[FoRK] On the unreality of bottom-up brain simulation
jebdm at jebdm.net
Tue Aug 17 22:06:31 PDT 2010
On Tue, Aug 17, 2010 at 3:49 PM, Russell Turpin <russell.turpin at gmail.com>wrote:
> On Tue, Aug 17, 2010 at 12:20 PM, Jebadiah Moore <jebdm at jebdm.net> wrote:
> > I agree with what you're saying, but he also was saying that you couldn't
> > bottom-up brain simulation without understanding all of the complex
> > interactions between proteins and whatnot, which isn't true so long as
> > simulate whatever causes those interactions to happen.
> That's a possibility, but cuts across my grain in how simulation
> generally works. Let's take the example of a nuclear blast. It's a
> hard thing to understand. There are simulations of blast effects --
> computationally intense programs -- built by physicists on the basis
> of lower-level physics, e.g., fluid mechanics. Running and validating
> those simulations explains blast effects and provides an understanding
> of them. So to say, "oh, simulate from the bottom up, without
> understanding" .... puzzles me. Maybe. But I'd like to see some
> examples of that.
I don't mean completely without any understanding, of course. But just as
you don't have to understand and describe all the patterns that can emerge
in Conway's Game of Life to simulate it, and you don't have to understand
all the common functions and idioms used by programmers to simulate a
computer, you don't have to understand all the particular functions of each
individual protein in its various roles to simulate a brain. If you
understand physics fully and can simulate it with reasonable efficiency, you
can simulate a brain without understanding anything about a brain in
Obviously this is the case, because otherwise computer models wouldn't be
(very) useful in academic study.
Your example of blast effect models is a good one. If scientists didn't
understand blasts qualitatively, but they did understand the underlying
mechanisms, they could nonetheless build accurate simulations of blasts via
the underlying mechanisms + data input. Then they could use the model to
learn something about the blasts at a higher level of abstraction, of
course, but they didn't need to work at that level to build the model.
I can give an example of something I worked on--plant hormone dispersal at
the cellular level--where I understood the basic concepts involved (passive
and active transport + the calculus), but had basically no experience with
the subject domain itself. I then implemented a working model of the
system, again without having really any idea what the simulation would look
like (other than an educated guess). Knowing about the simulated subject on
that higher level would have helped me get things working better/faster, in
retrospect, and you could probably implement a model at that level without
worrying about simulating the transport systems at all (using a simpler
stochastic model), but the higher level knowledge wasn't necessary.
Am I understanding your question?
More information about the FoRK