 |
|
|
 |
|
|
|
|
DanishMike |
| | | | | | posted 5/2/2010 13:02 |   | | | | | | | | | I'm wondering if anyone has ever succeded in making a neuron-by-neuron computer model of a very simple brain like that of an insect with only a few tens of thousands neurons?
With the term "succeded" I mean having *really* duplicated an insect brain and made it react exactly like a real one, thus beeing able to control an insect-like robot or computer-simulated insect moving about in a similarly computer generated physical environment.
In my opinion this would be the right - or at least a very obvious - way to start when trying to reverse engineer real biological brains. I'm not as such interested in research trying to *understand* how they work, but simply in experiments that "mindlessly" copies them in software form and succedes in making them react just like their biological counterparts.
Strangely, I can't seem to find any information about this and hope you migth be able to drop a link to some webzine or portal dealing with this subject.
Cheers,
Mike
| | | Last edited by DanishMike @ 5/2/2010 1:08:00 PM | | |
| | |
lordjakian |
| | | | | | posted 5/3/2010 07:40 | | | | | | | | | | http://www.youtube.com/watch?v=RLCT3wU4fek
Blue Gene as a computer is pretty old news, and doesn't have to do with insects so much as rats, but it is the closest thing I've heard of in response with your first question. The video says ten thousand neurons.
| | | Last edited by lordjakian @ 5/3/2010 7:43:00 AM | | |
| | |
hunt |
| | | | | | posted 5/3/2010 19:53 | | | | | | | | | | I think it's interesting that so many recent large collaborations are aimed at this goal: reproducing brains. I suppose because it seems a more tractable problem than "What is intelligence? Can we stick it in this computer?" he he he
I know there is a project aimed at simulating a cat brain: http://nextbigfuture.com/2009/11/ibm-has-achieved-cat-scale-brain.html
Then again, where there are high stakes, there is controversy: http://gizmodo.com/5411328/rat-brain-simulator-calls-ibms-cat-brain-simulation-bogus
| | |
| | |
DanishMike |
| | | | | | posted 5/3/2010 20:45 | | | | | | | | | | @Lordjakian
This wasn't quite what I was looking for, but thanks anyway.
@Hunt
Well, since after roughly *seventy* years of experiments and research (Turing onwards) we still don't have a clue how to build genuinely intelligent and sentient machines, reverse engineering real-world brains is beginning to seem like a pretty good idea :-)
The problem seems to be that researchers try to simulate too big brains which returns dubious results. Why even start with a cat's brain when there's more than enough of a challenge in simulating that of an ant or a spider?
I assume that someone by now has done a neuron-by-neuron map of an insect brain, like for instance the banana fly which if I remember correctly has about 50.000 brain cells. Simulating a brain of that size should be doable on a modern supercomputer, right?
In my opinion doing that would be the first real step towards developing truly intelligent and sentient machines.
cheers,
Mike
| | | Last edited by DanishMike @ 5/3/2010 8:48:00 PM | | |
| | |
hunt |
| | | | | | posted 5/4/2010 00:23 | | | | | | | | | | "Well, since after roughly *seventy* years of experiments and research (Turing onwards) we still don't have a clue how to build genuinely intelligent and sentient machines, reverse engineering real-world brains is beginning to seem like a pretty good idea :-)"
You might be right. :)
"The problem seems to be that researchers try to simulate too big brains which returns dubious results. Why even start with a cat's brain when there's more than enough of a challenge in simulating that of an ant or a spider?"
I wonder though if the results would be equally dubious for an insect or spider. What would this brain do that would make me say "Aha! It *is* behaving like an ant." Neuron activity matching? When we invoke neural responses from animals, we usually begin by applying stimuli. How do I stimulate the visual cortex of the ant brain? How do I know I'm doing it in the same way as would happen in an actual ant?
I'm not saying these questions don't have definite answers. But I wonder at the degree to which scientists understand their animal model itself, let alone a simulated copy of it. Especially if they are audacious enough to try a mammal. There are so many question marks when it comes to the mammalian brain, that I don't think modeling will elucidate. Especially if the test for it being a *good* model is mimicry of something that is poorly understood.
It all seems rather circular.
"I assume that someone by now has done a neuron-by-neuron map of an insect brain, like for instance the banana fly which if I remember correctly has about 50.000 brain cells. Simulating a brain of that size should be doable on a modern supercomputer, right?"
Hmm, somehow banana fly doesn't have the same cachet when applying for funding, I imagine.
| | |
| | |
DanishMike |
| | | | | | posted 5/4/2010 10:01 | | | | | | | | | | Hi there Hunt,
"What would this brain do that would make me say "Aha! It *is* behaving like an ant." Neuron activity matching? When we invoke neural responses from animals, we usually begin by applying stimuli. How do I stimulate the visual cortex of the ant brain? How do I know I'm doing it in the same way as would happen in an actual ant?"
I'd say by simulating the *whole* ant and it's environment, not just the brain. Construct a computer simulated ant, complete with legs, feelers, and eyes, and place it in a computer simulated environment, where everything moves and interacts according to the laws of classical mechanics. Then you'd know if the simulation was succesful by letting a specialized biologist observe the ant - or preferrably ant colony.
"I'm not saying these questions don't have definite answers. But I wonder at the degree to which scientists understand their animal model itself, let alone a simulated copy of it. Especially if they are audacious enough to try a mammal. There are so many question marks when it comes to the mammalian brain, that I don't think modeling will elucidate. Especially if the test for it being a *good* model is mimicry of something that is poorly understood."
That's what so wonderful about computer simulations. You don't have to *understand* it in order to simulate it. You just need to be able to copy it. And even only qualitatively so, not quantitatively. There doesn't exist a single quantitatively exact simulation of - say - our galaxy, but still scientists are greatly helped in their attempt to understand its mechanisms through qualitative simuations, where large-scale phenomena emerge from simple physical laws. I'd say this probably applies for brains aswell.
"Hmm, somehow banana fly doesn't have the same cachet when applying for funding, I imagine."
Ok, this got me laughing out loud :-) If you want lots of funding for research, the topic *needs* to have soft fur and big eyes!
Cheers,
Mike
| | | Last edited by DanishMike @ 5/4/2010 10:03:00 AM | | |
| | |
hunt |
| | | | | | posted 5/5/2010 03:05 | | | | | | | | | | "I'd say by simulating the *whole* ant and it's environment, not just the brain. Construct a computer simulated ant, complete with legs, feelers, and eyes, and place it in a computer simulated environment, where everything moves and interacts according to the laws of classical mechanics. Then you'd know if the simulation was succesful by letting a specialized biologist observe the ant - or preferrably ant colony."
That would be an interesting endeavor. I wonder though how long such a simulation would take--of even just one ant. You'd have to break it down into some finite element model, I'd imagine. And carefully determine the number of nodes needed to successfully simulate each part: even something as simple as a feeler can be made ceaselessly more complicated. (Surely, not every seta needs its own node. And need we model at the level of the chemical receptors, or will phenomenological rules that guide feeler nervous system reactions to certain chemical stimuli be sufficient?)
It would be rather ridiculous if a microsecond of ant time took a week of computational time. But then, what are we building these giant supercomputers for, if not to dedicate to modeling more and more complex systems? I'd certainly be interested in the outcome of such a model.
"That's what so wonderful about computer simulations. You don't have to *understand* it in order to simulate it. You just need to be able to copy it."
I guess my skepticism is related to what rubric is used to determine that one has, indeed, successfully "copied" a system. Especially when the original system is so imprecisely understood.
"There doesn't exist a single quantitatively exact simulation of - say - our galaxy, but still scientists are greatly helped in their attempt to understand its mechanisms through qualitative simuations, where large-scale phenomena emerge from simple physical laws."
True--it is precisely such simulations that lead astrophysicists to smack themselves over the head and say, "This doesn't match reality at all. Matter must be missing!" And thus dark matter was born.
But while in galactic simulations the mechanism that guides the time evolution of the system is well-understood (minus that pesky dark matter), it is not so clear which mechanisms of neurological activity are necessary in a brain model and which are not. Though, I suppose, picking your favorite parameters and giving things a go is a nice "experimental" way of answering that question: either it will work, or it won't! (Or worst of all, it won't even be wrong...)
| | | Last edited by hunt @ 5/5/2010 3:08:00 AM | | |
| | |
DanishMike |
| | | | | | posted 5/10/2010 03:59 | | | | | | | | | | After having read a little up on IBM's blue gene computer and professor Modha's brain simulations, I'm starting to feel the same skepticism about the whole thing as you apparently do.
Here's Modha's blog describing his experiments on simulating mouse- and cat-scale brains - It's sounds very promising, as if cognitive computing is just around the corner.
http://p9.hostingprod.com/@modha.org/blog/2009/11/post_3.html
And then there are these responses. Here's Henry Markram's - a colleague at EPFL - email sent to many other colleagues as a response to Hodhas experiments.
http://spectrum.ieee.org/tech-talk/semiconductors/devices/blue-brain-project-leader-angry-about-cat-brain
And then there are these posts over at the Al Fin blog:
http://alfin2100.blogspot.com/2009/11/more-on-ibms-bluematter-brain.html
http://alfin2100.blogspot.com/2009/11/actually-no-ibm-did-not-simulate-cats.html
So now I really don't know what to think. These questions swirl in my mind:
We can build a physical model of a galaxy with millions of bodys, where every object interacts (gravitationally) with every other object all the time. Each interaction requires a call to the square root function, which is computationally expensive.
So, why can't we also simulate an insect's brain with only a few tens of thousands of cells, and where each neuron only interactis with some of the other neurons, and where no cpu expensive square root calls are necessary?
How come a small mammal with a brain requiring only a few watts of energy cannot be simulated by a supercomputer like blue gene, which requires tens of thousands of watts? What the hell is going on?!?
Cheers, at 03:00 in the morning,
Mike
| | |
| | |
tkorrovi |
| | | | | | posted 5/10/2010 09:07 | | | | | | | | | | | |
DanishMike wrote @ 5/10/2010 3:59:00 AM:
How come a small mammal with a brain requiring only a few watts of energy cannot be simulated by a supercomputer like blue gene, which requires tens of thousands of watts? What the hell is going on?!?
| |
Likely, moving molecules takes less energy than switching the electronic gates... Consider, molecules are the smallest entities which can stay in one place, that, and a network-like structure which enables to move them in an organized way. But when even this takes too much energy, then the only alternative are the quantum effects, or some combination of molecular movements and quantum effects.
| | |
| | |
hunt |
| | | | | | posted 5/11/2010 02:42 | | | | | | | | | | "After having read a little up on IBM's blue gene computer and professor Modha's brain simulations, I'm starting to feel the same skepticism about the whole thing as you apparently do."
I think the problem really boils down to publicity vs. science. Science needs the publicity to spur both interest and funding. Unfortunately, outlandish claims of cat brains are more interesting to those outside the academic community, which is really a shame because what they're doing is quite impressive even without all the hyperbole.
"So, why can't we also simulate an insect's brain with only a few tens of thousands of cells, and where each neuron only interactis with some of the other neurons, and where no cpu expensive square root calls are necessary?"
Yeah, I agree. I'd like a clear answer on where current technology stands vs. the implementation of a model of this sort.
"How come a small mammal with a brain requiring only a few watts of energy cannot be simulated by a supercomputer like blue gene, which requires tens of thousands of watts? What the hell is going on?!?"
It boggles how truly inefficient modern computing is vs. the brain. And that's really what the answer is: computers are inefficient.
Think about this: you could simulate all the actions your computer does to make a calculation or display some text on its screen by running through the same algorithm using pen and paper manipulations. But you'd probably take all night to simulate the display of one post--that's a lot of coke and chips to get the job done! Many more calories than your battery is burning through, anyway. ;)
| | |
|
|
|
|
|
