🚀 go-pugleaf

Newsgroups: comp.ai
Subject: Re: "Intelligence"
Summary:
Expires:
References: <952@maestro.htsa.aha.nl> <1130@cbnewsh.ATT.COM> <>
Sender:
Reply-To: oded%H@wiscvm.arpa (Oded Maler)
Followup-To:
Distribution:
Organization: Weizmann Inst. of Science, Dept. Of Math, Rehovot, Israel
Keywords:


In article <> pluto@beowulf.UCSD.EDU (Mark E. P. Plutowski) writes:
>
>How about discussing its relation to another intuitive concept
>(learning) that has a well-defined theoretical analog (Valiant, et.al.)

Valiant's notion of learnability is well-defined but is not so "analogous"
to the intuitive concept for some reasons:

1) It is utility-free, the cost of classification errors is uniform.
2) Learning in this model (as well as in others) is done from "above",
   by someone who manipulates the *explicit* representation of the learned
   function according to the sample.
3) The success criterion is (as in the computational complexity tradition)
   the asymptotic behavior (i.e. ever increasing "size" of "concepts") which
   is not clearly the appropriate one for bounded mortals.

All this does not contradict the usefulness of the model for *certain*
purposes, nor the essentiality of adaptive behavior in "intelligence"
(or in "life" in general).


        Oded Maler
        Department of Applied Mathematics
        Weizmann Institute of Science
        Rehovot 76100, Israel
        (oded@wisdom.bitnet)