Will We Ever… Simulate the Brain?

21

For years, Henry Markram has claimed that he can simulate the human brain in a computer within a decade. On 23 January 2013, the European Commission told him to prove it. His ambitious Human Brain Project (HBP) won one of two ceiling-shattering grants from the EC to the tune of a billion euros, ending a two-year contest against several other grandiose projects. Can he now deliver? Is it even possible to build a computer simulation of the most powerful computer in the world—the 1.4-kg cluster of 86 billion neurons that sits inside our skulls?

The very idea has many neuroscientists in an uproar, and the HBP’s substantial budget, awarded at a tumultuous time for research funding, is not helping. The common refrain is that the brain is just too complicated to simulate, and our understanding of it is at too primordial a stage.

Then, there’s Markram’s strategy. Neuroscientists have built computer simulations of neurons since the 1950s, but the vast majority treat these cells as single abstract points. Markram says he wants to build the cells as they are—gloriously detailed branching networks, full of active genes and electrical activity. He wants to simulate them down to their ion channels—the molecular gates that allow neurons to build up a voltage by shuttling charged particles in and out of their membrane borders. He wants to represent the genes that switch on and off inside them. He wants to simulate the 3,000 or so synapses that allow neurons to communicate with their neighbours.

Erin McKiernan, who builds computer models of single neurons, is a fan of this bottom-up approach. “Really understanding what’s happening at a fundamental level and building up—I generally agree with that,” she says. “But I tend to disagree with the time frame. [Markram] said that in 10 years, we could have a fully simulated brain I don’t think that’ll happen.”

Written By: Ed Yong
continue to source article at phenomena.nationalgeographic.com

21 COMMENTS

  1. Everyone on the nay side of this debate acts like 100 billion (5 trillion if you include the glial cells, which were recently discovered to play a computational role) is a gigantic number of components beyond anything we could possibly reach in human-made computers. Have these people forgotten what has actually physically happened in the past few decades of Moore’s law? We could build this thing with present technology, given enough money and space. And 10 years is ages in computing.

  2. Billion Euros worth of research into consciousness, computer science, and engineering. Cool. Oddly enough, it might bring us closer to a cure for cancer than investing in cancer research directly.

    Interesting model, “bottom-up”. A recent article on this site discussed using evolution to engineer enzymes. This has had me thinking about evolution as an engineering approach in general, and I believe consciousness and simulating biological systems can benefit from this approach.

  3. Even if they could build such a detailed simulation, there’s no guarantee it would do anything interesting. Whole sections of interconnected neurons need to work in concert and a very specific, purposeful way to produce behaviour or thought. You can train neural networks to recognise patterns but this is a very guided process and not particularly conducive to interesting, independent behaviour. Not only this, but brains develop physically as they learn – the amount of neurons grow slowly as they organise. (Perhaps Markram’s strategy will take account of this, I don’t know). What a lot of AI research has shown is that some of what we consider to be intelligent behaviour is a consequence of embodiment – that is to say, the fact that the brain is connected to a physical body. Various tasks in animal behaviour are simplified greatly because a creature’s morphology is set up in a particular way. For this reason, I’m not sure that simulating a disembodied mind is the way to go. Admittedly this is my gut feeling as someone who used to study AI many years ago, so there have most certainly been advances in the field which I am not familiar with.

  4. In reply to #4 by Mister T:

    For this reason, I’m not sure that simulating a disembodied mind is the way to go.

    On the other hand, giving it a body and a way of interacting with the world is already super easy. Ideally we’d like a robot with the strength, speed, and physical reflexes of a human body, but Robonaut will do until then.

  5. I think the most important question is:

    Will it suffer?

    And if so, how is experimenting on a simulation of a human brain any different from experimenting on an ordinary human brain? At the very least we should need to obtain its permission first.

  6. In reply to #4 by Mister T:

    Even if they could build such a detailed simulation, there’s no guarantee it would do anything interesting.

    As I read the article it seems to me that its not their goal to “do anything interesting”. That’s not snark, by anything interesting I assume you mean replicating human performance, for example the way Watson can play at an expert level a game that requires lots or what used to be human specific skills, understanding analogies, pop culture references, etc. They aren’t trying to do that:

    “Izhikevich points out that technology is quickly outpacing many of the abilities that our brains are good at. “I can do arithmetic better on a calculator. A computer can play chess better than you,” he says. By the time a brain simulation is sophisticated enough to reproduce brain’s full repertoire of behaviour, other technologies will be able to do the same things faster and better, and “the problem won’t be interesting anymore,”

    The goal, as I understand it isn’t to create a workable AI that can function at or beyond human levels in one or more domains but rather to give us a better idea of how our brains work. For example, a very interesting topic of research is exploring the built in software (or maybe firmware might be a better analogy) humans seem to have for processing common sense information, things that people call “folk biology and mechanics” we seem to have built-in “modules” for understanding things like language (Chomsky’s Universal Grammar), cause and effect, agency, the way objects move, etc. Understanding how these kinds of capabilities develop and are organized in the human brain would be an amazing achievement and my reading of this is that is the kind of thing they are aiming for.

  7. In reply to #6 by Peter Grant:

    I think the most important question is:

    Will it suffer?

    And if so, how is experimenting on a simulation of a human brain any different from experimenting on an ordinary human brain? At the very least we should need to obtain its permission first.

    We’d have to first create the brain, only then ask for permission to experiment. Keep in mind it’ll have no body, and thus no sense organs, so it would only be able to experience emotional pain. (unless scientists stimulate the computer’s pain neurons.)
    At least we won’t have religious groups screaming about pulling the plug (a serious ethical question, actually), because it wouldn’t have a ‘soul’.

  8. I made a machine a trillion times stupider. It wanted to vote republican and believed the world was 6,000 years old…

    Love the boffins work. If Moore’s Law keeps on going, we humans are going to be superceeded very soon.

  9. I think it’s very interesting that he’s attempting to simulate a human brain, as opposed to designing a brain. The human brain is remarkable in that it is extremely quick at certain things, but extremely slow (relatively) when compared to some pure logic devices such as, say, a calculator. I wonder to what extent the human brain is ‘less efficient’ due to evolutionary idiosyncrasies, and therefore what this study would reveal about creating a streamlined virtual or manmade brain. Ie, if you were going design an efficient brain starting with a blank sheet of paper, what fundamental design moves would you make, how would they differ, and how much better could it be? I keep thinking of Deep Thought here….

  10. I wasn’t referring to replicating human performance when I said “do anything interesting”. What I meant was I had my doubts it would do much of anything useful. They may learn a few things along the way, but even that may be very limited in terms of actually understanding brain behaviour with the approach they are using. The majority of simple animal behaviours, even various insect behaviour is beyond our current capabilities to replicate, so my standards of “interesting” are not so high that I would expect human-level performance by any means. Nor I am particularly impressed with feats of pure computation, such as the Deep Blue chess-playing computer – as a technical achievement sure, but not as an example of interesting AI. Even as an exercise in “simply” furthering biological neuroscience, I’m not sure how much they will actually achieve. Admittedly I say this as someone with very little expertise in neuroscience who has studied AI and evolutionary robotics for a time.

    In reply to #9 by Red Dog:

    In reply to #4 by Mister T:

    Even if they could build such a detailed simulation, there’s no guarantee it would do anything interesting.

    As I read the article it seems to me that its not their goal to “do anything interesting”. That’s not snark, by anything interesting I assume you mean replicating human performance, for example the way Watson can play at an expert level a game that requires lots or what used to be human specific skills, understanding analogies, pop culture references, etc. They aren’t trying to do that:

    “Izhikevich points out that technology is quickly outpacing many of the abilities that our brains are good at. “I can do arithmetic better on a calculator. A computer can play chess better than you,” he says. By the time a brain simulation is sophisticated enough to reproduce brain’s full repertoire of behaviour, other technologies will be able to do the same things faster and better, and “the problem won’t be interesting anymore,”

    The goal, as I understand it isn’t to create a workable AI that can function at or beyond human levels in one or more domains but rather to give us a better idea of how our brains work. For example, a very interesting topic of research is exploring the built in software (or maybe firmware might be a better analogy) humans seem to have for processing common sense information, things that people call “folk biology and mechanics” we seem to have built-in “modules” for understanding things like language (Chomsky’s Universal Grammar), cause and effect, agency, the way objects move, etc. Understanding how these kinds of capabilities develop and are organized in the human brain would be an amazing achievement and my reading of this is that is the kind of thing they are aiming for.

  11. In reply to #14 by Mister T:

    I wasn’t referring to replicating human performance when I said “do anything interesting”. What I meant was I had my doubts it would do much of anything useful. They may learn a few things along the way, but even that may be very limited in terms of actually understanding brain behaviour with the approach they are using. The majority of simple animal behaviours, even various insect behaviour is beyond our current capabilities to replicate, so my standards of “interesting” are not so high that I would expect human-level performance by any means. Nor I am particularly impressed with feats of pure computation, such as the Deep Blue chess-playing computer – as a technical achievement sure, but not as an example of interesting AI. Even as an exercise in “simply” furthering biological neuroscience, I’m not sure how much they will actually achieve. Admittedly I say this as someone with very little expertise in neuroscience who has studied AI and evolutionary robotics for a time.

    In reply to #9 by Red Dog:

    On chess playing, one thing I find interesting is that as late as the 1970′s when I was studying philosophy of mind there were still AI critics who held up chess as an example of something a computer could never do. “Oh sure it can master the basic moves but a computer will never play at the Grand Master level because its not capable of the deep strategy and intuition that a human has” people made that claim in all seriousness. Then a computer beat the reigning world chess champ.

    I actually share some of your skepticism about this project although I don’t know enough to have a strong opinion either way. But I’ve seen similar “big leap forward” types of projects, where people asked for and got orders of magnitude more money then computer science usually receive and they tend to not give a very good research ROI. The best/worst example was the CYC project led by Doug Lenat at MCC. They were trying to solve a very different problem but the approach was the same, give us lots of money and we’ll make a huge leap forward. CYC contributed almost nothing of lasting value and the ironic thing is that the areas they were working in: knowledge representation, common sense reasoning, managing large ontologies for heterogeneous domains, all that stuff was intensely relevant to the Internet that was just emerging at the same time as CYC. Similar technologies as were developed for CYC are very important to the emerging Semantic Web, things like OWL and Protege from Stanford. The difference is those technologies were developed in a much more collaborative fashion and with the goal of application to real world problems and CYC never really took real world application seriously.

    So I’m kind of skeptical of this kind of big ticket funding but unlike you I think the problem they are trying to solve is intensely interesting. Also, unlike you, I think AI has made some impressive achievements in understanding neurology and cognition so far. People at DARPA and MIT have designed robots that essentially have the computational power of insects. They can navigate, seek out “food” and light, etc. And we also know quite a bit about how information processing is done by some specific organs, for example how vision is processed. So much so to the point that we can design synthetic limbs that can actually communicate with the human nervous system.

    But we still know very little about how the brain does higher order cognition. How do neurons organize to make things like concepts, rules, and ontologies? Or do those paradigms even make sense when we talk about how the brain works? I think those are fascinating questions and if they can make progress answering them it will be quite an achievement.

    I’m curios what do you consider an example of “interesting AI”?

  12. In reply to #13 by EbeneezerGude:

    I think it’s very interesting that he’s attempting to simulate a human brain, as opposed to designing a brain. The human brain is remarkable in that it is extremely quick at certain things, but extremely slow (relatively) when compared to some pure logic devices such as, say, a calculator. I wonder to what extent the human brain is ‘less efficient’ due to evolutionary idiosyncrasies, and therefore what this study would reveal about creating a streamlined virtual or manmade brain.

    If by “creating a virtual brain” you mean “designing a really powerful (lots of CPU speed and memory) computer” then this research probably won’t have much to say on that topic. The architecture of the human brain (at least to the extent we understand it) is very different then a modern digital computer. Computers are rigid, they have very specific modules for different tasks (number crunching, graphics crunching, CPU, short and long term memory, etc.). Brains seem to be very general. We’ve been able to identify certain regions that are related to certain types of function such as language but only in the most general way. And even in those cases if a human has damage to the language center of their brain they can usually eventually relearn most of their language skills and repurpose them to remaining working parts of the brain. Computer can never do that. If your math processor goes down you can’t divide what it does with what the graphics processor does.

    The analogy AI people often use is birds and aerodynamics. Engineers who design planes don’t try to design them like birds. But it is sometimes possible to study birds and by understanding general principles of aerodynamics apply those principles back to airplane design. A good example is the WWII Corsair fighter and the wing of a seagull. So its unlikely that a direct impact of this research will directly lead to better computers. However, it is possible that by learning more about how the brain organizes and processes information we can learn some general principles about the topic and apply them back to computer design.

  13. On chess playing, one thing I find interesting is that as late as the 1970′s when I was studying philosophy of mind there were still AI critics who held up chess as an example of something a computer could never do

    That is indeed interesting, because it showed that something that we previously thought of as one of the intuitive pinnacles of human intellectual endeavour, could in essence be distilled into a database and “number-crunching” exercise. It highlighted the dichotomy between the algorithmic, Symbolic AI approaches and the more biologically-inspired techniques such as Artifical Neural Networks and adaptive systems. In more recent times, researchers have combined both approaches to achieve some interesting results in various applications.

    … unlike you I think the problem they are trying to solve is intensely interesting

    I think you misunderstood what I was saying. The problem itself, I agree is intensely interesting. I was merely expressing skepticism that their actual findings will result in particularly interesting discoveries given the specific approach they are taking. I freely admit I haven’t read their actual research documents (I’m sure the biology side of it in particular would go over my head). It just seems that they are attempting to create an extremely complex model which is not that well understood, leaving out certain problematic elements (which may be crucial) and then hoping something noteworthy happens which can somehow be related to biological brains. This may turn out to be a valid approach but to me it seems somewhat unfocused and possibly a little naive. Fine for a undergrad/postgrad project but does it really warrant so much funding and attention? It’s appears like one of those New Scientist stories which exaggerates and sensationalizes the importance of particular research and gives the lay-public an erroneous impression about the state of the art. I am fully prepared to eat my words if they find out something new and significant, by the way.

    Also, unlike you, I think AI has made some impressive achievements in understanding neurology and cognition so far

    I don’t know why you’d think I find these unimpressive, I certainly did not say I did or intend to imply as much. Which is why I was talking about specific examples. I’m enjoying the discussion but I would kindly ask that you don’t put words in my mouth.

    People at DARPA and MIT have designed robots that essentially have the computational power of insects. They can navigate, seek out “food” and light, etc

    Yes I am aware of these, having studied some of this research in detail some years ago and made some efforts in that direction myself. There still remains however a vast array of insect and animal behaviours which are beyond the current capabilities of AI modelling and simulation. I am confident that we will see even more impressive advances in these areas over time.

    “But we still know very little about how the brain does higher order cognition. How do neurons organize to make things like concepts, rules, and ontologies? Or do those paradigms even make sense when we talk about how the brain works? I think those are fascinating questions and if they can make progress answering them it will be quite an achievement.”

    I agree, these are exciting and stimulating areas of research. While it would be great if their work makes some headway, I just don’t have much faith this particular work will take us very far in that respect. I would certainly like to be shown wrong.

    I’m curios what do you consider an example of “interesting AI”?

    I am interested in the things you have mentioned regarding neuroscience and cognition, I am equally (if not more) interested in the apparently complex behaviours observed in animals and insects. From the social and cultural behaviours of ants e.g. swarm intelligence, to the “higher level” cognition seen in larger-brained animals and humans.

  14. In reply to #17 by Mister T:

    Sorry, I thought I was picking up a general “AI sucks” attitude in your original comment, shows how easy it is to misinterpret things (or how bad I am at interpreting things).

    Anyway, yes I agree about the Chess example. It shows how hard it is to rigorously define what counts as intelligence and knowledge. For a long time many people (both AI critics and sceptics) thought that the only way to play grand master chess was to make a program that thought like a grand master and some researchers took that approach, doing knowledge engineering on grand masters and trying to replicate strategies like “control the center”.

    Meanwhile the more gear headed types just said “all we need is more CPU power and memory” and they turned out to be right. As you probably know the reason they called it Deep Blue (besides the reference to Hitchhiker) was because it worked primarily by doing brute force deep search (extrapolating many more alternative moves ahead then had been practical before).

  15. Actually I meant the exact opposite! :-) I meant what if you designed a brain from first principles, based on what we do know about how brains work but also on sound intelligent design principles? To what extent could a premeditated brain design be better (or worse) than a brain ‘designed’ by the Blind Watchmaker? Nothing to do with how computers work – they’re machines – but clearly using computers to simulate the dynamic environment within which a ‘designed’ brain might run

    In reply to #16 by Red Dog:

    In reply to #13 by EbeneezerGude:

    I think it’s very interesting that he’s attempting to simulate a human brain, as opposed to designing a brain. The human brain is remarkable in that it is extremely quick at certain things, but extremely slow (relatively) when compared to some pure logic devices such as, say, a calculator. I wonder to what extent the human brain is ‘less efficient’ due to evolutionary idiosyncrasies, and therefore what this study would reveal about creating a streamlined virtual or manmade brain.

    If by “creating a virtual brain” you mean “designing a really powerful (lots of CPU speed and memory) computer” then this research probably won’t have much to say on that topic. The architecture of the human brain (at least to the extent we understand it) is very different then a modern digital computer. Computers are rigid, they have very specific modules for different tasks (number crunching, graphics crunching, CPU, short and long term memory, etc.). Brains seem to be very general. We’ve been able to identify certain regions that are related to certain types of function such as language but only in the most general way. And even in those cases if a human has damage to the language center of their brain they can usually eventually relearn most of their language skills and repurpose them to remaining working parts of the brain. Computer can never do that. If your math processor goes down you can’t divide what it does with what the graphics processor does.

    The analogy AI people often use is birds and aerodynamics. Engineers who design planes don’t try to design them like birds. But it is sometimes possible to study birds and by understanding general principles of aerodynamics apply those principles back to airplane design. A good example is the WWII Corsair fighter and the wing of a seagull. So its unlikely that a direct impact of this research will directly lead to better computers. However, it is possible that by learning more about how the brain organizes and processes information we can learn some general principles about the topic and apply them back to computer design.

  16. In reply to #19 by EbeneezerGude:

    Actually I meant the exact opposite! :-) I meant what if you designed a brain from first principles, based on what we do know about how brains work but also on sound intelligent design principles?

    Well I’m 0 for 2 in understanding other people’s comments on this thread. I see what you mean now, I agree. It occurs to me it might be an interesting exercise to imagine what a human being “designed from scratch” might actually look like. Quite a bit different from the actual thing I imagine.

  17. In reply to #1 by Jos Gibbons:

    Everyone on the nay side of this debate acts like 100 billion (5 trillion if you include the glial cells, which were recently discovered to play a computational role) is a gigantic number of components beyond anything we could possibly reach in human-made computers. Have these people forgotten what has actually physically happened in the past few decades of Moore’s law? We could build this thing with present technology, given enough money and space. And 10 years is ages in computing.

    A fair point, but the continuation of Moore’s law will be running up against the quantum scale soonish, maybe within that ten years. It’s already started, though I’m not sure how long it’ll be before that prevents conventional computers doubling each 18 months. It’s been suggested (possibly predicted) before but logically it must happen some time. That’s not to say computers can’t keep getting faster: quantum computing is a well-established and very active field, and there’ll probably be new ways of increasing yield, dealing with heat problems, etc. However, neither is it certain that QC will simply be a replacement for the current trend of miniaturisation. These things don’t always work out as imagined, hoped or simple continuation of previous developments would imply: remember the histories of AI and nuclear power.

Leave a Reply