Re: behavior as mapping


JGCASEY wrote:


Such mapping of a whole input to a set of outputs
amounts to crude template matching and is not how
real animals or recognition machines work.



Motor mapping works for very simple cases, but the more complex the
behavior sets, the more poorly it fullfills the requirements for
intelligent adaptive behavior. Eg, in the frog, the visuo-motor maps in
the tectum are fairly straight-forward [albeit, even there they are
modified by diencephalic inputs], but by time you get to higher animals
where the eyes are movable in a movable head, direct mapping is
subsumed by more complex and flexible mechanisms. Among other places,
this is discussed in The Computational Brain, by Sejnowski and
Churchland, 1992.

.