[ExI] Blue Brain Project
rpwl at lightlink.com
Mon Feb 8 20:59:17 UTC 2010
Ben Zaiboc wrote:
> Richard Loosemore <rpwl at lightlink.com> wrote:
>> Ben Zaiboc wrote:
>>> Anyone not familiar with this can read about it here:
>>> The next ten years should be interesting!
>> Or not.
>> Markram is NOT, as many people seem to assume, developing a
>> biologically accurate model of a cortical column circuit. He is
>> instead developing a model that contains neurons that are
>> biologically accurate, down to a certain level of detail, but with
>> random connections between those neurons. The statistical
>> distribution of wires is supposed to be the same as that in a
>> cortical column, but the actual wires... not so much.
>> Markram's entire project, then, rests on his hope that if he builds
>> a randomly wired column model, the model will "self-assemble" and
>> do something interesting. He produces no arguments for what those
>> self-assembly mechanisms actually look like, nor does he
>> demonstrate that his model includes those mechanisms.
>> As far as I can tell, Markram's only reason to believe that his
>> model columns will self-assemble is ... well, just a hunch.
> This is pretty much what a human brain starts out as. The brain of a
> baby is *massively* overconnected, and 90% of the connections get
> pruned away as the baby starts to learn. Markram's reason to believe
> that his model columns will self assemble is that that's what they do
> in biological systems.
This is a non sequiteur, surely? Just because the connections are
pruned, it doe snot follow that they were random to begin with.
> Even so. Suppose this is not the case, suppose Markram's random
> interconnections fail to capture something intrinsic to the
> biological situation, it's still not a pointless exercise. We are
> constantly mapping the connections in the brain (the connectome
> project), and while hand-wiring them is out of the question, we will
> surely extract significant statistical patterns that should help with
> setting up the blue brain with more successful starting patterns.
> The blue brain project might not produce the results Markram hopes it
> will, but it can't fail to produce useful information, even if it's
> just "how not to build a brain".
I wouldn't think much of a billion-dollar project, using some of the
world's largest supercomputers, whose goal was to understand computer
chips by modeling randomly wired versions of them, with some vague
promises that in the future some other project will be supplying more
accurate wiring diagrams, and with the fallback position that it would
be "bound to produce some valuable information, even if it's just 'how
not to build a computer'.
Collecting statistical patterns is just an excuse to burn research money
without having to think.
More information about the extropy-chat