Re: COMP: AI substrate

Eugene Leitl (
Mon, 26 Jul 1999 11:24:55 -0700 (PDT)

Eliezer S. Yudkowsky writes:

> Okay, so it's a damn big pattern-catcher, but it's still just a
> pattern-catcher. I don't think AI is that simple. It sounds large, in

It's not a pattern catcher. It's an all-purpose architecture (tho' it has got a few blind spots on the performance scala). It is not an AI, just a framework for an AI/a part of boundary conditions for the emergence of one.

One would still need a good bootstrap plan.

> absolute terms, but relative to what already exists:
> 1. Raw power is about the equivalent of, what, a rat's brain?

It is difficult to link GBytes & MIPSen to neurons without calibrating this first (i.e. looking how this thing can solve problems vs wetware). My point was to illustrate what we can do NOW on modest budget with industrial commodity silicon, then just point to Moore's law, and giggle.

> 2. Self-programming algorithm... around the spider level, I'd guess.

It's not an algorithm, it's a meta-algorithm.

> If intelligence was that simple, rats would be sentient.

If that thing performed as intelligently and as robustly as a rat, I could get all the funding I want. Autonomous planes, tanks, robots, you name it.