[From Rick Marken (990716.2110)]
A model (like PCT) is not just a mathematical tool for cranking out
predictions; it is also a way to think about what we observe. PCT
(especially for the non-research oriented) provides a mental model
of the the processes that we _don't see_ (brain activity) that are
presumably responsible for what we _do_ see (the behavior of living
systems). The basic PCT model exists as the familiar control system
diagram (see, for example, _Mind Readings_, p. 73). But there are
many different ways to think about _how_ this model maps into what
we actually see people doing.
One suggestion regarding how to map the PCT model into the behavior
we see is to think of behavior as the result of error in the brain.
So when I see a person pushing a cart down the hall (as I did earlier)
I _imagine_ that there is an error signal in the fellow's brain --
a discrepancy between a reference and perceptual signal, per the
PCT control system diagram -- and that the behavior I see is an
effort to reduce this error.
But there are problems with this way of mapping PCT into observed
behavior. First, as Bill noted, in a high gain control system, error
is always very small, virtually zero. Second, as I noted, a control
system can behave (push a cart down a hall, for example) even when
error is _precisely_ zero. There are other problems, too, but these
two are enough to suggest that another approach to mapping the PCT
model t observed behavior may be desirable.
I would like to suggest another way to map the PCT model into
behavior. I suggest looking at behavior as _your view_ of
perceptions that the behaving system keeps matched to fixed
or (more often) time varying reference signals in the brain.
The reference signal can be viewed as a causal or independent
variable; the perception can be viewed as a dependent variable.
Think of the reference signal changing smoothly and continuously
(when it changes). This way, you can imagine changes in the
reference signal (such as changes in the reference for the
location of the cart) being reflected _instantly_ in changes in
the corresponding controlled perception (the perception of cart
changes right along with changes in the reference signal). The
dynamic (continuously changing) behavior we see reflects the
dynamic (and smoothly varying) behavior of reference signals in
the the brain.
So I suggest that the (continuously changing) behaviors we
see are our view of perceptions that are continuously changing
in response to continuously changing reference signals in the
brain of the behaving system. The behaving system is responsible for
its own behavior because the behaving system is the one setting
the varying references that cause the varying perceptions that
we see "from the outside".
Some of the behaving system's perceptions are being varied as the
means of protecting higher level perceptions from disturbance.
This is what happens when the kinesthetic perception of mouse
movement is varied as the means of protecting cursor position
from disturbance. But it is still OK to think of mouse movements
as perceptions that are varying in response to variations in
reference signals because this is what is happening. Mouse
movements are themselves a controlled perception (controlled
relative to a smoothly varying reference) which will be protected
(by variations in muscle force) from disturbances. The distinction
between act and result (at all levels in the hierarchy but the
very lowest) is simply a way of talking about the role of different
controlled perceptions in the hierarchy; the references for lower
level controlled perceptions are varied _as the means_ of protecting
higher level controlled perceptions from disturbance. But both acts
and results are typically controlled perceptionz. Only at the lowest
level of the control hierarchy (where output variables have direct
effects on the environment) is an act (output) an uncontrolled
variable.
So, in summary, here's my way of mapping the PCT model to the
behavior I see. When I see behavior, I imagine that in the
person's head are a bunch of varying reference signal "arrows"
pointing _out_ toward the sensory surface (the eyes, ears, nose
ligaments, skin, etc). Corresponding to each of these reference
signal arrows is a perceptual signal arrow pointing _into_ the
body (toward the reference signal arrow); these perceptual signal
arrows come from the sensory receptors. A reference signal arrow
can "move" (change value) and when it moves, its corresponding
perceptual signal arrow moves right along with it. What we are
seeing as "behavior" is our view of the "back" of the perceptual
signal arrows. Of course, we don't know which of the arrow "backs"
we see correspond to arrows (perceptions) that are actually under
control; that's why we have to test for controlled variables.
That's a pretty good cartoon of how I look at behavior as the
control of perception. I usually take it for granted that the
perceptual "arrows" are forced to "track" changes in their
corresponding reference "arrows" by the operation of the control
loop (which keeps the perceptual arrow exactly on the reference
arrow, protected from disturbance). But I don't waste a lot of
imagery on the operation of the control loop itself; once you
know how the loop works you can pretty much take it for granted.
What I think is important about behavior (looked at from the
perspective of the PCT model) is that perceptual "arrows" are
kept precisely matching reference "arrows". Reference arrows
(and their corresponding perceptual arrows) are constantly
changing as the means of keeping higher level perceptual arrows
matching higher level reference arrows. The fellow who was pushing
the cart (making the cart position perceptual "arrow" match a
smoothly varying cart position reference "arrow") was presumably
doing so in order to keep some higher level perceptual "arrows"
(such as the perception of "delivering") matching its reference
"arrow".
I'll have to talk to Disney about turning this into an animated
"true life adventure" short.
Best
Rick
···
---
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/