Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sun-barr!apple!well!nagle
From: nagle@well.sf.ca.us (John Nagle)
Newsgroups: comp.ai.philosophy
Subject: Re: AI - the real problem
Message-ID: <23399@well.sf.ca.us>
Date: 28 Feb 91 20:05:45 GMT
References: <1434@ucl-cs.uucp> <5219@media-lab.MEDIA.MIT.EDU> <23176@well.sf.ca.us> <6341@rex.cs.tulane.edu>
Lines: 29

feldman@rex.cs.tulane.edu (Damon Feldman) writes:

>	Nonsense.  The brain is structured fundamentally differently
>from a computer and has capabilities that use this structure to its
>maximal advantage (I beleive for a variety of reasons).  Modeling the
>brain with a computer is no more feasible than having a human perform
>all the tasks of a computer.

>	The problems I refer to are things like having 10^10 (or
>something like that) connections among neurons.  A computer cannot
>keep track of 10^10 little peices of information in the first place,
>and if it could, it could not model the interactions that occur among
>all of them many times per second.

      If all we needed was to store 10^10 "little peices [sic] of
information, it would be easy enough.  RAM prices are down to around
$100/MB, so for a million dollars, one can store 10^10 bytes.  And maybe
some could be paged out to disk.  Disk drives are down below $5/MB.

      Hardware is not the real problem any more.  If you believe Moravec's
figures on brain complexity, a Connection Machine has more than enough power to
emulate a mouse.  Yet we still have trouble doing ants.  

      Squirrel-level AI is worth striving for.  Good eye-hand coordination,
good balance, good performance on manual tasks, and about three orders of
magnitude below human brain size.  When we can do a squirrel, we will be
close to human-level performance.

					John Nagle