Path: utzoo!attcan!uunet!unisoft!achut
From: achut@unisoft.UUCP (Achut Reddy)
Newsgroups: comp.ai
Subject: Re: Limits of AI
Message-ID: <1368@unisoft.UUCP>
Date: 9 Nov 88 23:13:53 GMT
References: <1651@ndsuvax.UUCP> <1666@cadre.dsl.PITTSBURGH.EDU> <3802@cs.utexas.edu> <2413@cs.Buffalo.EDU> <3833@cs.utexas.edu> <2149@bucsb.UUCP> <3876@cs.utexas.edu>
Reply-To: achut@unisoft.UUCP (Achut Reddy)
Lines: 15

In article <3876@cs.utexas.edu> berleant@cs.utexas.edu (Dan Berleant) writes:
>Given this definition, if we can build a machine more intelligent
>than we, we can have a machine whose intelligence is the theoretical
>maximum value of intelligence.

Not necessarily.  There is no guarantee that a more intelligent entity will
be able to build a more intelligent machine than ourselves.

The following analogy will make this clear:
Take the best optimizing C compiler for some new CPU, say the MC88100.
Compile this compiler with itself.  You *may* get a better compiler
(you also may not).  Repeat until fixed point is obtained.  The final
compiler is not necessarily the best that can be achieved.

Achut Reddy