[From Rick Marken (920825.0930)]
PCT has been around for quite some time. The basic idea of PCT
is rather easy to state:
Organisms are closed-loop negative feedback systems that control various
aspects of their own perceptual experience. Observable actions are the
visible means by which organisms influence their own perceptions and
compensate for those influences of the environment (disturbances)
that would tend to move perceptions from reference levels specified
by the organism itself. Thinking, planning, memory and imagination
are pocesses that involve controlling perceptions without going through the
external environment. Learning is the processes of reorganizing the existing
means of controlling perceptions.
Basic PCT is still best summarized by the title of Powers' classic:
Behavior: The control of perception.
We have been hearing a lot in the last year about many of the "hot" new
approaches to understanding the behavior of living systems. Some are
schools of thought, others are the approaches of individuals or labs.
Examples are: Beer's bug, Artificial life, interactionism, Agre/Chapman
Pengi system, Brooks' subsumption architecture (and general approach to
robotics), fuzzy logic, chaos, dynamical systems, Artificial Intelligence,
etc, etc.
What ALL these approaches (and many, many others) have in common is
that NONE of them explicitly recognizes the basic organizing principle
underlying the behavior of living systems -- that they CONTROL
PERCEPTUAL INPUT VARIABLES. Thus, the essential insight of PCT
is missing from EVERY ONE of these hot approaches. In fact, many times
it seems like these investigators are going OUT OF THEIR WAY to avoid
concluding that organisms control perceptions. This hit me while reading
the Agre/Chapman paper. A&C said at lot of things that were consistent
with PCT. For example, they correctly pointed out the problems for
programmed output type models -- 1) unpredictable changes in the
environment, 2) complexity of computing all required outputs [we'd call
that the "inverse kinematics" problem] and 3) the unpredictable effects of
planned activities (I'm being generous -- it sounds to me like A&Cs
problems 1 and 3 are the same). So A&C state (albeit quite unclearly) the
basic reason why programmed output systems will not produce life-like
behavior -- [the reason. of course, is disturbances, in PCT terms]. They even
say that behavior cannot be generated by a plan that "determines an agent's
actions" -- which is EXACTLY what PCT says. But after all that they end
up proposing a model that is described as one that generates actions -- when
it is, in reality, one that generates perceptions. They just won't even SAY
IT; the model acts to produce intended PERCEPTIONS.
A&C claim that their model doesn't control actions -- but they won't say
what it controled What they do say is that the model has a mental plan that
"serves as a resource that an agent can use in deciding what to do". How
does the system use this resource? It turns out that it generates predicted
outcomes (intended perceptions -- but they don't say it), it then generates
actions and compares the actual outcome (as PERCEIVED) to the predicted
outcome. The plan is a resource in the same sense that a reference signal
is a resource -- it specifies an intended state of its own inputs.
So the A&C model (after all the verbal nonsense) is a control model -- and it
acts to produce intended perceptual outcomes. But C&A don't realize this.
The same was largely true of the Beer model and, as I recall, Beer seemed
quite ademantly opposed to the suggestion that his bug was controlling
perceptions.
Does it matter that these hot modellers don't understand that their models
are controlling perceptions (when they are)?? Or is it just a verbal quirk --
PCTers like to say that behavior is the control of perception and hot research-
ers like to say things like "behavior is the result of a communication process
between plans and outcomes". I think the failure to even SAY that behavior
is controlled perception IS a FUNDEMENTALLY IMPORTANT error --
not just a linguistic quirk. By failing to point to this important aspect of
behavior (or by refusing to notice it it) those working on these hot
approaches to behavior 1) fail to see all the great work that already has
been done on modelling the control of perception 2) don't understand that the
behavior of the system depends largely on how they design the perceptual
functions 3) get lost with attempts to program output once the lower level
perceptual control systems have been built and 4) don't understand how
much of their own understanding of the situation is going into their
modelling.
The moral: I just don't think there is much to be gained by trying to sift
through the chaf of the hot approaches to understanding behavior in the
hope of finding much wheat. Why do it anyway, when you have all the
flour you need sitting right there in front of you with PCT?
The question: Why is there such reluctance on the part of those working on
the hot approaches to behavior to even consider the possibility that behavior
is the control of perception? Why do people in all these areas seem to
ACTIVELY avoid even MENTIONING this as a possibilty? Has anyone
ever seen the phrase "control of perception" used anywhere but in a paper by a
PCTer -- whether it has been used correctly or not? What's the problem
here?
Regards
Rick
ยทยทยท
**************************************************************
Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
E-mail: marken@aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)