From: Mark J Young <mjy@maxwell.arc.nasa.gov>
Subject: Re: TECH: Are sensory inputs a help or hinderance
Date: Mon, 22 Feb 93 11:06:41 PST


I disagree with those inclined to think that a direct interface
between a sensorium simulator and neural pathways would
be a simpler VR technology. At our current level of knowledge
about biological perception, and well into the foreseeable future,
I don't think that a direct neural interface would simplify simulation
design.

It is very difficult to be sure what information at the sensory
receptors is involved in high level percepts. As an example, notice
that the current understanding of human 3D vision is characterized by
a good deal of disagreement about what the internal representation
of 3D objects and space actually is. This leaves undetermined what the
the form of the computations of 3D structure from 2D image measures are,
and, thus, leaves undetermined what the specific 2D measures are. You then
realize that there is a long way to go until you can determine what is
the necessary and sufficient set of pixels (as functions of time) that
will result in the targeted 3D percept. It is easier to simulate more
than you need (model the world instead of the specific cues) because
the analysis of what you don't need is intractable, as of yet.

Even at a lower level, understanding of what 2D (projection of 3D)
spatial information (as function of time) is input to the sensory
cortices is incomplete. It has taken about three decades just to
characterize the linear aspects of sensory information processing
and that is just the beginning of the story. A precise specification
of the sensory systems will require much than more a linear specification
(although that work has also been going on). So again, if you model
sensory input for specified world phenomena, except for extremely
simple stimuli (e.g. sine-wave gratings), it is difficult to predict
what the transformation between sensorium and cortical input is.

Mark J. Young.