Early PCT Research

[From Bruce Abbott (950603.1340 EST)]

Bill Powers (950602.1505 MDT) --

Your expanded description of the series of experiments answers all the
objections I raised. Considering the thorough way in which you tested
for variables that the rat acted to control, the main question left in
my mind is "what kept you from tumbling to PCT?" Was it simply the
custom in that field of interpreting variables controlled by the rat's
actions as variables that controlled the rat's actions? If that were all
there were to it, I should think that the transition to PCT would be
relatively easy for behaviorists. As you look back on your own frame of
mind during those experiments, can you find any illumination on this
subject?

This is a good question and a difficult one. I think the best answer I can
give is that work such as this is done in a context: a particular set of
questions being asked from a particular point of view. Within that context
what you are doing seems to make sense; it fits in with previous research
and builds on it by providing new pieces to the puzzle that is being
assembled. To use Kuhn's descriptive, you are doing "normal science," the
ordinary pick-and-shovel work within an accepted paradigm.

In this case the context was traditional learning theory dating back to the
work of Thorndike, Pavlov, Hull, and Skinner and supported by nearly a
century of conditioning studies. The work we were doing in Pete Badia's lab
in the mid-1970s was an extension of a line of investigation begun by Pete's
mentor Charlie Perkins, who had predicted on theoretical grounds that
subjects would prefer a shock schedule in which shocks were signaled over an
identical schedule in which shocks were not. We were evaluating Perkin's
hypothesis (preparation) against an alternative we favored (safety signal)
and a few others, while trying to generate clear, systematic data
(functional relationships) whose value would stand (we hoped) independent of
any particular theory.

The fact was that rats given a choice between the two otherwise identical
schedules chose the signaled one (as Charlie had predicted) and that with
the schedule parameters we were using this preference was powerful. In the
context of traditional reinforcement theory, this meant that there was
something about the signaled schedule, _relative to the equivalent
unsignaled schedule_, that was acting as a powerful reinforcer for lever
pressing in our operant conditioning paradigm. From the theoretical point
of view this was extremely interesting because computation of the so-called
reinforcing values of various elements of the two schedules suggested that,
under traditional assumptions, the rats should be indifferent between the
two options; thus the observed preference seemed an anomaly. In this
context we viewed our studies as a search for the source of (differential)
reinforcement. What was motivating lever pressing was the greater overall
reinforcing value (equivalently the lesser punishing value) of the signaled
schedule relative to the unsignaled one. What was the source of this
difference?

It wasn't until I was nearly finished with my graduate training that I
encountered B:CP; by that time I had become very unhappy with the
traditional view and was already beginning to explore the notion that a
control-systems approach might provide the alternative I was looking for (I
was reading Ashby, Simon, and others). Unfortunately I was also in the
middle of preparing for qualifying exams and running three different
studies, including my dissertation; I found B:CP very appealing but did not
really have an opportunity to sit back and think about the broader
implications. It seemed to me at the time that the two views (traditional
reinforcement theory and PCT) were dealing with the same observables, but I
needed to step back and analyze how the concepts of PCT and of reinforcement
theory mapped onto one another. That is, I wanted to understand how control
theory would explain those findings for which the traditional view offered
what others were accepting as a satisfactory account.

Perhaps another difficulty that got in my way was that I did not really
understand the distinction between functional models (models of the behavior
of a system) and mechanistic models (models of the system itself). Without
this appreciation one can be impressed with models (such as the matching
law) that seem to explain observed functional relationships but which turn
out to be mostly exercises in curve fitting. There were plenty of such
models to admire in the late 70s.

There is a sense in which viewing a set of results such as those obtained in
the signaled versus unsignaled shock studies is like looking at one of those
"reversible figures" such as the Necker Cube. Such figures can be
interpreted in alternative ways that are each self-consistent but mutually
contradictory. You can see the Necker cube as if viewing it from above and
to the left or as if viewing it from below and to the right. Similarly, you
can view our experiments as methods to identify the source of reinforcement
for the observed lever-pressing behavior or as as methods to identify the
perceptual variable being controlled by the rat via its lever-pressing
actions. If you've aready been trained to recognize the
cube-seen-from-above in your data, it may require considerable effort and
coaching before you can see things the other way. Even if you can see both
cubes, both may appear to be perfectly reasonable representations of the
sense-data, at least until you have made an effort to evaluate the deeper
implications of each view. In the late 1970s I was beginning to see that
other cube, but had not yet done the work required to appreciate its full
implications for behavior analysis.

At this point, with what I think is a much better grasp of PCT and its
implications, I can flip back and forth between PCT and traditional
reinforcement-based descriptions and recognize where they are actually
dealing with the same phenomena and in some cases, using essentially the
same methods. In those cases it is indeed only a small step from the
traditional view to PCT, but for those used to seeing the cube in its
traditional orientation, getting the cube to reverse can be a difficult--and
disorienting--perceptual task.

I trust you misspoke when you included "being able to prevent it" among
the conditions to which the rats were indifferent. If a rat could press
a bar before the shock occurred and thus prevent it, it would learn
quickly to do so -- my rat paper was a study of just such a situation,
an experiment done by Verhave, a "Sidman avoidance schedule." But that
would not raise the issue of a preference for an abstract capacity to
control; it merely would verify that rats will control shocks if they
can. It wouldn't tell us what they think about being able to do so, if
anything.

No, I did not misspeak. In the avoidance experiment the rat performed on a
Sidman shock-avoidance schedule on one session and the actual temporal
pattern of shock delivery was recorded (rats on this schedule occasionally
make mistakes and receive shocks); this pattern was "played back" on the
next session, in which the rat had no control over shock delivery. As with
escapable versus inescapable shock schedules, the rats failed to resist the
disturbance when the apparatus switched them from avoidable to unavoidable
shock schedules or vice versa. The key here is that during training the
rats had learned that the shock frequency was the same whether they
controlled shock delivery let the apparatus determine when shocks would be
delivered.

Bill Leach 950602.18:33 U.S. Eastern Time Zone

[From Bruce Abbott (950602.1455 EST)]

Actually, if the rat had done nothing it would have been in the
unsignaled condition 0% of the time during the testing sessions.

Huh?

Oops! Sorry, I meant to say "in the _signaled_ condition 0% of the time."

As to your rather extensive comments in this post, I'm a little pressed for
time at the moment, so I'll just respond to the suggestion you raise that
doing PCT research on animals is expecially difficult because the researcher
has no way of knowing what variables the animal is actually perceiving and
controlling as opposed to those the experimenter perceives in the situation.
This problem has long been recognized by researchers of animal behavior, but
it is not an insurmountable one as one can always devise tests specifically
designed to determine what these variables are, as was done, for example, in
pigeon navigation to determine what variables the pigeon monitors and
controls when homing. On the other side of the coin, it may be easier to
map out the control systems of animals that lack the higher levels of the
control hierarchy found in humans. I never found a rat that chose, say,
higher over lower shock intensity because it thought I was trying to assess
its ability to "take" the higher shock level and wanted to prove to me that
it could.

As to the rat rolling over on its back to avoid receiving shock, the fact
that it did so was no reason to throw out the data so long as I could then
prevent that behavior and thus collect valid preference data on choice
between the two schedules. If you want to know whether Johnny would choose
apple over cherry pie, allowing him to choose cake spoils the test and fails
to give you the answer you seek, even though it might be of paramount
interest in another context. So while Johnny is making his choice, you try
to keep the cake out of sight.

Regards,

Bruce

<[Bill Leach 950603.23:43 U.S. Eastern Time Zone]

[From Bruce Abbott (950603.1340 EST)]

... _signaled_ condition

Thanks, that sure corrected an error!

I think that I am gaining an appreciation for at least some of what you
have been doing.

Some of the information that you seek is not itself relevent to PCT (ie:
does Johnny prefer apple pie over cherry pie when indeed he prefers cake
to either). The difficulty here is that to do the test you may need to
create a situation that Johnny would never permit if you did not
overwhelm his control systems. That is, given free control he might
never be willing to eat either apple or cherry pie.

This is the danger that I think that you always face in such experiments
(to some extent unavoidably). If you run your test and he chooses Apple
pie you could well have an item of "knowledge" that might actually be
useless (unless you are planning to incarcerate Johnny, offer only apple
or cherry pie and want to know which to order)!

Of course not all experimental result data will necessarily be useless
but just about anytime that you have to prevent the subject from doing
what the subject wants to do you are in danger of making a generalization
that will really only be valid when applied to the conditions that
existed in the experiment.

I DO NOT think that your rat experiment was of this sort however. While
this will be conjecture on my part, I think that the results of you
experiment support the hypothesis.

Sometime back Rick made the statement that skeptical curiosity had to be
taught (I think I have that right). I retorted that it did not but
rather WE do supress it in children. My position was that it was not so
important to teach children to use or how to use skeptical curiosity but
rather to teach adults to recognize that such a characteristic in
children should be valued and not surpressed by the adult.

My basis for this starts with the idea that newborn infant humans begin
live by activating every muscle that they have. Soon they are observed
moving their own limbs (erratically at first) within their field of
vision. Later they are observed moving just about anything that the can
grasp and move into their field of vision.

Their entire life seems composed of eating, sleeping and exploring. This
remains their primary activity set for several years. Some activities
will become regular as they discover things that they learn to enjoy, but
the exploring will continue to death (if they are fortunate).

The exploring "sees" its' first curtailment when some large control
system physically prevents exploration (overwhelms the smaller control
system). Further curtailment occurs when a consistant link is discovered
between certain exploratory activities and an error associated with the
reference value of pain caused by overwhelming control actions by one of
these large control systems.

These probably do not supress the exploratory activity much but the next
disturbance probably does major damage. At the verbal level of
exploration an "normal healthy child" can ask more questions about more
subjects that any other living being. Unpleasent experiences in this
exploration, including the often irrational explainations provided by the
large control system(s) likely begin to affect the systems concepts in a
negative way with respect to seeking understanding of what is observed.

I posit that this continual barrage of "social conventions" ("good little
boys and girls don't ask questions like that" or whatever) results in the
formation of systems concepts that supress the otherwise close to
intrinsic desire for understanding of observations.

I am sure that exploration of our environment must be closely related to
some intrinsic reference(s). Virtually all animals that I have had
experience in observing will explore extensively when introduced to a new
environment. The exploration activity for individual animals varies
greatly but this could well be due to "bad experiences" in previous
exploration attempts.

Most people (though certainly not all) will admit to some discomfort in
"unfamiliar" surroundings.

Certainly a "case" can be made for this characteristic existing because
of the consequences of "natural selection" but regardless of why it
exists I really do believe that at least all humans are "born with it"
and it takes a pretty dedicated, persistant and overwhelming force to
surpress it.

-bill