COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Topics for a talk
C00009 ENDMK
C⊗;
Topics for a talk
Retinex,Horn,Marr
AI,philo,models,direct perception
Computer magazine article
Posner
examples of prowess
general AI, problem of organism



Language is apparently based on models for perceptual processes, rather than the
converse (Wharfian hypothesis).


The organism has access only to its models of the environment, not to
the objects out there.  Language is thus referential only in that it
refers to the models, not to the objects directly.  It is clear though
that we really think as though there were objects out there.  When we
refer to an electron, we are really referring to something which allows
us to pull together a lot of experiments in our atomic theory.  But that
is equally true of an orange; when we say we see an orange, we mean that
certain photons enter our eyes.  We interpret this as certain reflectance
(orange color).  There might be some demon switching objects in front of
us faster than we can see.  In fact, we make great leaps of trust in
assuming that the object is the same when we watch it, or turn our backs
and then see it again.  Or again, it might only be some great dream
going on in our solitary heads.  It turns out to be useful for survival
to make these models and assume they correspond to something in the world
out there.  In a real sense, our assumptions about an orange are very
like those about electrons, only much more vivid and immediate because
we carry our experimental apparatus with us, and have other experiments
like manipulation for consistency.

Basically models allow us to do thought experiments instead of just
blundering randomly along.  For example, if I am chasing a rabbit,
I predict where he is going next.  The rabbit has evolved to run
erratically, but sometimes maintain their course long enough that our
predictions are accurate.  It is clear that our energy towards understanding
and building models has great survival value.

To understand intelligence is to understand the models we make of the
world around us.  
The models serve to allow us to do thought experiments, rehearsing
as Freud and Perls say.  Some models are tightly organized, such as
physics, even the physics which we all know without schooling.  There
are quite a number of models, topology,logic,geometry, algebra, number theory and 
various forms of mathematics, chemistry, physics, economics, social
structures, psychology, etc, etc.
Some models are rather vague, like the scripts we acquire on how to act
to get certain payoffs.
The point is that to use and understand language, to exist in the world,
to be intelligent is to have these models (representations).  

We are in the field of understanding human intelligence and of understanding
how machines can be intelligent.  That means we want to understand human models
and calculations and how machines can accomplish the same thing.  Often this
means that they must have equivalent models, although we are not bound to do
things the same way.  The similarity is really that of the similarity of the
models of different people.  The world out there is the same world, even if
two people are different, and models of fairly obvious real world things
have very similar outward behavior for different people.  The most efficient
form of calculation may be very different for a machine and a living being.
For example, human flight was achieved by means very different from the
flight of birds. 

retina, retinal layers, lgn,cortical layers, gestalt organization
camera, retinex, color, edge detectors, organization, description

revolutionary nature of semiconductor technology

understandings of line drawings

representation of shape

stereo and depth measurement techniques

technical applications
	assembly
	computerization of industry--rationalization