[From Bill Powers (931219.0930 MST)]
On modeling behavior using control theory
In discussions of control theory with engineering control
theorists on the net, it has become clear that engineers have
developed certain ways of viewing the problem of modeling
behavior, and have settled on particular ways of representing it.
To a great extent, the conception of what has to be explained is
dictated by the form in which control diagrams are customarily
drawn, and by the mathematical methods already available for
analyzing systems with a certain organization. The question of
how to apply control theory to behavior thus becomes the question
not of how behavior is organized, but how the existing methods of
analysis can be used without change, together with their
underlying assumptions about what behavior is. The engineers have
been arguing, in effect, that the methods they learned in school
are completely adequate for the analysis of behavior, and that
PCT introduces nothing new. As long as that belief persists we
will get nowhere. I think we should focus on the real issues, or
give up.
···
------------------------------------------
The most common engineering assumption is that behavior is the
externally visible effect that an organism has on the physical
world around it. The engineering assumption is that there is a
"plant" in the world outside the organism, and the objective of
organismic control is to bring this plant to a certain static or
dynamic state and maintain it there in the presence of
disturbances.
The PCT view, in contrast, is that organisms know nothing
directly of the properties of the environment. All information
they have about the environment is found in the primary sensory
signals resulting from the impingement of physical stimuli
directly on sensory nerve endings, and in perceptual signals that
represent further processing of the primary signals. As a result,
all that the organism can control is a representation of the
world in the form of neural (or chemical) signals. There is no
way for the organism to determine the actual effects of its
outputs on the external world, so it has no way of altering its
own actions or organization to create systematic effects that are
not represented as perceptions.
This immediately rules out "open loop control" as a possible
model of behavior. If the organism can't ascertain the effects
its output are having, there is no way for it to adjust its
outputs to produce any particular effect on the world. All
outputs have particular effects on the world, but the organism
can control only those effects that appear in its sensory world.
All control is closed-loop control, in a model of organisms. If
sensory information is lost, some information dependent on the
output signals must be substituted if any form of control is to
continue. And when such a substitution takes place, the
correspondence between the perceptual signals being controlled
and variables in the external world is lost. How much difference
that makes depends on the rapidity with which the substituted
information departs from the information that would have been
received if the sensors were operating normally.
--------------------------
In the engineering approaches described on the net, there seems
to be a tendency to think of control behavior as one single
process, often of great complexity. The PCT approach is to
consider behavior as the aggregate of a great many simple control
processes, working in parallel and also hierarchically organized.
The engineering approach is a matter of preference and custom,
undisciplined by the actual organization of the neuromuscular
system. The PCT approach is forced on us by the facts we know
about the neuromotor systems.
We know of a great many simple control systems in the human
organization, which deal with simple scalar variables, their
integrals, and their derivatives. These control systems are used
in all kinds of behaviors, in controlling all sorts of variables.
The 600 to 800 control systems that control the musculature
directly are employed by higher systems in every behavior that
involves overt action. Their organization remains constant while
the behaviors in which they are involved vary over the whole
range of human experience, from scratching an itch to multiplying
two polynomials together using pencil and paper.
Furthermore, behavior is clearly organized in larger units which
also retain their character over a wide variety of behaviors. We
learn specific skills like walking and speaking and typing,
riding bicycles and driving cars, handwriting letters and
numbers, opening doors and windows, putting things down, picking
things up, moving things from one place to another, throwing
things, pulling things apart, putting them together, and so on
and so on into the hundreds or even thousands of elementary
control processes. These elementary skills remain the same even
though they are used as the means of controlling more abstract
variables of all kinds, in all sensory modalities, under all
kinds of external conditions.
Representing all these control processes as a single complex
expression is simply not possible, and it is probably not
possible that they are physically realized as a single complex
process. It isn't useful to try to find a general expression that
will cover all these varieties of behavior and all these levels
of organization. We need to understand the details, because each
control system is a general-purpose device that can be called
upon in a variety of combinations by higher systems of very
different nature. A lower control system provides the means for a
higher system to bring one variable of experience to a specified
value in a reliable way, without the higher system needing to
know how to accomplish that end. The higher system simply
generates a signal that is an example of the desired perception,
and the lower system takes care of all the necessary details in
bringing that perception to the requested state.
The single-system single-level approach is simply too limited to
explain all of human behavior, even in principle. It requires
that we either understand the entire human system, or understand
nothing. Under the hierarchical approach, we can analyze behavior
systematically, starting with the least units of organization and
progressing to the more complex ones. This is the only practical
way to model the entirety of human behavior, and it is also
probably the way behavior is actually organized.
-------------------------------
To model behavior we must first observe it carefully. I do not
see the required degree of care being taken in the engineering
approaches. Too much is taken for granted, too much is left out.
Common-sense assumptions are used where a model is really needed.
Diagrams leave out essential functions and connections.
Disturbances are either ignored, or assumed to have convenient
characteristics that allow them to be handled in a statistical
way instead of in detail. Too many loopholes are left, too much
is assumed without demonstration. The approach to PCT is
defensive, not open. There is too much blind reliance on
mathematics, and not enough attention paid to identifying the
constants, variables, and functions with observables. The
orientation is theoretical, not experimental.
These are all defects that we have labored, and still labor, to
remove from the PCT approach. Until control engineers take them
seriously, it will be hard for them to see why their approach is
inadequate to the problem.
---------------------------------------------------------------
Best to all,
Bill P.