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!europa.eng.gtefsd.com!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: 82
Date: Tue, 20 Sep 1994 11:55:07 +0000
Message-ID: <780062107snz@chatham.demon.co.uk>
Sender: usenet@demon.co.uk

I posted a response to this once before, but it vanished into intenet limbo. If 
it suddenly shows up, there will be two versions on display and I shall appear 
a prat. Yet I am large, I contain multitudes.

--------------------------------------------------------------------------

Goethe's study was not entirely obscure: it stimulated the school of 
pointillism most associated with Seurat. The point in respect of relativism is 
now known to be true, however: Land (of Polaroid fame) demonstrated this some 
decades ago. If, for example, an observer looks at a isolated panel made up of 
strips of colour (figures which Land called "Mondrians", after the Dutch 
abstract impressionist) and sees these illuminated with white light, all 
such observers will tend to point to the red bits and say "red" and the green 
ones and say "green", unless RG colour blind, when of course they do not. This 
aside, if the light balance is now gradually changed, the observer of this 
isolated scene sees no change: red remains red, when there is not a photon of 
light bouncing into his or her retina which would normally be associated with 
redness. Our perceptions are relative, based on the dynamic set up between what 
we perceive rather than primarily driven by the absolute values of those 
signals.

We know quite a lot about colour (and other) forms of perception. It seems the 
case that qualities (redness) are linked to a percept in a specific part of 
the visual cortex. In this, a dimension is evoked and linked to other 
"primitives", such as roundness, circularity and the like. One can see each 
of these primitives as being more or less expressed, such that in a space 
spanned by N such orthogonal or related dimensions, there are a set of 
distinct zones within this N dimensional space:

    {so much red, so much roundness, smooth edges}       evokes "ball"
    {less red, same roundness, more fractal/ frilliness} evokes "rose"

In such a percept space, each generic ("ball", "rose") would have a more or 
less discrete space that would resonate - become excited, operational - when 
perceptions placed a vector within it. Other areas of perception (odour, 
weight, social context) would also be aroused both by their incoming data and 
by the parallel arousal of linked areas of percept space that had, 
experientially, been linked to them by past learning or which were hard coded 
by genes or design. If 90% of the areas of arousal which are involved in there 
being a ball about were buzzing, the remaining 10% would also be aroused, 
producing fill in, anticipation, generalised seamlessness: what we experience 
as our awareness. (This does not tell you what "I" is but that's another story)

This is, of course, a potentially as true a picture of what happens amongst
conceptual as well as of sensory linkages. The progress of a thought could be 
seen as the resonance of linked areas of arousal which evokes associations and  
cross-references in set of parallel structures. One notes that "red" emerges 
from a data stream but is also a quality with which these higher order 
abstractions operate. We also know that our literal sensitivity to red is 
enhanced when we look for red things, or think about the colour. Systems are, 
therefore, self-referential: what allows X to be seen involves previously 
learned things about X.

People who work with fuzzy logic will see something that they recognise in 
this. I suspect that one can say something about AI if it is to emulate these 
aspects of NI:

1: It will be installed (taught to) massively parallel machines.
2: The machine will be hybrid, at least in emulation. There will be neural net-
   like elements, which filter and resonate with others which do the same 
   thing; and more symbolic actvities which enhance linkages which show 
   themselves to be actually or potentially useful and which attentuates or 
   deletes duff connections.
3: There will be a program in the sense of a set of goals structures and 
   objective functions against which these elements operate. These may be 
   innate to the design, hard wired in; or they may be more traditional 
   programs; or they may consist of real-time guidance and teaching.
4: The the progress of events in the machine will consist of a network of 
   mutual arousal which flows back and forth across the structure of 
   connectivity which has been established by past events. Thins will not be 
   "on" or "off" so much as more or less contributory, more of less related, 
   more or less aroused. Individual systems will learn to apply thresholds, to 
   cope with nonlinearities; and specialised structures may exist to force 
   indeterminate systems into closure.

It may be that all of this is emulable in commodity silicon. Then again, it
may not.
 
_________________________________________________

  Oliver Sparrow
  ohgs@chatham.demon.co.uk
