From: Ben Goertzel (email@example.com)
Date: Sun Jan 28 2001 - 12:45:11 MST
> I *know* evolution is a general and powerful process. It's the process
> that produced me, and I acknowledge that. I just think that intelligence
> is a *more* general and *more* powerful process. If evolution gets in the
> way of my declarative goals, or the declarative goals of a Friendly AI,
> then we'll *deal* with it, and without too much trouble. The emergent
> forces of evolution are unintelligent. I can outsmart them; a Friendly
> seed AI can outwit, outclass, and outgun them; a superintelligence can
> simply crush the forces involved via brute brainpower.
And this is what I don't believe.
When you say "brute brainpower", you're referring to a rather underspecified
collection of cognitive algorithms
My mathematical conjecture is that the ONLY effective algorithm for general
inference control is EVOLUTION.
I actually tried a while to prove this, and failed -- the mathematical
just doesn't exist
I could be wrong of course...
> So when people argue that the character of post-Singularity reality is
> *still* going to be determined by evolution, on the grounds that evolution
> is going to sneak in and stomp on the goal systems of the transhumans that
> will enact the new world, I just don't buy it.
My argument is different: the thought process is intrinsically evolutionary,
and so the mutation and "genetic drift" aspects of evolutionary optimization
will be present
in the course of development of AI's even if there is no evolution on the
but only within the long-term development of individual Ai minds.
> Sure. Hedonic neurons, directed evolution in the human immune system...
> like I said, evolution is a very powerful method. Intelligence is *more*
> powerful. You just don't get to see that unless you have access to
> general intelligence - not algorithms, not classical-AI heuristics, but
> true general intelligence. Wait until Webmind starts writing vis own
> treatises on seed AI before you conclude that an evolutionary goal system
> is the best possible method...
Well, you could be right. But I doubt it. We'll see ;>
This archive was generated by hypermail 2.1.5 : Thu May 23 2013 - 04:00:20 MDT