Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!ix.netcom.com!netcom.com!jqb
From: jqb@netcom.com (Jim Balter)
Subject: Re: What's innate? (Was Re: Artificial Neural Networks and Cognition
Message-ID: <jqbD3r4G1.F7r@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <3g673d$7pl@mp.cs.niu.edu> <D3nJ8n.5rv@spss.com> <jqbD3pCI6.Cu4@netcom.com> <D3pFrv.DAp@spss.com>
Date: Thu, 9 Feb 1995 21:16:48 GMT
Lines: 53
Sender: jqb@netcom8.netcom.com

In article <D3pFrv.DAp@spss.com>, Mark Rosenfelder <markrose@spss.com> wrote:
>In article <jqbD3pCI6.Cu4@netcom.com>, Jim Balter <jqb@netcom.com> wrote:
>>In article <D3nJ8n.5rv@spss.com>, Mark Rosenfelder <markrose@spss.com> wrote:
>>> <jerrybro@uclink2.berkeley.edu> wrote:
>>>>Meanwhile, human language does
>>>>not seem to be the sort of thing that a computer can really
>>>>learn, until the day when it becomes far more powerful than it is today.  
>>>
>>>No problems here, except that the statement that "human language does not
>>>seem to be the sort of thing that a computer can really learn [now]"
>>>would presumably surprise AI supporters, since it's basically a denial
>>>of "strong AI".
>>
>>I don't understand.  "strong AI" is a matter of principle; how can a matter
>>of practice (whether computers are powerful enough today) be relevant to it?
>
>Right: strong AI implies that algorithms, not computers, are intelligent, and 
>that an intelligent algorithm could be run on any machine equivalent to a TM
>(tho' some, such as an Apple II-- sorry, ][-- or Searle sitting in a room, 
>might do it very, very slowly).  jerrybro's statement seemed to imply that 
>some fact about the hardware would make present-day computers unable to run 
>an intelligent algorithm; but that contradicts the hardware-irrelevance of 
>strong AI.
>
>There's a loophole, of course: TMs have to have an infinite tape, and 
>jerrybro may be opining that an intelligent algorithm would require more
>disk storage than any present-day computer could scrounge up.  Possible, but
>rather boring. :)  It'd be more fun if he did mean that something about 
>present-day architectures or hardware precluded intelligent algorithms.

"power" in computers often refers to scaling relative to human activity.
Higher level languages are said to be more powerful than lower level languages
because it is easier for humans to produce algorithms in the former, not
because of some theoreticl difference in computability.  A Cray is powerful
because of its ability to solve problems within some time limit, not because
of some theortical computational advantage.

Computers may not be powerful enough to learn simply because it would take too
long to feed enough inputs or too long to see the results to be practical in
terms of human or real-world limitations.  Perhaps no current computer could
complete a learning task faster than its MTBF.  Or, if "computer" means a
computer system that includes software, this may simply be a comment upon the
power of the algorithms that we have so far developed.  If by "power" you
simply mean theoretical computational power, then the word loses all meaning;
pdp8's, Pentium's, and Cray's are all equally "powerful".

This discussion of course has implications for Searle's comments regarding
computers made of catgut and eggshells or whatever.

"Get real."
-- 
<J Q B>

