Newsgroups: comp.ai.philosophy
From: ohgs@chatham.demon.co.uk (Oliver Sparrow)
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!news.duke.edu!convex!cs.utexas.edu!swrinde!pipex!demon!chatham.demon.co.uk!ohgs
Subject: Re: Is Common Sense Explicit or Implicit?
References: <357e2u$f8k@mp.cs.niu.edu> <35alp4$4r9@agate.berkeley.edu> <Cw8oK7.BB7@gpu.utcc.utoronto.ca> <35d6i2$5fd@agate.berkeley.edu> <CwE3Gq.3xH@gpu.utcc.utoronto.ca>
Organization: Royal Institute of International Affairs
Reply-To: ohgs@chatham.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.27
Lines: 29
Date: Tue, 20 Sep 1994 11:14:02 +0000
Message-ID: <780059642snz@chatham.demon.co.uk>
Sender: usenet@demon.co.uk

People seem to be using "language" in a very restrictive sense. THose
who analyse the structure of human affairs often talk about "semiotics",
about the underlying structures and icons from which a text is built and
to which it can be deconstructed. The degree to which this can be believed
is a matter of taste, but the fundamental point - that meaning comes in
complex little capsules which we mix and blend in order to create ideas (and
that the specificity of words is a post hoc evocation of this inaccessible 
mental event) seems undeniable. It is undeniable not least because means of
comunication which transcend words - images, odours, music, diagrams - are
so much more immediate and powerful (if less precise) than words. 

Common sense knowledge, the subject of this interminable thread, is as 
embedded in these icon bundles as it is expressible in words. A dog learns, 
a goldfish learns: not maybe, but repeatedly and in ways open to manipulation.
They do so in ways that differ from machine learning, for they discover what
it is about their environment that matters, discover how to detect instances
of this and build patterns of behaviour which are adaptive to these 
circumstances. Machines do this, provided someone supplies the objective 
function - "back propagate against this! Not the following flag! Here is a 
data hierarchy!" Machines have not - in all but a few instances - been 
designed to have a "!" detector, something which decides that a noteworthy 
event in in progress and that data streams should be searched from cross 
correlation, related forms of arousal and the like. Herein the weakness: NI 
has learned how to learn, to define its own contextuality. AI has not.
 
_________________________________________________

  Oliver Sparrow
  ohgs@chatham.demon.co.uk
