Building a brain inside a supercomputer

Blue Brain is an IBM computer built to simulate a human brain. It's powered by 2,000 microchips, each acting as a single neuron, that enable it to execute 22.8 trillion operations per second. Based at the Ecole Polytechnique Fédérale de Lausanne, the project launched in 2005 to much controversy and skepticism. Modeling the complexity of the brain in a computer is considered a holy grail to some, and hubris to others. The Blue Brain Project isn't an attempt to build an artificial intelligence, although it could someday inform such an effort. That's because the scientists are hoping to use the machine to understand physiology, brain chemistry, and even intelligence and consciousness. The project's stated goal? "To reverse engineer the brain." Here's Markam talking at TEDGlobal this year:

Already though, Blue Brain has simualated the brain's neocortical column, containing 10,000 neurons and 30 million synaptic connection. "The column has been built and it runs," project director Henry Markram told Seed Magazine. "Now we just have to scale it up." In two years, Markram hopes to have modeled a complete rat brain that he will then load into a mobile robot. From SEED:
When listening to Markram speculate, it's easy to forget that the Blue Brain simulation is still just a single circuit, confined within a silent supercomputer. The machine is not yet alive. And yet Markram can be persuasive when he talks about his future plans. His ambitions are grounded in concrete steps. Once the team is able to model a complete rat brain--that should happen in the next two years--Markram will download the simulation into a robotic rat, so that the brain has a body. He's already talking to a Japanese company about constructing the mechanical animal. "The only way to really know what the model is capable of is to give it legs," he says. "If the robotic rat just bumps into walls, then we've got a problem."

Installing Blue Brain in a robot will also allow it to develop like a real rat. The simulated cells will be shaped by their own sensations, constantly revising their connections based upon the rat's experiences. "What you ultimately want," Markram says, "is a robot that's a little bit unpredictable, that doesn't just do what we tell it to do." His goal is to build a virtual animal--a rodent robot--with a mind of its own.

But the question remains: How do you know what the rat knows? How do you get inside its simulated cortex? This is where visualization becomes key. Markram wants to simulate what that brain experiences. It's a typically audacious goal, a grand attempt to get around an ancient paradox. But if he can really find a way to see the brain from the inside, to traverse our inner space, then he will have given neuroscience an unprecedented window into the invisible. He will have taken the self and turned it into something we can see.
Blue Brain Project (EPFL)

"Out of the Blue" (Seed)

Brain Playground Day


  1. ahhh. It’s impossible!!!

    It won’t work. It won’t really be able to think! It won’t have a soul, it won’t really be alive ahhhhhhhh no no no no no no.

    God made humans not computers.

    etc etc

  2. Almost every neuroscientist I’ve spoken to thinks this project is a really unfortunate waste of the talents of gifted people like Markam and those who work for him. We don’t know enough about how the brain works yet for the model to do anything special (like control a sophisticated robot), and we certainly won’t be able to reverse engineer the brain at this stage. Modeling has its place in science, but it’s to demonstrate a mathematical principle, or to show what we claim about our empirical results are true. These things are better done with smaller models where more time is spent thinking about the inputs to the model and the math behind it. Just making the model big doesn’t really accomplish anything special…

  3. Why bother with the messy and inelegant hardware end of the robotic rat? Why not just have the AI run a virtual rat in a rich simulation? Any virtual rat body is going to be better than any real world working model anyway… Also, when this gets further along, virtual models could be run at faster-than-real-life speeds. Faster generations, evolution, super-rat intelligence, then super-human, faster AI overlords, etc.

    1. I think that the reason you build the “messy physical” bit, is that it teaches you whether or not your model is in fact realistic. If you dont do this part you may very well miss important aspects of the problem that are not represented properly in a virtual model. Given the controversial nature of this experiment and this model in particular, it would expecially important.

  4. @tyrell_turing – I disagree. I looked into this project a bit, and it seems that they are creating very careful models of neurons (down to the chemical reactions causing the charge to travel along the axon) which simulate real brain behavior quite closely.

    I agree that the model can’t be considered totally accurate, but if it behaves the same way (same output given specific input) as a real neuron does a large percentage of the time (which they claim it does), then I believe it can definitely provide us with some very useful data and opportunities for experimentation.

    We don’t know how complex an information system has to be to be a “brain”. Some people claim that there is important things going on all the way down to the quantum level within each neuron. Other people say that the only functionally important thing is the ability for a system to have enough logic gates in parallel, thus allowing a more abstract simulation to be functionally equivalent to a real brain (making that simulation a real brain itself).

    The harder part of testing a brain like this is the fact that it is modeling an animal brain. Animal brain’s require experience and environmental input (sensory input) to become fully functioning brains. This is of course possible (one can create a virtual world for the virtual brain), but then that virtual life experience needs to be complex enough to provide the brain with a decent degree of variety within a consistent context of the environment. Also that environment would need to have similarities to ours if we ever want to eventually compare the brain’s development.

    This may be the first baby steps in creating a system to fully recreate an existing brain, with personality, memories, and all the other experiential connections already intact. Woo immortality (prolly not in my lifetime).

  5. @Vagabond

    No, it’s a pity Markram isn’t his middle name and his first and last names aren’t both “I”

  6. Hail the singularity!

    Kurzweil’s predictions of technological advancement say there should be a supercomputer capable of simulating one human brain by 2010, so I guess we’re not that sophisticated yet.

    I do wonder about the physical robot, why not just make a “Matrix” for it to live in. Is it even kosher to call it a living being?

    As Kurzweil’s essay continues, we’ll probably be able to build devices that read our brain’s own synaptic wiring, and then duplicate it into a computer, i.e. do a brain download…

  7. If someone looked into this and found some references in scientific peer review journals I would be very grateful if they were posted.

    I have dabbled in a research project in neuroscience modeling before I settled on a topic for a phd thesis in applied mathematics; from my experience the sortof mainstream academic projects at places like the Courant institute at NYU etc. have goals such as modeling some small part of the visual cortex or some insects olfactory system. This project is most certainly out of the range of what almost any applied mathematician/neuroscientist working in the field would tell you is feasible. Although I’m very skeptical I’m glad there are people dreaming big.

    @p.s. at NETSHARC: Kurzweil’s full of it

  8. Kurzweil said that there should be a supercomputer “capable” of simulating one human brain by 2010. He didn’t say that it would be simulating it. You just witnessed a supercomputer capable of simulating a human brain it just hasn’t been fully programmed yet. The singularity is closer than anyone thinks..

  9. We can’t put the rat in a “matrix” because we don’t know enough about the brain, even a rat’s, to simulate the nerve impulses that would replicate the sensory input, or interpret the impulses being sent for motor control.

    Success with these experiments may get us closer to that, however.

  10. How would a real newborn rat brain develop if it were removed from its body and hooked up to a robot with far more limited sensory and motor capabilities? The results of the experiment could be interesting, but even if the brain simulation is fairly accurate I don’t think the robot can be expected to develop anything near the kind of behavioral complexity seen in real rats.

    1. Been done, actually — with a real rat brain, no less. It eventually got sensory deprivation, and then killed itself.

  11. Are they able to program some basic instincts in this thing first? Because if its a completely blank slate it will probably just sit there. The natural state of all things is to seek the lowest energy state, unless there’s some impetus not to (e.g. to make a higher energy state more attractive).

    Even newborn rats have some basic instincts that they inherit. In mammals, suckling is one of those things built in. Also newborns up until about 6 months or so instinctively close their mouths underwater.

    But probably the most basic instincts in all life are a reward/punishment system (endorphins released on “good” stimuli as a reward to repeat the behavior, and negative reactions to try and stop it), as well as a basic curiosity which causes creatures to try new things. In babies its often “stick an object in the mouth”. In toddlers its mobility and testing with the sense of touch and sound.

    I’m genuinely interested to know, because I do find this fascinating.

    1. > “Are they able to program some basic instincts in this thing first? Because if its a completely blank slate it will probably just sit there.”

      what do you think an instinct IS? it’s neural wiring. they are simulating neural wiring…so yes it would in fact have instincts.

  12. @aspasia We really don’t know what would happen with that many interconnections. If heart muscle cells are any indication(the ones that spontaneously started beating) with a close simulation of the brain things might start happening. Or we might have to trigger a little cascade to get things started. I simply doubt that we won’t get “strong” AI going.

  13. guess it means go is still intractable, else ibm would get more pr mileage by doing that instead.

    this sounds like an ai / fusion / flying cars / call back in the future thing.

    “We have estimated that we may approach real-time simulations of a NCC with 10’000 morphologically complex neurons interconnected with 10×8 synapses on a 8-12’000 processor Blue Gene/L machine. To simulate a human brain with around millions of NCCs will probably require more than proportionately more processing power. “

  14. “It’s powered by 2,000 microchips, each acting as a single neuron”

    Nope, there’s no relation between the chip and the neuron and the implication that the computer is behaving physically like a brain (or vice versa) is false.

    If you had better algortihms for solving the chemical processes you’d need fewer microprocessors. That is, this is a virtual brain inside the mind of a computer.

    “We don’t know enough about how the brain works yet for the model to do anything special”

    We don’t know enough to know whether it WILL do anything special, which is precisely why it should be done.

    I was puzzling over why this story waited a year and a half since SEED (what, David, you can’t send Markham an email?) to deserve a story when I realized it was sponsored. Still it’s a wonderful thing.

  15. “We have estimated that we may approach real-time simulations of a NCC with 10’000 morphologically complex neurons interconnected with 10×8 synapses on a 8-12’000 processor Blue Gene/L machine. To simulate a human brain with around millions of NCCs will probably require more than proportionately more processing power. ”

    My guess is that an awful lot of computing power is being spent on physically irrelevant processes. The tendency when you have a computer that can keep track of all the details is to keep track of EVERYTHING. A little experience shows you where to pare back.

    Remember that it took years to sequence the full genome, something celebrities now do for fun.

  16. @The5thElephant – They are trying to model neurons very closely, but the fact is we still don’t know very much about how neurons actually behave in vivo, so at best the model would be a small window into our current understanding. Markram would be contributing more to neuroscience were he to focus his time on experiments and less on this model (because he is an excellent experimentalist).

    @teufelsdroch – What could it do? It takes huge gobs of processing power to run the numerical integration necessary for these sorts of physiological models, processing power that would be better spent on elegant, simple artificial neural network models that aren’t physiologically realistic. The only way a model of the brain could do something special is if we knew that our modeling was close enough to reality to capture the special aspects of brain function. At this point in time, that is simply not the case.

    1. “They are trying to model neurons very closely, but the fact is we still don’t know very much about how neurons actually behave in vivo, so at best the model would be a small window into our current understanding.”

      I think both approaches are pretty good. You can look at efforts to make “flying machines”. Early models were often bird like, but didnt fly worth a damn. Recreating a bird’s flight abilities were clearly and are clearly beyond our capabilities, even then we didnt have anything like a working model of what birds really do, nevertheless I can still get from Austin to LA faster than my car can drive me thanks to those early efforts. Here there is clearly value in not only creating thinking that more closely resembles what human minds can do (eventually) that is artificial, but in also creating a better model of the thinking humans actually do. The bit about better modeling the effects of certain drugs in the video point out why. Both are valuable, and a succeess of the former and failure of the latter would still be a huge advance.

  17. At my university (University of Canterbury, Christchurch, NZ) there is a Blue Gene supercomputer. There is a human brain programme underway (named the Brain Research Group) that is attempting to model the workings of the human brain. A mathematics PhD student based at the same research institute as myself (the Van der Veer Institute for Parkinson’s and Brain Research) is developing a model of neurovascular coupling – including right down to ion channels and the ways cerebrovascular bloodflow changes as a result of brain pathology (such as Alzheimer’s dementia). It’s an incredibly complex project that will involve scores of researchers all contributing a little part. She can connect to the supercomputer over a network from the research institute and run her programs remotely. It’s far removed from the punch cards my dad used to run his research back in the late 70s :)

  18. This sounds like a great idea. I think we need to test this with the proper rat analog, like maybe our nuclear missile system. Either that, or giant 50′ rats with fricking lasers.

  19. This is hubris!

    A brain is not designed. It is not an artefact. Neither the phylogeny nor the ontogeny of a brain is the product of design. You don’t “build” a brain. A brain is grown. It is evolved at the intersection of blueprint and environment.

    You think you are close to a solution, but it is your conceptual horizon that is close.

  20. It’s almost endearing to see, over and over again, this touching, hopeful, faith that all we have to do to cause “consciousness” to arise is to make things complex enough. Descartes would be *so* proud of his progeny. The first commentator is (despite himself), in fact, correct: this programmed agglomeration will be neither alive nor ensouled, although it may have a fair imitation of both on loan. It will then be a really nifty mechanical doodad capable of doing really nifty mechanical things. Perhaps it will clean my house. It will never take it into its head to decide one day that the ottoman would probably work better in the living room, or that the wallpaper in the bath was a really bad choice. This is because, although it may do things based on “visual” cues, it will never be doing that complex thing called “seeing,” which is a phenomenon not reducible to its component sensory apparatus. And I know how that thought simply galls the true believers who cling to the self-refuting thought that there simply *is* nothing beyond atoms and molecules shoving around other atoms and molecules. Would that it were so simplistic.

  21. otoh, it’s all harmless fun. and if they end up with a good enough model of their neocortical columns, then presumably there’ll be no need to stick wires into rats/cats/monkeys brains in the name of science, which would be a pretty winner result for the rat/cat/monkey community.

  22. Just because you have the ability to do something doesn’t mean you should do it. Should I use the gun inappropriately because I own it, because I have the power to do so? Maybe it’s from watching too many Scifi movies, but somehow a successful AI rat or other creation may have outcomes not anticipated by these scientists. Did anyone think to ask the question, if these people are successful what will be the outcome? Once you learn to model a rat, you are 90% to modeling a cat, then 90% to modeling a dog, then 90% to modeling a human. Do you really want a virtual human? A virtual human (AI) by definition MUST have emotion because not to would be an intelligence without conscience, i.e. pathological. An industrial robot can not have emotions and AI by definition would. The outcome is an intelligent slave. Are we prepared to live on the backs of slaves?

  23. An intelligence is more than the sum of a number of IF/THEN statements of Boolean Logic. It’s one thing to program a repetitive outcome for a robot, but to create an AI would lead to unintended consequences (by their creators) that are totally predictable – slavery. Where slavery exists, revolution follows.

  24. im all for taking the first steps to ai just think at what could be done with greating artificial brains you could creat a new creature but it would be in a robot bodie but who knows what will come in the next 50 years or so maby a dragon bot or something that will act like a dragon or other creatures you would have to have safty measures but still the possibilitys are endless at what you could make with a artificial brain i dont think they will try to harm there makers unless we try to harm them in some way

  25. A virtual lab-rat could be quite useful, especially if you could restore the state, or rewind the simulation and try something different, with a perfect control that had the exact state and previous learning at the beginning of every experiment.
    Using this technology to make a virtual housekeeper that can spontaneously decide to rearrange the furniture seems a bit far away. It would probably be easier to start with something easier, like simulating a lawyer or an advertising copywriter.
    Until someone invents a way to retrofit humans with the kind of telepathy that would allow them to experience an artificial entity’s sense of self, I really don’t care if a simulated neural network is “seeing” in exactly the same way that a person “sees” (which is a different neural networking problem than these researchers are working on) or whether it takes “visual” cues and simply processes the optical input and reacts in a similar way.
    Then again, I’m just a moderately complicated bio/chemical/mechanical doodad capable of doing moderately complicated things.

  26. I tell ya, pretty soon humans are going to be able to fly across the ocean in mechanical birds. Our team has just successfully built giant flapping wings that will allow our birdplane to take off and fly just like a real bird!
    Ok, for all those that don’t get it. To simulate a brain, your objective should be to duplicate its functionality, NOT its structure. Notice how our planes can fly WITHOUT flapping wings. In other words, BRAIN FUNCTIONALITY SIMULATION FAIL

  27. At any rate, I don’t think arguing about the plausibility of the project on the internet has any significant affect on the project’s success. (I’ve seen more intelligent things written on the bathroom stalls). There seems to be this fervent doubt that we cannot duplicate true brain function, as if it were out of reach as too complex (or sacred, which is a seeming sound moral objection), but this is a contradictory notion, as this presupposition of unreachable complexity condescends our own intellect. Are we smart enough to understand ourselves? And, though logic principles and intelligence is arguably possible, is the recreation of sensation? Are emotions rudimentarily chemical reactions we’ve come to endear? There are unresolved philosophical issues of empiricism and reality, and this might shed light on such.

    However, I believe the bigger question is not a matter of “can?” but “should?”. This project evokes ideas of icarus, adam and eve’s “apple”, and Mary Shelly’s frankenstein, among others. The issue seemingly blasphemous to some almost seems moot, outside of the “rat/cat/monkey community”.

    Is this the gravest sin or the holy grail? I was wondering if this would be the door to immortality, for if we could operate indefinitely sustainable (or replaceable) neurons, couldn’t we exist in “our reality bubble” forever?

  28. Nicely said (by previous commenter). However, I don’t see why we shouldn’t. I don’t have anything to say directly against the argument that, for some reason, we aren’t meant to understand ourselves. But the application for such an entity, an emulation of us in a computer system, are ENORMOUS.

    I believe that creating a human brain in a computer (i.e. creating software to replicate it) would be more useful than just to understand more clearly how the brain works – although this should definitely be a major goal.

    What limits are there to the evolution of a computer based brain? If our problem is increasing the size of our brain within the limited space of the skull, a computer that can have additional processors added (and hence more neurons simulated) indefinitely.

    Anyway, if we can create an entity inside a computer, we would be able to directly integrate it into a computer system. If you would give this more thought, why can’t we have computers that can speak with us (as seen in Sci-Fi movies and futuristic games)? I’m probably not expressing this very clearly, but, if you had a human mind directly integrated into a computer system, we would have the easy human-computer interaction.

    If you don’t understand what I’m trying to say (which is probably my fault as I don’t know how to express it), just imagine the possibilities of being able to directly integrate human minds into a computer and be able to tap directly, almost literally hardwired into all the resources technology has made available to us.

  29. Okay, for a while, forget about trying to create a fully functional rat, cat, ape or human brain. Just consider what takes place in the tiny brain of a sparrow as it flies between branches of a tree in order to land safely on it’s nest or evade an attack by a hawk. Or picture a Purple Martin as it maneuvers to snag a mosquito out of the air. Try creating control systems with computers, as we know them today, that could guide a car or airplane with such precision.

    If Markram and his team succeed in creating really sophisticated processing/control systems based on how the brain really works, it would revolutionize safety systems of all types. A functional brain is deeply concerned with protecting its own existence. So picture these types of devices controlling planes, ships, cars, utilities systems. or dangerous manufacturing processes.

Comments are closed.