dagblog - Comments for "Build(ing) a Supercomputer Replica of the Human Brain"

In my opinion computational

Verified Atheist — Tue, 21 May 2013 19:15:29 +0000

In reply to Thanks for the info. Suppose

In my opinion computational constraints are not the large pole in the tent. Neuroscientific knowledge is, as you allude to. I think that we might actually be able to model the vast majority of our 10 billion neurons as Izhikevich type neurons, which we already have enough computational power to do (on super-computers, that is). The problem is, what parameters do we use for those models? How do we connect them? How do they respond to coincidental firing (i.e., how do they learn)? An approach I used in my research was to take a wide range of possible values for many of those parameters and let evolution find viable solutions for me. That doesn't mean those parameters are going to necessarily model the internals of a rat brain, but they'd model how the brain behaved, where by "behaved", I mean mimicked the neural firing pattern observed in rats under similar stimuli. My results were interesting, but I wouldn't exactly call them earth-shattering. Just don't tell that to my dissertation committee, as they thought it was good enough for my PhD.

Thanks for the info. Suppose

Michael Wolraich — Tue, 21 May 2013 17:44:45 +0000

In reply to The million dollar question

Thanks for the info. Suppose computational constraints weren't an issue. Are there other limitations? How do you read the entire neural structure of an actual brain?

The million dollar question

Verified Atheist — Tue, 21 May 2013 12:08:36 +0000

In reply to Having studied a little bit

The million dollar question (or possibly billion dollar) is how much detail do we actually need to reproduce the cognitive behavior of the brain. Can we get by with neurons as abstract as McCulloch-Pitts, which don't even have integrate-and-fire capabilities? My guess is no. Can we get by with Izhikevich neurons which require about 10x as many computations as McCulloch-Pitts, but significantly fewer than Hodgkin-Huxley? I hope so. It's possible that not even Hodgkin-Huxley or even models based on cable theory will be sufficient. Will we actually require quantum computers to mimic quantum behavior of the brain? My guess is no, but at this point, it's just a guess.

Once we can adequately model a rat brain, we're only a few years from being able to model a human brain, in my opinion, as most of the fundamentals are essentially the same. A significant difference, I suppose, is that we're in a much better position to notice when the models of the human brain are slightly off, so that we might fool ourselves into thinking we've adequately modeled a rat brain when we actually haven't.

Having studied a little bit

Michael Wolraich — Mon, 20 May 2013 22:51:33 +0000

In reply to As someone who has worked in

Having studied a little bit of AI from the psychology-philosophy side, I definitely think biological models are the way to go. The folks who want to reproduce the functionality of the mind from heuristic models are totally clueless.

That said, do you think we're even close to being able to model a rat brain, let alone a human brain? My wife studies T-cell receptors, and she doesn't even have enough computing power to model a complex molecule. So I have to assume that the models Markram has in mind are relatively abstract, and I wonder if they'll actually simulate the biology effectively.

As someone who has worked in

Verified Atheist — Mon, 20 May 2013 18:47:47 +0000

In reply to Build(ing) a Supercomputer Replica of the Human Brain

As someone who has worked in the field making my own models of regions of the mammalian brain, I feel compelled to state that until he's finished it's still a hypothetical debate. I'm confident that we'll eventually get there, but I'm highly skeptical that we have enough neuroscientific knowledge to get there within a decade. (I do think we have the computing power, however.) The good news is that trying to get there within a decade should add significantly to that neuroscientific knowledge. (I suppose the bad news is that we're far more likely to create all kinds of "insane" models before we create our first "sane" one.)

What is really scary is that

EmmaZahn — Mon, 20 May 2013 13:19:54 +0000

In reply to Build(ing) a Supercomputer Replica of the Human Brain

What is really scary is that he is first going to build a rat brain. Kind of adds a whole new dimension to the Terminator franchise, no?