Re: (OPE-L) Re: The Church-Turing thesis

From: Ian Wright (ian_paul_wright@HOTMAIL.COM)
Date: Thu Jan 22 2004 - 18:22:04 EST


Hello Chris,

Yes, chess programs don't play like humans, and tend to rely on brute-force
methods of heuristic search (though not exclusively). It's easier to beat a
human in this domain with extra processing power rather than trying to
simulate human reasoning. But it doesn't follow that we've learnt nothing
about human mental processes from AI and cognitive science. Apologies for
the length of this message, but allow me to give one example of something
that has been learnt.

Take your example of "gestalt". The Necker cube illusion:
http://www.yorku.ca/eye/necker.htm
causes the observer to flip between two different kinds of "gestalt"
interpretations of the scene. AI can explain this. Research in computer
vision has discovered that processing a visual scene is indeterminate as the
visual field does not contain sufficient information to uniquely interpret
it (Kant would be happy). Some AI vision systems parse the visual field by a
process known as constraint satisfaction, in which a set of constraints,
usually types of connections between identifiable lines at various points in
the field, are labelled as "in front" or "behind" each other in a mutually
consistent manner. In the case of a Necker cube, there is more than one
consistent interpretation, and the constraint satisfaction process does not
settle down, but flips between the two possible global labellings. This
algorithmic finding can then integrate with neuropsychology and related to
the organisation of the visual cortex to determine whether this is a good
theory of aspects of visual processing in humans and animals. This theory of
gestalt, unlike the contemplative or speculative version, has the advantage
of being sufficiently well specified to be constructed in artificial systems
that replicate aspects of human vision. So not only has something been
learnt about the human mind, but something practical has been built.

-Ian.

_________________________________________________________________
The new MSN 8: advanced junk mail protection and 2 months FREE*
http://join.msn.com/?page=features/junkmail


This archive was generated by hypermail 2.1.5 : Sat Jan 24 2004 - 00:00:02 EST