From: Eliezer Yudkowsky (sentience@pobox.com)
Date: Mon Dec 30 1996 - 16:19:15 MST
> Though AM was pretty impressive for its time, I think further
> investigation by grad students trying to reproduce Lenat's results
> showed that the "new" theorems it produced were actually latent
> in the heuristics that Lenat seeded the system with. Something like
> "look for patterns when you divide integers by each other", then
> presto, AM "discovers" prime numbers.
I've heard that a lot, and judging by AM's source code, that's not the
case. The example you gave above, for example, is *impossible*, since
AM didn't start out with the concept of "numbers", just set theory.
"Numbers", to AM, are "bags-of-ones". AM had heuristics like
"investigate extreme cases". Having built numbers from bags,
multiplication from addition, and factoring from division, AM looked for
extreme cases: Numbers with zero, one, two, or three divisors. It
discovered that no numbers had zero factors, only one (1) had one
factor, and that a set of numbers (later called "prime", but to AM,
"bags-of-ones-with-two-inverse-of-multiplications" or whatever) had
two. It also investigated numbers with three divisors (squares) and
numbers with *lots* of divisors.
It is true that AM ran out of steam after a while; Lenat concluded that
this was because AM's heuristics were inadequate. So, after years of
work, he came up with a system that could improve its own heuristics.
It's called EURISKO, and it beat the living daylights out of humans in
that trillion-credit-squadron game.
--
sentience@pobox.com Eliezer S. Yudkowsky
http://tezcat.com/~eliezer/singularity.html
http://tezcat.com/~eliezer/algernon.html
Disclaimer: Unless otherwise specified, I'm not telling you
everything I think I know.
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:35:56 MST