Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!rutgers!att!cbnewsh!mbb
From: mbb@cbnewsh.ATT.COM (martin.b.brilliant)
Newsgroups: comp.ai
Subject: Re: Therapy for Carbon-Based Neural Networks and Silicon-Based Machines
Message-ID: <1617@cbnewsh.ATT.COM>
Date: 21 Jun 89 14:07:31 GMT
References: <1420@lzfme.att.com>
Organization: AT&T Bell Laboratories
Lines: 39

From article <1420@lzfme.att.com>, by jwi@lzfme.att.com (Jim Winer @ AT&T, Middletown, NJ):
> .....
> Forcing an abreactive state for a machine intelligence would involve
> simulating an earlier surround in which there is a highly charged
> emotional state. For sequential computing devices, this seems
> unlikely. For a neural net this might correspond to a period of
> intense negative feedback.....

It shouldn't be that hard.  The therapeutic situation Jim described is
basically what happens when a learning system makes an invalid
generalization.  Suppose it mistakenly concludes that a certain set of
conditions is too dangerous to allow at any time.  Sure, if it is a
simple system you go in and do a manual adjustment, and a man-made
system should be built so you can do that, but suppose for some reason
you couldn't do that.  You would have to force the system to go back
and re-evaluate that supposedly dangerous situation.

> ...... Thus, in a complex net, changing a strong
> early pattern of response might have interesting effects on the
> response to later patterns (if not totally destructive). I wonder if
> there would be any behavior changes that might relate to the type of
> behavior changes that result from putting a human through an
> abreactive crisis?

Yep, I think that's the right question.  I'm suggesting that the
machine would resist therapy at first.  If it consented to review its
assumptions, it might have to unlearn and relearn everything it learned
after it made the incorrect generalization.  It might behave rather
immaturely for a while.  I'm not thinking necessarily of a net, but
more of an expert system that learns from its Q and A's.  To stay sane,
such a system would have to remember when and how it learned what it
thinks it knows.

M. B. Brilliant					Marty
AT&T-BL HO 3D-520	(201) 949-1858
Holmdel, NJ 07733	att!hounx!marty1 or marty1@hounx.ATT.COM

Disclaimer: Opinions stated herein are mine unless and until my employer
	    explicitly claims them; then I lose all rights to them.