Re: Mike Perry's work on self-improving AI

Matt Gingell (
Wed, 8 Sep 1999 02:35:16 -0400

From: Eliezer S. Yudkowsky <>

>There's a subjective factor involved in judging between what constitutes
>a "knowledge base" and what constitutes "intelligence", but any sane
>AIer should be able to distinguish them easily. Previous AIers will
>argue interminably about philosophical definitions and triumphantly
>point out minor crossovers; also, they'll conflate the distinction
>between intelligence and knowledge with the distinction between
>procedural and declarative knowledge. The paradigm of pragmatism lets
>us ignore them, since the distinction is obvious most of the time. The
>practical distinction between intelligence and knowledge is easy enough,
>even when both take the form of declarative data. If the static data is
>used in analogies and similarity analysis, that's knowledge; if the data
>is used to direct operations on other data, if it contains the pattern
>of nontrivial procedures that can operate on data, that's intelligence.
>Again, those are just practical guidelines, not definitions.

I think youíre making a bogus distinction Ė whether a piece of data is knowledge or, to use your word, intelligence depends on how itís being used. If I pass a string of bits to a compression algorithm, itís data. If I feed the same string through a Turing machine, itís an algorithm.

To use a high level example, a concept like Ďcarí is a pattern matching elements of the infinite set Ďcars.í The same pattern can be used generatively to produce instances of that set, perhaps matching constraints imposed by some creative process. Such a pattern must surely be described algorithmically.

Elements of the learning substrate would certainly treat such a concept as data, however. For instance, If there were two sets, one matching red cars and one matching every other kind, the learning system would have to notice the abstract overlap and unify the two patterns.

I donít know why you think itís useful to talk about classifying these things. Learning generates knowledge which in turn influences future learning, to the point where itís angles on a pin to ask whether itís data considered in the process of learning or programs driving the learning machinery. In a very real sense, we learn how to learn.