brain, and thus not in conflict with this article at all? Or does he mean implement an accurate model of all functional aspects of the brain? Because computers aren't like that now and we have no evidence they ever will be.

The simple fact is that arguing that the brain is analogous to a Turing machine is a dangerous thing to do. Philosophers have created theoretical machines capable of solving the halting problem (for the uninitiated that's a problem computers can't solve). The brain may be a realisation of some super-Turing machine. It is true that any parallel arrangement of Turing machines can be modelled by a single machine, but it is not certain that the brain can be modelled by a collection of parallel Turing machines.

I think that the brain probably can be modelled by a Turing machine. But not yet.

Lastly, I love comment number 10. Go J. J. Gibson.

Posted by: Steve G | March 29, 2007 7:59 AM

Jonathan: Unlike Chris, I'm not buying most of your criticisms.

Yeah, we've built "artificial neural networks", but most of those are research simulations! Simulating analog processes on a digital system (or vice versa) tends to pull in huge overheads, worsening the basic order of the computational costs -- and it still isn't "exact".

Simulating massively parallel systems on CPU-based systems is worse, and less reliable. The CPU version fundamentally has time cost at least linear to the number of nodes and connections, whereas a true parallel system does not.

It might well be possible to make something like "content-addressible" memory in the RAM model, but it would be a "bloody hack" with no connection to our usual programming schemes, or to a biological-style memory.)

Then too, our ability to "program" neural nets is frankly humbled, by the ordinary development of almost any vertebrate's nervous system.

These aren't issues I'm willing to wave off as "implementation details"....

Posted by: David Harmon | March 29, 2007 11:48 AM

And here is one dimension along which the brain is like a computer.

See O'Reilly, 2006:

"Computer models based on the detailed biology of the brain can help us understand the myriad complexities of human cognition and intelligence. Here, we review models of the higher level aspects of human intelligence, which depend critically on the prefrontal cortex and associated subcortical areas. The picture emerging from a convergence of detailed mechanistic models and more abstract functional models represents a synthesis between analog and digital forms of computation. Specifically, the need for robust active maintenance and rapid updating of information in the prefrontal cortex appears to be satisfied by bistable activation states and dynamic gating mechanisms. These mechanisms are fundamental to digital computers and may be critical for the distinctive aspects of human intelligence."

Posted by: Brian Mingus | March 29, 2007 12:39 PM

I think this list is all correct, except for the last item. It seems obvious to me that the chances of anything even *close* to a biologically accurate simulation of the brain being necessary to get the relevant aspects of its functioning are tiny. Much more likely is that the vast majority of the complexity is accidental. This isn't to say that we don't need to understand most everything that goes on before we can get a good model, it's just to say that current estimations of brain computation capacity *aren't* much too low. If Moore's law holds, I don't see why we can't have supercomputers within ten years that could simulate a human brain, given a model that we probably won't have for another 15 years.

Posted by: pdf23ds | March 29, 2007 1:15 PM

Re: 'computer' defined...
It appears inevitable to speak in analogies when discussing AI, so the analogy I'd offer is this: the brain is like what we would call a computer network. It has areas of specialization of function, but it operates overall by combining data, memory and processing to arrive at decisions. While today's networks are still quite primitive, perhaps self-evolving networks using nanomachines to construct their own development/evolution would ultimately become 'intelligent'. Of course, one gets into the philosophy of human intelligence and...

"Contemplate this upon the Tree of Woe."-Conan the Barbarian

Posted by: Richard | March 29, 2007 2:04 PM

I think it's important to note that Chris is not criticizing or "targeting" computers. His target is the erroneous thinking that people working in cognitive science, AI, etc. have committed over the years because of a faulty analogy between "a brain" and "a computer". Or more accurately, "their conceptualization of a brain" and "their conceptualization of a computer."

From what I read, it seems that part of what created said thinking was a mistaken idea of how a computer is built, and so in this article Chris must explain a few points about how typical computers work in order to help dispel the incorrect portions of the analogy. Don't take that to mean that he is making authoritative statements about computers and what they can be or that his article is meant to target shortcomings of computers. Any criticism is targeted at a tendency for some researchers and lay people to think of a brain as being like the computer they sit at every day.

So yes, there are alternate computer architectures that more closely resemble a brain. And certainly computers don't have to work the way Chris describes. But the model of "computer" that he uses very strongly matches that used by people who subscribe to the erroneous analogy he is attempting to debunk.

Posted by: Roy Huggins | March 29, 2007 3:13 PM

Your points are well-taken and very relevant when people start talking about resemblances between brains and computers. But it seems a little misguided to compare brains to computers, whose current form, was a result of a lot of historical and technical factors: technology available, standardization decisions taken, and so on. I think computers could have taken other forms -- for instance, if the von neumann architecture had not proved to be so influential, isn't it possible that computers today would be massively parallel? -- that would have made them more structurally (although superficially) similar to the brain.

The key difference, it seems to me, is the difference between brain processes and computational processes. What is the role of brain processes in our interaction with the world? And at what level of description are brain processes computational (if at all)? And so on.

Still, great post!

Posted by: shreeharsh | March 29, 2007 3:35 PM

Sorry for the delay in posting all these very thoughtful comments - I have been camping in Utah for a couple of days, and just now got access to the intarweb:)

I'll go through them in more detail soon.

Posted by: Chris Chatham | March 29, 2007 4:04 PM

Sorry for the belated reply, but here goes...

As for Minsky & Papert, I'll defer to your knowledge, as I'm shaky on the concept of linearity in neural networks. I understand that they made the mistake of only considering single-layer networks when pouncing on perceptrons; if they had considered multi-layered networks they would have seen that things like XOR are possible. Linearity and analog systems notwithstanding, I can say with the hindsight of a huge generational gap that it just seems silly to me that they didn't consider multi-layered networks.

We can know for sure because all modern day digital computers are Turing-equivalent, meaning any program implemented on one can by implemented on another and be computationally equivalent despite differences in system design. Just as the brain only has hardware (as you said, there is no software that is the mind running on top), the only thing that counts when programming a mind is the software. The high-level software need not be concerned with memory registers when representing knowledge, and "pointers" can be implemented on a system that uses only non-volatile memory.

I think the only real problem here is whether or not digital computers can simulate the continuous nature of the brain. If it is the case that a discrete state machine is not hindered by this, then the brain's architecture with all of the intricacies of neuronal activity can be implemented to the fullest extent with no other problem (although we'd of course want to abstract away as much complexity as possible). However, if digital computers cannot simulate continuous structures with sufficient robustness, then I think AI would have to start putting more research into analog circuits. But I don't think we currently have enough evidence yet to make the case for either.

So yes, brains and PCs have different architectures, but that doesn't mean you necessarily cannot implement a mind on a computer.

BTW I'm looking forward to your upcoming article on Turing Machines vs Brains. :D

Posted by: Jonathan | March 29, 2007 4:06 PM

Thomas & Kurt - I agree with both of your comments, but again my point here was to contrast brains with modern PCs (think of a standard dell laptop). If I am guilty of a straw man fallacy, then this is doubly true of those who take issue with this post for the reason that the brain may *in principle* be the same as a "computer" (i.e., an information processing device). I never argued that the brain is not an information processing device, but for some reason many think that's what this post is about (e.g., Tyler, whose points I completely agree with - b/c he's not criticizing my opinion, but rather one that I've never actually seen anyone endorse. But perhaps we could push shreeharsh in that direction?)

Paul: I've been following your posts on cognitive/epigenetic/developmental robotics for quite a while, and I am also very interested in it. On the other hand, while there is substantial reason to believe that embodiment is important, a lot of the arguments used to support this claim are far too philosophical for my taste (and indeed I am currently collecting experimental evidence against one of the strongest claims for the importance of embodiment in developmental psychology). You'll notice the evidence I present in #10 actually pertains to immersion rather than embodiment (a logical fallacy I permitted myself;). I believe embodiment is important, but I don't think it's actually been proven.

Brian - I saw Randy's Science paper and (probably like yourself) was very surprised. On the other hand, he himself admitted in a recent prosem that the FPGA metaphor for PFC "may or may not" be deeply informative. So I have some difficulty taking that paper's perspective very far.

Lisa & Richard both make the very interesting point that metaphorical reasoning appears to be a necessary component for understanding complex things, in particular the brain.

Posted by: Chris Chatham | March 29, 2007 5:29 PM

"Are you trying to tell me that the amount of RAM available will affect how we traverse a neural network lookup table?" --Steve G

I sure don't remember telling you that...I was responding to architectural differences and the non-issue of how RAM works. Other than that I'm not quite clear on what you were saying.

"Jonathan: Unlike Chris, I'm not buying most of your criticisms.

If you reread what I wrote then you'll see that I stated my belief that it is an open question whether analog can be simulated by digital.

Also when you take issue with "time cost," you're arguing against computational power. Again, I was arguing against the problem of architectural differences. Yes, clearly a massively parallel system would make for a much nimbler machine. I would never argue against that.

And yes, we obviously haven't built neural networks that reflect the robust behavior of living systems. I was only making the claim that it is *possible* to implement real nervous systems (with some amount of complexity abstracted away). In fact I'm only arguing that it *might* be possible.

Posted by: Jonathan | March 29, 2007 6:39 PM

Ah you kids, arguing analog vs. digital while I sit here remembering the hybrid system (two PACE analogs and a Scientific Data Systems Sigma 7 digital) that we used for aircraft simulations back in the 1960s. How soon they forget. :(

Posted by: RBH | March 29, 2007 7:33 PM

Philosophers have created theoretical machines capable of solving the halting problem (for the uninitiated that's a problem computers can't solve). The brain may be a realisation of some super-Turing machine.

IIRC Scott Aaronson argues on his blog and in papers that P != NP could eventually become accepted as an observed fundamental constraint like 2LOT since it seems impossible to find a proof and/or physical systems that would go against. Invoking this would lead to a couple of reasonable and simplifying no-go corollaries, like no closed timelike loops.

If I understand correctly, super-Turing machines has pretty much the same basis. In this perspective it would seem really surprising if they exist.

Posted by: Torbj�rn Larsson | March 29, 2007 8:37 PM

1) It is an "open question" whether any part of reality is actually analog.

2) If reality IS analog, so that human brains are analog, then computers can be constructed in such a way so that they are analog as well. We build computers whose information processing is (approximately) digital because that's convenient for us - but the underlying substrates are the same. Anything we can do, they can do - no matter what speculations we may entertain about brains being beyond Turing capabilities.

Posted by: Caledonian | March 29, 2007 9:54 PM

I believe embodiment is important, but I don't think it's actually been proven.

If you consider that the human being has output systems, be it in movements, writing, speech, or imagination, etc, it stands to reason that the architectural demands of those systems are likely to have an effect on design of a resource constrained brain, as the thoughts and plans are to be enacted through them. On the other hand, the little I've heard about embodiment seems to go too far, and the above is not really evidence as much as it is logic. Embodiment would be part of internal context, btw...

Good posts, keep it up.

Posted by: Martin Persenius | March 30, 2007 4:28 PM

"Are you trying to tell me that the amount of RAM available will affect how we traverse a neural network lookup table?" --Steve G

I sure don't remember telling you that...I was responding to architectural differences and the non-issue of how RAM works. Other than that I'm not quite clear on what you were saying.` --Jonathan

I'm arguing that the way the RAM works is important, because one doesn't need to have the same limitations in RAM behaviour that one has the brain. You *can* model the limitations on a sufficiently powerful computer. But like I said before, you can model a raisin with a housefly.*

The key words there are sufficiently powerful. Because unless you were recreating the wetware, an emulation of the brain on any other architecture is going to be sub-optimal. Why would you want to handicap your amazingly powerful computer?

*What do you call a fly with no wings?
A walk.

What do you call a fly with no wings and no legs?
A raisin. :D

Posted by: Steve G | March 30, 2007 6:52 PM

I am now considering a second post entitled "10 important similarities between brains and Turing machines" haha.

Oooh! I can't wait.

Posted by: The Science Pundit | March 30, 2007 8:27 PM

History is another essential difference, often overlooked. The organization of computers is explicitly designed by people. The structure of brains is not designed but emerges from contact with the world over evolutionary time, developmental time, and in the immediate moment.

Posted by: J D Raoul | March 31, 2007 1:08 AM

If I understand correctly, super-Turing machines has pretty much the same basis. In this perspective it would seem really surprising if they exist.

You're right. It would be really strange, but I'm just pointing out that it's still theoretically possible at this point. To say "We can *probably* simulate the brain in a computer, therefore the brain *is* like a computer and it's just a question of getting a proper interface to hardware" is not a valid argument.

Emotionally, I find the idea that the brain isn't a Turing machine only slightly less distasteful and unlikely than violating the physical domain in explaining consciousness. But the idea of Peyton Manning winning a superbowl is distasteful and unlikely, but it happened too.

Posted by: Steve G | April 1, 2007 5:50 AM

I think, that the post addresses a certain, probaly ruling methaphore of brain computation. This comes from the Turing-model of computers and reflects the understandings of the second half of past century. In this sense, the post urges for metaphore change, and I fully agree with that.

If you go out to everyday informatics you'll hardly find the low-level structures and mechanisms that are used in cognitive sciences, even common systems are much better modelized and understood (see. Universal Modeling Languge, Design Patterns, etc.).

I think, cognitive sciences would gain a lot by recycling the models, patterns, analogies that are used in designed and partly designed systems (like the Internet, and yes, Google is functionally something like the hippocampus). Anyway, in these systems, to get them work (OK, more or less), problems (like asynchronous mixed inputs, information relaying, updating and evolving strucures, interconnection of heterogenous environments, growth, continous work, bottlenecks, prioritizing activities, etc.) must have been resolved in some way. My first candidate is, of course, the Internet, NOT as an analogon to the brain, but as a warehouse of patterns enough rich to reach for better metaphores in the understanding of cognitive processes. Here, you have the unique opporunity to study the evolution of solutions, what forces and processes led to change, what goals could be achieved and what not, and why not, what competing solutions existed, what was the winner, and why (not allways the "better", the Internet "has body"), etc.

I'm less optimistic about the usability of new hardware architectures, or dedicated brain models, they have a couple of design points in place, so can be good to demonstrate a certain aspect, not more.

Posted by: incze | April 1, 2007 11:15 AM

A really great post Chris indeed.
But an "AI-believer" is throwing the gauntlet : A challenge to the AI-deniers.
I cross-linked this posting over there already.

Posted by: Kevembuangga | April 1, 2007 4:19 PM

I think the substrate-neutrality requirement of computationalism should be re-examined. Computationalists work hard to maintain a definition of computation that includes both computing devices and people. But what falls out of that definition is not particular enough to talk in a meaningful way about organismic knowing. That human cognition can be modelled in a limited way using hardware and software does not mean that computation is the best analytic framework for understanding what cephalized organisms do and how they go about doing it.

Posted by: J D Raoul | April 2, 2007 9:15 AM

I am not sure the change blindness literature supports the notion that visual memory is sparse. Dan Simons, a leading change blindness researcher, has recently argued against the notion of a sparse representation. Instead Simons and Ambinder (2005) and other researchers have postulated that people have the information, but it is inaccessible. Hollingworth and Henderson (2002) found memory for changed objects when participants failed to report the changes. Similarly, Mitroff, Simons, and Levin (2004) found memory for both the pre-change and post-change objects. Object viewing times in other studies have also shown what appears to be covert change detection (e.g., changed objects had longer second looks than objects that were not changed). Simons, Chabris, Schnur, and Levin (2002) also found memory for changes in a real-world environment when changes were not reported. Conversely, another reason we may not detect the changes is simply that we perceive a coherently richly detailed world with everything perceived simultaneously, when in fact we do not.

Posted by: Mark Thomas | April 2, 2007 1:13 PM

Mark, thanks for the very interesting comments about change blindness. You clearly know more about that research niche than I do - but I am confused by your last sentence, which seems to support the idea that visual memory is indeed sparse (the point I was originally trying to make): "another reason we may not detect the changes is simply that we perceive a coherently richly detailed world with everything perceived simultaneously, when in fact we do not"

Perhaps my definition of "sparse" includes temporal aspects, whereas yours does not?

I'd be really interested in further clarification on this or other points from you.

Posted by: Chris Chatham | April 2, 2007 1:22 PM

You are correct. I was not including in my definition the exact spatial or temporal aspects of a scene. Instead, we seem to do a good job of quickly grasping the gist of a scene, which I suspect may contribute to failures to report changes. Often, depending on how it is perceived, the gist does not change in change blindness paradigms. Researchers have found that whether a change is detected, or not, is probably a function of how long an object was looked and whether the object's image fell on fovea which is thought to imply that it was attended. By our perception of a rich detailed coherent scene, I mean when we look at the world, we 'see' color all the way across our visual field, but in the periphery we do not have the cones to detect colors. So, we somehow put the color into the part of the scene in the periphery. We also think we have a crisp clear image in front of us, but once you get away from the highest region of visual acuity, we do not see things clearly (the fovea is only 1 to 1.5 degrees of visual angle). Provided it is not too big, if one has a hole in the photoreceptors, the visual system fills in the hole with the background. So with vision, we often know more than we think we know and it may be the researchers, like me, have not figured out how to measure change blindness. I think we have a much better system than I could write code for. But, I would still like to know how we know to look at something if we have not already looked at it and attended to it.

Posted by: Mark Thomas | April 2, 2007 2:16 PM

I put abstract about this post in my weblog in portuguese.
http://www.gluon.com.br/blog/2007/03/31/computador-cerebro/
If you dont like I can remove it.
Thanks

Posted by: Luis Brudna | April 4, 2007 12:13 PM

Ho.. Cool information,

but please translate to Spanish...!!

THANSK!

Posted by: Paris | April 8, 2007 9:59 PM

I think the fact that we can even have conversations at this level of detail shows that we are in fact very close to being able to reproduce the brain, or something able to compute like the brain in the near future. A post like this would have been much, much different even 10-15 years ago, a very short time in the grand scheme of things.

Another thing to consider, is maybe the solution isn't so much in recreating the brain in a computer - I like to use flight as another technology. We've created rockets and jets and shuttles, etc, which acheive flight, but not in the same way that birds do. A simple example, perhaps, but I think if you spoke to scientists 150 or 200 years ago, maybe they would be focused on all the reasons why flight might be impossible, pointing to all the details of why we can't make a machine that is like a bird.

Posted by: John Walter | April 26, 2007 9:56 AM

If I might be so presumptuous, I've recently got round to writing a post on embodiment which I think is quite relevant to the discussion of point #10 (Brains have Bodies). It looks at two fairly extreme views of embodiment, and whether they may lead to "strong AI". It's over at my blog (not including link, as that's just rude! :-) ).

Secondly, if I could respond to John Walters comment on flight as an analogue. Minor point perhaps, but I think that there is a huge difference between trying to recreate the brain and tring to achieve flight. One of the main points of creating an 'artificial brain' is the lessons that may be learned for medical purposes through experimentation which is simply impossible in a biological system. The implementational detail is thus of paramount importance. In trying to achieve flight, we (the human race) wanted the end product - flight - where the implementational details were completely irrelevant. After all, who's going to learn anything about bird flight by examining a rocket? In terms of the creation of an 'intelligent' artificial entity though, I agree with you completely - there is no fundamental necessity to tie the artificial system to a neural system (although the human brain does serve as the best example of an 'intelligent' system).

Posted by: Paul | May 4, 2007 6:52 AM

This is a great article, it really shows where we are in regards to computers. I had previously thought that we had mad great strides in AI emulation, but I know see we still have a long way to go.

Posted by: Dell Computers | May 21, 2007 9:41 AM

This is a great article!

I would like to add one more difference, though it really overlaps your #9 and maybe #10: Computers are constructed from components, while brains are produced by a developmental process. (Furthermore, that developmental process is not only interactive with the environment, but broadly dependent on various features thereof.)

With respect to #3 (modularity), be upshot is that while there are certainly "critical areas" for various brain functions, these don't represent "modules" in the engineering sense. A "module" is a discrete section with a specified function and interface. The various brain regions instead seem to be "ad hoc" processing sectors, with multipath datastreams all over the place.

Posted by: David Harmon | June 28, 2007 5:07 PM

Indeed, as our aircraft don't fly like birds, an AI computer built like the brain will have emergent mind-like properties that are quite different. In fact we already have a computer that is brain-like and self-organizing called the Internet, with PC's and servers acting somewhat like big fat neurons.

Posted by: Robert | June 28, 2007 6:08 PM

Wow. That's a new point of view: an Internet model of the brain, with computers are neurons. Very interesting!

Posted by: MoonShadow | June 28, 2007 11:28 PM

Computers are devices made by man.Try to program a computer to write a good book and achieve to win a Nobel Prize of Literature.We talk about different order of concepts.

Posted by: Jorge Gajardo Rojas | September 20, 2007 6:07 PM

"Difference # 7: Synapses are far more complex than electrical logic gates"

Why did evolution choose such a convoluted route? I know evolution is a blind, random process and doesn't strictly 'choose' anything, but the point is, I would think it would take much longer to evolve a full-fledged neuron than to evolve a simple organic equivalent to the transistor and then replicate it in quantity to match the 'computational' power of a neuron.

Moreover, considering just the space and weight penalties, wouldn't something that seems as unnecessarily convoluted as a neuron soon be replaced in evolutionary competition by something as elegantly simple as a network of organic transistors? Yet that never happened in billions of years of evolutionary history, and if there's a good reason for that, it may not happen now, either.

Posted by: Joe S. | January 16, 2008 2:55 PM

This is a great post. Just bookmarked your site, good stuff.

Posted by: Tech Deck | February 25, 2008 10:38 AM

Great article and a good read.
Although it can be argued (clearly) is the brain not an evolution of the computer? Albeit an incredibly advanced one, but when compared to a computer, the real difference is that we think for ourselves.

Now granted I may be the only one here without a science degree, but it's more me thinking out loud. But if you take away our ability to think for ourselves, a sort of total control sodium penethol, then do we not just become computers that can be controlled, and in turn control a system of devices etc?

Along those lines, you have the software side of us, which I read you argued against, but if you look at it in the sense that we do control our brains like a machine. Subconscious body functions are like the hidden tasks, things we don't do ourselves (because we don't know how to, not intimately knowledgeable of our brains, kind of like Vista...), while our senses act as inputs.

While I do note the mind spawns from the brain, would it not seem logical that a complex biological computer like us, that is assumingly self aware before the development of our mind, would actually go and develop a mind that could take full advantage of the body it is in? Rather than standing around somewhere (ie. mindless human) to die from starvation etc and not increasing it's chances of survival by multiplying? (Although that latter sounds far more virus like than computer like)

I'll cut it off here, I'm trying to curb down on my ramblings. Some of my 2c on the issue anyway.

Posted by: JS | March 23, 2008 8:12 AM

great overview... this reminded me of Hyperion-Endimion novel by Dan Simmons where AI is using human brains to calculate UI (ultimate intelligence - god) :D

Posted by: Goran | March 23, 2008 9:28 AM

Super interesting post I found from Digg. Just like to point out that almost all of this discussion about "computers" is taking place at a very high level of abstraction. At cutting edge technology processes there is nothing simple about a "simple" electrical logic gate. They are subject to many similar continous effects as neurons and much time and effort is spent trying to restrict their behavior to produce "digital" behavior. In this sense, ALL computing is analog, even the digital stuff. It seems that most of the analysis here is overlooking the essential differences between abstract CS and hard EE.

Posted by: Giles | March 23, 2008 10:44 AM

This article has explained in simple terms to normal users, that it is impossible to make some device which could perform at least 10% of what our brain can do.

I guess, it might take around 200 more years...to see some prototype.

Posted by: Mani | March 23, 2008 1:01 PM

Another difference is I can turn my computer off.

Posted by: Matt | March 23, 2008 1:33 PM

There are obvious and many differences between computers and brains in underlying "hardware" and "wetware". However, how the brain works at a macro level, a symbolic level, and ultimately gives rise to self-consciousness for example, has only little to do with the detailed working of neurons and synapses as studied in neuro science. To state that understanding the differences between hardware and wetware may be crucial in the ultimate creation of artificial intelligence is likely an overstatement. The knowledge gap is our lack of understanding of the "symbolic" system model of the brain, and the bottom up model(s) that connect the macro level model to the neuroscience of neurons and synapses at a micro level (e.g. perception, ).

For a more in depth treatise on this subject see the book "I am a strange loop" from Douglas R. Hofstadter.

Posted by: GG | March 23, 2008 1:35 PM

Amazing and a quite interesting read. I have thought for quite some time that the next jump in human evolution will be when the human brain is interconnected with a CPU and associated resources. We may become somewhat of a Borg, but are math skills will be incredible. It may not be so much evolution, but the end results would become truly revolutionary.

Posted by: Jon | March 23, 2008 10:17 PM

Difference #11: Many Republicans have computers.

-Crow

Posted by: crow | March 23, 2008 11:02 PM

Difference #6 it is interesting to see that the brain has a hardware=software coexistance...this is also very similar to an FPGA (field programmable gate array) which changes its structure dependant on the software that it is running

Posted by: Adam | March 23, 2008 11:34 PM

Regarding #6 -- #6 is the one that interests me, as it seems to be the place where "something else" comes into the picture -- that is "something else" besides the brain.

DOES the mind emerge directly from the brain? It seems to me there is something more required -- that is, the brain is part of a living being, whatever that is. Clearly a brain, on its own, without "life", without being part of a whole body, doesn't generate a mind, right?

Where does awareness play into all this? That's the bit that interests me most. I realize that some fiction has postulated that at some point when a computer is big enough or complex enough, it might have awareness. It's an interesting idea, but seems doubtful.

Is awareness a pre-requisite for a mind? Or is awareness generated by a brain/mind? Or both?

I have the distinct idea that having awareness -- or conciousness -- or even, more specifically, self-conciousness -- is a sign of a "being" or "self" being present. That is, I think it is not just a result of having a brain. I think there is "someone" there.

I'm curious is there is any research in this direction, that is, if there are even ways to do research on something I can barely manage to even state!

Posted by: MM | March 24, 2008 5:35 PM

Great. Still accepts the mechanistic (i.e. human) views of a brain-inclusive mechanism per se. It is still a replica of AI as we know it, with all the differences listed.
"Brain" is a tool of mentality, with the rest of it (unknown so far).
Atemporal, aspatial. Concentrating on neurons is a stingy selection.
AI is being designed on the pattern of a so far discovered brainfunction mentality, so the reverse is untrue.
I did not have the guts to read all those lengthy comments. Sorry.
JM

Posted by: JAMikes | March 24, 2008 6:11 PM

Great post. I take issue with just one statement: "short-term memory seems to hold only "pointers" to long term memory." This is inconsistent with cases where brain damage made it impossible for victims to create new, long-term memories while short term memory seemed unaffected. I'm open to plausible explanations; it just seems problematic on its face.

Posted by: Kavan Wolfe | March 25, 2008 12:21 AM

Thanks for this post, and thanks to all the commenters!

Are we presuming all transistors are simple gates, or are we presuming multi-gate transistors?

It seems a neuron is really a simple analog computer. Maybe some research into the old tube pentodes and their marvelous integrating circuits might be in order.

It's like forgetting the average citizen in 1776 was literate in a way we're not now.

Posted by: Ormond Otvos | March 25, 2008 1:49 PM

Your article made the size and quality of the gulf between the brain and computer mimickry even larger than I thought. I posted a comment to Jonah Lehrer's recent article "Blue Brain" in Seed (http://scienceblogs.com/cortex/2008/03/blue_brain.php) to point out some recurring problems with those equating a computational process to consciousness.

Quoting from the article article: "There is nothing inherently mysterious about the mind or anything it makes," Markram says. "Consciousness is just a massive amount of information being exchanged by trillions of brain cells. If you can precisely model that information, then I don't know why you wouldn't be able to generate a conscious mind."

Well, if that is true then would a machine processing somewhat less than massive amounts of information fall short of being conscious? Markram is going too far, logically, in asserting that a set of electrical impulses become consciousness simply because a similar set of impulses is observed to coincide with awareness in a human brain. At most, one could say it is a necessary condition that some such process be present for the associated experience of consciousness to occur in a living brain.

In any case, no process can be said to actually be consciousness, any more than a digital piano can be called a musician, even if it has a massive repertoire with complex programs that make it indistinguishable from a human player. Basically, no quantity of observed correlations between a brain, the information it processes, and what that brain's owner experiences is adequate to get us from the world of encoded information to the qualitative world of someone being informed. Consciousness could be an innate property of all living organisms, or of matter in general, or even be independent of matter, operating as a field that is concentrated and finds an interface in neural networks through some as yet unknown principle.

With a machine merely mimicking a brain's logical processes, however numerous or complex, it is truly preposterous to believe that a mind is at work inside the box. We don't even have a scientific model of what a thought is, let alone a measure of consciousness.

Posted by: Morphix | April 13, 2008 12:31 AM

I recently read a very interesting book called 'Understanding Thinking' which has a chapter (chapter two) on the differences between the ubiquitous general purpose computer and the emergent properties of our neural networks. The author's intention, I think, is to try to wake educationalists up to the fact that neuroscience / AI / neural network simulation has discovered new and previously unimaginable principles about the way we perceive the world, trap experiences, generalise, abstract, learn, etc., which have profound implications for education. It also suggests that we need to start developing a new vocabulary to better convey these ideas as our old ideas about rationality, memory, awareness, consciousness, mind, matter, etc., are rather misleading. It covers some of the same issues as this essay, but from a slightly different perspective, and identifies some other difference as well. You can read that section for free (preview) at www.lulu.com/content/01224917 or you can 'search inside the book' at www.amazon.com (search for 'Understanding Thinking'. Bart

Posted by: Bart Turnwell | May 20, 2008 6:26 AM

Interesting read. Now I'm more confident that robots won't take over the world in the future.
Thanks for the interesting comparisons

FatLoss4Idiots

Keith Maven
FatLoss4Idiots Editor

Posted by: Keith Maven | October 3, 2008 2:19 AM

To be honest i think u'd need a supercomputer just to model one brain cell, and it would still probably fail.

I could make an OR gate with some water tubes and levers... would i call that an 'intelligence'?

We just dont have the technology. In a few hundred years people will laugh at throwing a buncha little OR and NAND gates together and thinking that could ever have intelligence.

Posted by: quinn | November 20, 2008 4:22 AM

You make some good points here. I still like to use the analogy of RAM and short term memory people (though of course they are not the same) because it shows people how important it is as far as processing things. People know that if you increase RAM you can do more tasks at once, where short term memory is similar because it determines how much information you can hold in your head at once. Great article and it was very informative.

Posted by: David Rahimi | November 25, 2008 7:54 PM

There are so many topics to be sorted out here, time is too short.
I would however like to defend Minsky & Papert's work on perceptrons. When that was written (about 30 years ago), perceptrons were being promoted as a realistic paradigm for neural modelling, and in fact were the only neural-modelling paradigm that had been fully worked out. M & P did a superb and devastating job of analysing them better than anyone before them and showing conclusively that they were inadequate; and this was, and remains, a model of intellectual clarity and original thought. M & P did not consider three-layer networks because nobody had considered them at that time, and because they were, after all, talking about perceptrons.

It may be true that their critique had the effect of making neural networks in general such an unpopular topic that the modern analysis of three-layer networks was delayed by a few years, but one can hardly blame M & P for that, or for not developing or presaging an entire new field of research, one that they were were both early to acknowledge to be of immense importance. Science advances by bad, but influential, ideas being conclusively refuted, as well as by great ideas coming to the fore.

Posted by: Pat Hayes | December 3, 2008 9:25 PM

Each of the ten differences is absent in at least one computer design. Thus, the differences are only valid for common classes of computer, as you cannot build a computer to perform non-computable tasks. (Alan Turing proved that one.) Thus, if none of the ten differences introduce non-computability, the differences are purely implementation decisions. As any Turing-complete device can perform the tasks of any other Turing-complete device, those implementation decisions are largely illusionary, they do not alter what can be done, merely how efficiently it can be done.

Posted by: imipak | March 5, 2009 12:44 AM

I really like the article, it was very informative.
Yes, computers are powerful and great. They are super smart and they are capable of things that humans are not normally capable of. But let us not forget that humans were still the one who invented them. So if someone would say that computers are much better than humans, just think that an invention would not be that powerful if it weren't for the inventor.

Posted by: Computer Rental Company | March 20, 2009 7:43 AM

Great article!
The computer can be compared to human brain because of its logical capacities. I would like to think that the computer was modeled after the capability of human thinking. However, the downside of computers is its incapable of imagination. That is one of the obvious demarcation between man and computer.

Posted by: Computer Rental Company | March 26, 2009 5:04 AM

Imipak (#74), you make a very interesting point about Turing equivalence but it relies on your strong premise that these implementational decisions are immaterial to whether the brain is Turing equivalent. In addition, it sidesteps the more important question about whether AI could be achieved without making these same implementational decisions. It reminds me a little of the proof that a sufficiently large 3-layer neural network can approximate any function to arbitrary precision, but that there are not necessarily learning algorithms to determine the necessary weights. Similarly, it may be possible to arrive at some kind of strong AI without making these same implementational decisions, but there's no guarantee that we'd ever be able to do it that way!

Posted by: CHCH | March 30, 2009 9:41 PM

Just off the cuff here, but building on the previous statements regarding Moore's law(which he himself said can't continue forever and doesn't actually translate into doubled processing power), assuming it holds true long enough; we could be looking at 225 million billion calculations per second in just over 100 years. If we base it on the 128 gigaflop Fujitsu "Venus" processor.

Posted by: Will Kerr | May 27, 2009 2:10 PM

Hello and thanks for this...

Not a computer scientist, indeed not a scientist at all, but following links from Neuroanthropology because the brain.computer analogy gives me the serious crazy. Thanks for addressing it...but..I can't understand most of it. [e.g. I don't know what 'isomorphic' means except that it has to do with forecasting the weather. And I don't know how serial differs from parallel, nor what a system clock does in a computer] I know this is meant to be a science blog and maybe aimed over my head, but I would really love to have some ammunition when I get hit over the head with the 'hard-wiring' stuff. Does anyone know of a link to a less technical discussion of this?
thanks.

Posted by: Stephen | September 7, 2009 1:01 PM

Hi Stephen - http://www.thinkquest.org, has a under-19 site in its library called The Computer versus The Brain, created in 2000, that is far less technical, and covers many of the same points. I use it as a reference for my high school computer literacy class. This posting challenges my more knowledgeable students, as the Thinkquest site lacks the breadth and depth of this, as well as the posts that follow.
Hope that helps.

Posted by: Aletta | September 28, 2009 6:23 PM

I think after you study yourself enough we pretty well are computers, I think the ultimate computer would eventually end up being us. being able to develop a machine that can have conscious is a dream for many scientists but they say it will be possible one day. One way to gather more information and increase your intelligence is to try some brain training, good blog with a list of what's new in brain training: http://www.allreviews.com/brain-training/

Thanks,
Kris

Posted by: Kris | October 25, 2009 4:44 PM

Developing Intelligence

Profile

Search

Recent Posts

Recent Comments

Archives

Blogroll

Currently Reading

10 Important Differences Between Brains and Computers

TrackBacks

Comments

Post a Comment

ScienceBlogs

Search ScienceBlogs:

Go to:

Channels

RSS Feeds

More ScienceBlogs

Editor's Picks

Science News from NYTimes.com »

From NYTimes.com