From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!think.com!news.bbn.com!papaya.bbn.com!cbarber Thu Dec 26 23:58:21 EST 1991
Article 2384 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!think.com!news.bbn.com!papaya.bbn.com!cbarber
>From: cbarber@bbn.com (Chris Barber)
Newsgroups: comp.ai.philosophy
Subject: Re: Scaled up slug brains
Message-ID: <4004@papaya.bbn.com>
Date: 23 Dec 91 17:28:58 GMT
References: <352@idtg.UUCP> <40782@dime.cs.umass.edu> <1991Dec18.135620.16540@news.larc.nasa.gov> <357@idtg.UUCP> <4001@litchi.bbn.com> <1991Dec21.055305.23986@morrow.stanford.edu>
Organization: BBN Systems and Technology, Inc.
Lines: 43

In article <1991Dec21.055305.23986@morrow.stanford.edu> 
        dow@nova1.stanford.edu (Keith Dow) writes:

>However an excellent question is how well do the best neural networks
>simulate the brain?

The brain as whole: not at all.  Some networks developed by or in
conjunction with neuroscientists have done a reasonable job on some
of the better understood circuits in the brain.

>Certainly the neural network game can be played without any knowledge of
>brain physics or chemistry.

You seem to be presuming that all of the "computation" going on in the brain
can be explained purely in terms of axonal connections and furthermore that
these have been completely understood and embodied in some type of artificial
model.  None of these assumptions are correct. Of course, in a sense you are
right: the neural network "game" can indeed be played without reference to
actual brains - but not in a way that is relevant to human cognition.

>Also the results of some simulations mimic aspects of brain behavior well.

But many of these simulations are functionally non-interesting because
the only thing they do is to simulate some simple behavior without showing
it to be the consequence of a particular functional approach or linking it
with any other types of behavior. For example, someone (I don't remember who
and my notes aren't in the office) once wrote a "Freudian Slipper" program
that would alter text by replacing one word with a similar one as if someone
were making slips of the tongue. The program produced these slips in the same
proportion that humans do.  Note that these slips were not really "Freudian"
- the word substitutions did not reveal any hidden goals or suppressed
thoughts on the part of the program but were only simple word substitutions
like the usual word-replacement typos one sees all the time on the net. This
program was successful in mimicking the behavior it set out to simulate but
did not teach us anything at all about what really is going on when real
humans make slips of the tongue.  Such simulations are toys only (it is too
bad AI 'research' is rife with this kind of stuff).



-- 
Christopher Barber
(cbarber@bbn.com)


