Re: Genius Dogs

Nicholas Bostrom (bostrom@mail.ndirect.co.uk)
Fri, 10 Oct 1997 17:22:55 +0000


Hal Finney wrote:

>[snip]
>Interesting, although of course we all know that the infinitely fast
>computer doesn't exist anyway.

How do we know that? (Do you mean that we know that no infinitely
fast computer can exist physically, or that no such machine will in
fact ever exist in our universe, or that the decendants of humanity
will never build such a machine?)

> A brute force technique, then, would have to simulate not only the brains,
> but potentially a significant subset of the biosphere. This would be many
> orders of magnitude more difficult. Volume considerations alone would
> suggest 4 or 5 orders of magnitude (what fraction of the biosphere is
> made of brains?). A complicating factor is that the kinds of simulation
> necessary or appropriate for things like muscles, trees and rivers may be
> fundamentally different than the presumably neural-net based simulations
> we would prefer to use for brains (and which probably formed the basis
> of Drexler's estimate).
>
> The idea of evolving intelligence does seem to have potential, and may
> turn out to be the easiest way of developing AI. I'm sure that Drexler's
> example is not meant literally, but is intended to motivate the intuition
> that given nanocomputers, it should become practical to evolve AI. This
> may still be the case even if the example doesn't go all the way towards
> establishing the point.

Even apart from the fact that Drexler's estimate doesn't seem to
have taken into account that the simulated creatures would need an
environment to live in, it is clear it could never be a certain way
of quickly evolving AI. There are something like 10^18 habitable
planets out there, and as far as we know, intelligent life has
evolved on only one, and we don't know how implausible that was.

Another point is that I think that even though the volume of the
non-brain parts of the biosphere are many orders of magnitude greater
than the brain parts, it doesn't follow that it would take so much
more computing power to simulate it. Brains are the most complex
systems in the known universe; other biological phenomena could
presumably be sufficiently closely approximated at a relatively low
computational cost. Of course, if we start to make idealisations then
the estimate loses its plausibility as an absolutely certain upper
bound on how difficult it is to create AI.

However, I don't see an reason why we should have a realistic
biosphere in our simulation at all. Since the simulate
evolution-argument doesn't seem to work anyway, we can just as well
focus on more realistic ways of creating AI. Some kind of strategic
multi-player game in an environment where some sort of simulated
"physical" things can be constructed would seem a better place to
evolve intelligence in than any realistic model of primordial earth.
But even that is not how I expect AI to come into existence.

Nicholas Bostrom
http://www.hedweb.com/nickb