From newshub.ccs.yorku.ca!torn!cs.utexas.edu!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!agate!netsys!pagesat!spssig.spss.com!markrose Fri Oct 30 15:17:56 EST 1992
Article 7421 of comp.ai.philosophy:
Newsgroups: comp.ai.philosophy
Path: newshub.ccs.yorku.ca!torn!cs.utexas.edu!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!agate!netsys!pagesat!spssig.spss.com!markrose
>From: markrose@spss.com (Mark Rosenfelder)
Subject: Re: grounding and the entity/environment boundary
Message-ID: <1992Oct28.180841.1257@spss.com>
Sender: news@spss.com (Net News Admin)
Organization: SPSS Inc.
References: <1992Oct27.205040.117959@Cookie.secapl.com> <1992Oct28.000703.5993@spss.com> <Bwu9M3.9CM.1@cs.cmu.edu>
Date: Wed, 28 Oct 1992 18:08:41 GMT
Lines: 39

In article <Bwu9M3.9CM.1@cs.cmu.edu> rudis+@cs.cmu.edu (Rujith S DeSilva) writes:
>In article <1992Oct28.000703.5993@spss.com> markrose@spss.com (Mark
>Rosenfelder) writes:
>>Actually it may be human memories that outperform computers'.  How often
>>does your heap get corrupted?  How often do you crash due to a page fault?
>
>My computer goes down pretty often, but the nice people at facilities bring it
>back up to exactly the same state each time.  :-)

If they re-created *exactly* the same state, it would crash again, eh?

>You're missing the point that human and computer memories fail in
>fundamentally different ways.  With computer memories, it's possible to get
>any desired degree of reliability.

I'm not missing any point-- I'm disagreeing with you.  The notion that
computer memory is reliable over timespans in which human memory is not
is completely absurd.  Not only have computers never been tested at the
70-80 years over which humans can operate, but in fact computer memory is 
*very* fragile.  If you don't think so you're obviously not a C programmer...

Disk drives are more reliable than RAM, of course; but if the computer
can appropriate the reliability of its peripheral storage devices to itself,
so can I: humans have reliable external storage devices called "books"
that are good for a few thousand years at least.

However, the interesting comparison is not between humans and computers but
between humans and AIs, and of course this cannot be done till we have an 
AI in hand.  Once we do we might not find so much of a difference.  A
neural-net-based AI might have the same kind of impressionistic, flexible,
unrigorous memory that we do.  Even if not, we might find that selective
forgetting is necessary for the AI to keep its knowledge manageable and
within its storage capacity.

Of course, even a connectionist AI might find it useful to have access to 
a CD full of encyclopedic data.  But then, we might like that ourselves.
In the future, the most with-it AIs and humans each might have a memory
that's an amalgam of neural net (for storing and interpreting real-world
experience) and RAM (for storing vast masses of digital information).


