[From Rick Marken (930908.0800)]
Gary Cziko (930908.0153 UTC)--
I think both Bill Powers and Rick Marken are
missing the essential point (at least as I see them) of what encodingism is
and why it is fatally flawed as a system of representation/perception.
Hopefully I can make this clearer this weekend.
It will be EONS until then.
I think I understand the flaws in what you describe as "encodingism" --
and I agree that they are flaws; they are 1) assuming that some of one's
own perceptions have a special ontological status -- as events which
are coded and 2) assuming that these events are represented by neural
events that are in some way "like" the coded events.
I think the PCT version of "encodingism" avoids these problems.
Perceptions are simply signals in the brain (from the model's
perspective; from your perspective they are aspects of "the world").
Your experience (perception) is what things look like when you ARE a
neural signal. In theory, these neural signals vary as a function of
variations in aspects of the world outside the brain. The most
successful representation of what that world is like (success
measured by quantitatively precise experimental verification of
model predictions) is embodied in the models of chemistry,physics
and neurophysiology. So we include those models (where necessary)
in our model of the brain (one of the basic "philosophical" tenets
of PCT is to be consistent with other successful models of "boss reality").
Last night (in order to avoid writing but also with an eye to developing
a paper on the PCT view of perception) I was playing with a simple little
model (and experiment) that I wrote some time ago that controls either the
area or perimeter of a rectangle. In order for the control system to work I
had to have it perceive the controlled variable -- in one case X*Y and
in the other 2X+2Y (X is width of the rectangle; Y is height). Both control
systems have a perceptual function that has two variables as inputs
(X and Y); the "area" function takes the product of these inputs, the
perimeter function takes the appropriately weighted sum, and produces p.
(By the way, the control system has only one output degree of freedom, o,
which affects both X and Y; X and Y are also influenced by independent
disturbances, dx and dy, so X = o + dx and Y = o + dy). The system
work great; if p = X*Y it controls area; if I change the perceptual
function so that p = 2X+2Y it controls perimeter (I also have to change
some dynamics to stabilize the system -- which is interesting; when I do
the experiment and switch mentally from controlling area to controlling
perimenter I experience a change in the dynamics that I must use
to successfully control the new perception).
The point of all this is just to show that, in order to model this control
process, I have to make some guesses about what variables are being
controlled.
I assumed that p =f(X,Y), where X and Y are other perceptions of mine (they
are perceived symbols in a program); for various, poorly thought out
reasons I have considerable confidence that X and Y are measures of the length
of strings of pixels oriented at right angles to each other. My brain aches
when I try to probe why I accept all of my assumptions about X and Y being
reasonable inputs to the perceptual funcitons; but I do know that when I use
these variables in a program, the program produces output that is very much
like mine when I try to control what I would describe as the "area" or
"perimeter" of the rectangle on the screen.
Note that the model's perceptual function (a simple multiplier) that produces
what I call a perception of "area" is almost certainly NOT the same as the
perceptual function in my brain that produces the same perception. The model's
perceptual function knows nothing about shape or enclosed space or whatever;
if X and Y were two vertical lines next to each other the model would still
control the product of their lengths -- but it would not look like control
of area any more.
Modelling the way the brain computes perceptual variables,p, from the variable
aspects of "boss reality" that are represented at the sensory inputs to the
nervous system, s -- that is, discovering how the brain implements the f()'s
which convert s's into p's -- is one of the most important (and most
difficult) tasks of PCT modelling (it is important because you need to under-
stand f() in order to understand what a control system is controlling, ie. to
understand what it is "doing"). If that's encodingism than I really want
to know what's wrong with it -- before I waste too much more time on it.
Best
Rick