Newsgroups: comp.ai.alife
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!news-feed-1.peachnet.edu!gatech!howland.reston.ans.net!pipex!demon!chatham.demon.co.uk!ohgs
Subject: Re: On the color of "red" (was: Re: The Meaning of Life)
Distribution: world
References: <35d8uo$q5k@Germany.EU.net>
Organization: Royal Institute of International Affairs
Reply-To: ohgs@chatham.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.27
Lines: 52
Date: Mon, 19 Sep 1994 12:47:21 +0000
Message-ID: <779978841snz@chatham.demon.co.uk>
Sender: usenet@demon.co.uk

Goethe's book was not entirely ignored, insofar as Seurat (etc) founded
pointillism in order to explore this "true view" of perception. As you 
indicate, however, perceptual research has shown that the subjective 
experience of colour has little to do with wavelength and everything to do 
with the balance between wavelengths. Land (of Polaroid fame) showed that you 
could start with a collage of primary colours (which he termed a 'Mondrian' 
after the Dutch abstract impressionist) which one viewed under white light. 
One noted red, green, yellow. The lights could then be changed such that there
was not a photon of "red" in the illumination and the viewer would continue to
report the scene as before, unchanged. Red stays red in relation to its 
neighbours. We calibrate our percept in *relation* to other things.

Herein lies the message. Hard truths exist as relationships: as balances
struck between things, as polarities which we identify, as dimensions that we
derive from an evolving explanation of our experience.

No doubt someone has already done the following. Our percepts trigger 
detectors which have beeen created by learning processes (or which are hard-
wired into the brain; or some combination thereof) such that certain 
primitives (red, round, rose-scented) trigger neurological events. This we 
know to be true, both in the sense that we can measure this and also 
manipulate it. Each detector represents a dimension which is more or less 
expressed in percept space; and the strength of its detecting red-ness, 
circularity or odour is a quantity measured on that dimension. The 
neurological instant is, therfore, represented by a vector which spans the 
space created by these various (nested, resonating, hierarchical) detectors. 
At one point in this space lies the concept that is evoked when the vector 
rests upon it: a rose. On another, a red ball. The symbolic chunks into 
which we dissect our percepts can be thought of as points in such a vector 
space, where the orthogonal vectors are the things which we have learned to 
recognise. Fuzzy logic people will recognise similarities.

Prescription: htch up a sea of neural networks to a series of spaces within 
which fuzzy logic rules operate. Generate "symbols" out of the nonlinearities
which emerge and use symbol heuristics to link these together. Use error 
minimisation as the route to feed back by which the nets learn that ! this !
bundle of percepts matter (autocorrelate with  ! those !) and you would be 
(well, might be) on the way to a genuine learning system that discovers
for itself how to represent knowledge in useful ways. 

I suspect that one would also need objective functions (analogous to pleasure, 
pain, fulfilment and frustration) in order to drive such a systsem: error
minimisation might be insufficient. Add delta from gaol as a proxy and supply 
the initial goals from outside, therefore.




_________________________________________________

  Oliver Sparrow
  ohgs@chatham.demon.co.uk
