[From Bill Powers (960410.0700 MST)]
Chris Cherpas (960408.1534 PT) --
RE: PCT: selectionist?
Since I've done EAB work, and have now been exposed to PCT, I
intuitively read something like, "behavior Y is reinforced by some
property of X" as something like "behavior Y probably controls some
property of X."
I'm glad you brought these two ideas together in just that way. I don't
think anyone has yet stated the real difference between PCT and
reinforcement theory: that PCT explains HOW any given behavior could
possibly occur, while reinforcement theory tries to explain only WHY it
occurs. Reinforcement theory is about learning or adaptation, not about
performance. There is nothing in reinforcement theory that explains how
a rat can lift a paw, place it on a bar, and depress the bar. If you
ask, "why does the rat press the bar?" the reinforcement theorist can
say only "because it has been reinforced for doing so." The control
theorist, on the other hand, is more likely to answer "It is varying
reference levels for its perceived relationship between paw position and
bar position, in order to make its perception of food intake match an
internal reference level for food intake."
These are completely different kinds of answers. If you ask a control
theorist _why_ a given behavior occurs, the answer is likely to be in
terms of _how_ am already-learned higher-level control process works.
The answer is NOT likely to be given in terms of reorganization. To find
a real basis for comparison, we would have to find a control theorist
who says that the rat presses the bar "because its former control
organization wasn't sufficient to maintain its state of nutrition, and
it had to reorganize to discover some other controllable means of
maintaining nutrition." In other words, the control theorist would have
to be thinking in terms of why there was a change in the organization of
behavior, a subject specifically related to learning or adaptation.
Reinforcement theory says that the organization of behavior is changed
by reinforcers which act on the probability that a response will occur
in the presence of a discriminative stimulus. Reorganization theory says
that the organization of behavior is changed by the organism until it
results in a consequence that reduces the intrinsic error that is
causing the organization to change. Under reinforcement theory, the
cause of changes in behavior is an external event or thing called a
reinforcer. Under reorganization theory, the cause of changes in
behavior is an error inside the organism, plus a mechanism that converts
the error into changes in brain functions. When we consider only what
causes the organization of behavior to change, reinforcement theory and
reorganization theory are obviously different theories. Reinforcement
theory puts the efficient causes of organizational change outside the
organism; reorganization theory puts them inside the organism.
The other basis of comparison is the _performance_ theory -- the theory
that explains behavior once it has reached a steady state. Here, the
theory offered in EAB is embedded in reinforcement theory, and does not
actually explain how behavior works. To say that a response is emitted
in the presence of a discriminative stimulus is specifically to avoid
any claim about a causal relationship between the discriminative
stimulus and the response. The response is simply "emitted," and there
just happens to be a discriminative stimulus present at the same time,
or perhaps just a little bit previously. EABers, in attempting to
divorce themselves from S-R theory or reflexology, deny that the
discriminative stimulus acts to cause the response; it merely signals
the conditions under which a given response is likely to produce
reinforcement.
So in sum, the theory of operant conditioning contains no theory of
performance at all for comparison with the performance theory of PCT. It
offers no explanation of HOW an organism can produce just the response
that will produce the reinforcer; it can only say THAT IT DOES SO.
This makes explanations based on reinforcement theory seem somewhat
trivial to anyone interested in modeling the actual physical processes
involved in behavior. Asked to explain the upright stance of bipeds, the
best the EABer can do is to say that adopting this stance must have had
consequences that were more reinforcing than the consequences of
adopting other stances. Nothing at all can be said about how it is
possible for a biped to balance itself on a series of stacked bones held
together by elastic bands. The EABer can offer a theory for WHY the
biped stands erect, but can offer no explanation for HOW it can possibly
do this.
In PCT, both the why and how questions are answered by a single
hierarchical model in which all processes, both learning and
performance, are carried out by functions inside the organism. The
environment enters this process only in terms of physics and chemistry;
it contains no special influences on organisms that are not present
without organisms. There are no "stimuli" which have special
discriminative properties; there are no "reinforcers" that have
specialized capacities to alter behavioral organization. All such
special effects are explained in terms of how the organism views various
physical events and processes, and what the organism does to affect its
experience of the world. Behavior and the change in organization of
behavior are explained in PCT entirely in terms of functions inside the
organism. The environment provides only the setting in which these
functions are carried out.
Now, as to the question "Is PCT a selectionist theory?" If you're
speaking of performance, the answer is NO. The subject of performance is
covered by control process which act systematically to control
perceptions, not by using fixed behaviors or responses, but by using
control systems having fixed parameters. If you're speaking of learning
or adaptation, the answer is YES. There are mechanisms inside higher
organisms that can vary the parameters of existing control systems and
create new control systems; these mechanisms, in their simplest form,
operate by blind variation and selective retention.
Of course both types of mechanism are always working, so we can see both
of them operating at once. When the environment is relatively constant,
the learning/adaptation process, driven by intrinsic error, alters the
parameters of control systems and brings new ones into being, until the
control processes can maintain the environment in a state that reduces
intrinsic error essentially to zero. When that occurs, we find that the
control processes are highly regular and systematic and the variation-
and-selection process is nearly inactive. In a hierarchical system as
complex and competent as that in a human being, the systematic control
processes can be so efficient that we rarely see any reorganization. In
less capable animals, like rats, even modestly complex environments
strain the ability to keep intrinsic errors near zero, and the
reorganizing component of behavior becomes larger, or even predominates.
At the higher levels of the human system, we probably have organizations
similar to Hans Blom's model-based control systems, in which the blind-
variation-selective-retention process results in an organization that
can use _systematic_ variation and _systematic_ retention. This adds an
order of magnitude of control skill, reducing even further the chance
that intrinsic errors will become large enough to start the non-
systematic process up again. Of course this is still all conjecture; it
will be a long time before we understand the lower-level systems well
enough to be able to isolate phenomena due to such higher-level
processes.
So -- ask a short question, get a long answer.
···
-----------------------------------------------------------------------
Best,
Bill P.