Newsgroups: comp.ai,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!udel!news.sprintlink.net!demon!chatham.demon.co.uk!ohgs
Subject: Re: Is Common Sense Explicit or Implicit?
References: <1994Sep1.115848.20788@unix.brighton.ac.uk> <344o0o$qcd@mp.cs.niu.edu> <34o30t$5kc@percy.cs.bham.ac.uk> <34qanf$qvt@mp.cs.niu.edu> <34uqtd$fkg@percy.cs.bham.ac.uk>
Organization: Royal Institute of International Affairs
Reply-To: ohgs@chatham.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.27
Lines: 41
Date: Mon, 12 Sep 1994 08:19:15 +0000
Message-ID: <779357955snz@chatham.demon.co.uk>
Sender: usenet@demon.co.uk
Xref: glinda.oz.cs.cmu.edu comp.ai:24218 comp.ai.philosophy:20306

In article <34uqtd$fkg@percy.cs.bham.ac.uk>
           A.Sloman@cs.bham.ac.uk "Aaron Sloman" writes:
  
  >  Whether believing that P is a matter of degree is not to be settled
  >  by verbal or philosophical or psychological research, but by
  >  designing different architectures for agents capable of having
  >  belief-like states and then seeing whether the architecture permits
  >  anything other than binary options. 
  >  [chop]
  >  These sorts of considerations lead me to conclude that here as in
  >  many other contexts the words of ordinary English with which people
  >  formulate questions about mental states and process (in AI,
  >  psychology, philosophy, etc.) lack the depth and subtlety for the
  >  task. They need to be replaced by architecture-based concepts that
  >  relate to underlying states and processes.

I agree entirely with this view. The interesting thing is, to me, that the
architecture is self-referential: it is made out of the things to the
description of which it contributes. (In Aaron's earlier entry, locating
a point in a city depends on other points that one knows about in the city, 
themselves predictaed upon the point in question). I can only believe that
such systems arise by considerable iteration (indeed, learning) during which
the features which matter are extracted.

This need not be the source of alarm, however, for there are many tools by 
which we identify these features: cluster and principle component analysis, 
reversed monte carlo econometrics, for example. There are many "softer" 
methods which are also in use: concept and linkage mapping, dimensionality 
extraction: all of these are intended to make explicit an underlying model. 
They do so for reasons quite distinct from the defintion of or reverse 
engineering of cognitive processes, but point to the degree to which such 
procedures could be built into a designed system such that it would have the 
properties that allowed for the directed expansion of its linked models of 
the operating environment in which it was required to function. It suggests, 
however, that focus upon how to learn, not on the embodyment of information, 
which should be the key issue of interest. 

_________________________________________________

  Oliver Sparrow
  ohgs@chatham.demon.co.uk
