Re: Topologi of a SI AI (Was:Re: extropians-digest V7 #341)

From: Eugen Leitl (eugen@leitl.org)
Date: Sat Dec 14 2002 - 10:50:25 MST


On Sat, 14 Dec 2002, Robert J. Bradbury wrote:

> Not really, one of Eric's nanocomputers has the capacity of between
> 100,000 and 1 million human brains and its up against the power supply

In ops (bit flux), not in total bits stored. Nitpick not applicable to
this particular statement; yet handy to keep in mind when using Drexler's
estimates for 'how many people in a glass of water' arguments. One
person/glass of water at the very least. One person/sugarcube? Maybe,
maybe not. Besides, these are current, skinny people. A future person (if
indeed there is an organization hierarchy level comparable with today's
people) could be cubic meters, or cubic kilometers.

> and cooling capacity of known physics. So it isn't "impossible" to
> understand. Its simply way beyond current capacities. But I can grok
> 10^6 people -- its somewhat smaller than the population of Seattle or
> Boston and I've lived in both of those places.

http://www.kokogiak.com/megapenny/
One MegaMoo: http://www.kokogiak.com/megapenny/twenty.asp

> There are distinct differences between architectures and the capacities
> of Jupiter Brains and Matrioshka Brains (I designed MBrains to deal with
> a larger spatial distribution and some of the possible cooling problems
> that JBrains might have). Anders has discussed several other possible

Things do become considerably denser and faster with strangletronium (some
jokers have purported strangletronum clump passage through Earth
manifesting themselves as a linearly propagating quake s in duration).

> architectures in his JET paper from a few years ago (though he has
> to depend on some pretty creative physics if I recall). Both Michael
> Frank and Seth Lloyd have done some interesting work over the last
> few years regarding the theoretical limits.
>
> See: http://www.aeiveos.com/~bradbury/Authors/

Right, no one mess with physics of computation. It's as real as science
gets.

> We will never be able to explore the vectors of "intelligence" completely
> (even with an MBrain). The phase space of what can be constructed using
> robust nanotech is simply too large. The best we will ever be able to
> do is pick some promising directions and try to explore those as best
> we can.

Behaviour is evolution in configuration or state space. Which is pretty
much damn infinite, yes. Flagging it all 'intelligence' is missing the
mark wildly.



This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:58:44 MST