Applying PCT

CSGnet Archive CSG1998
41

[From Bill Powers (2000.09.24.0233 MDT)]

Rick Marken (2000.09.22.1600)--

Bruce Abbott (2000.09.22.1620 EST)

We end up disagreeing with each other because we have no firm
_data_ with which to test one formulation versus another.

To some extent this is true. But this is not quite the problem
I had in mind.

I remember sitting in my first undergraduate class in _Learning_,
at Ucla trying to learn _how_ to perceive behavior in S-R terms.

This is a nice lead into further explanation of my reasons for being
somewhat reluctant to teach people "how to see through PCT glasses." When
you were sitting there being taught, did anybody say to you that the S-R
framework was a _theory_? Or did they just teach you how to interpret
observations in terms of that model? When they called some things stimuli
and other things responses, did they explain what the underlying model was?
And did they try to _test_ that model in any way, to see if it was correct?

I know that nobody ever did that for me in my psychology courses, either
undergraduate or graduate (I did have a year of graduate school in
psychology at Northwestern). I was taught by example how to think about
behavior in terms of stimulus and response, but nobody ever worried about
showing that this was the _right_ interpretation, or righter than any
other. It was just the truth. Well, there were set arguments against
purposiveness that were repeated from time to time (the future can't affect
the present, and all that), but that was all.

It's possible to learn PCT the same way. You learn that behavior is the
control of perception, and you learn to sneer at S-R theory because
behaviorists are the Bad Guys, but this just creates the mirror image of
the other approach. Enough of that is going on already; I don't want to
contribute further to it.

To me, the essence of PCT is not PCT, it's the collection of reasons for
accepting it, including all the tests that have been done. When I set up an
experiment, like a tracking experiment or an experiment for controlling
some perception like size or orientation or pitch of a sound, I always set
it up as a test of the theory. Most of the time I have written up the
results before ever actually doing the experiment, leaving blanks where the
detailed numbers (like gain, delay, and prediction error) will go. So every
experiment starts with a detailed prediction of what I will observe. And so
far, because I've kept things simple, every prediction has been spot on.
I've had to make very few changes in the writeups. And this isn't because
I'm smart. It's because the theory is right in its main features, or right
enough to go on with.

I know of only three people connected with PCT who have formulated new
control experiments, devised a control-system model to explain them, and
done the experiments to test the model against real human behavior: Rick
Marken, Tom Bourbon, and I. Bruce Abbott has done some modeling but only as
thought-experiments; though he is perfectly capable of doing it, he has yet
to devise a new control-system experiment with a human being, write the
model as a simulation, and commit the model to a test against reality.
Bruce has a great thrill coming: to go through the anxiety of a real test,
and to see with astonishment just how accurately a control model can fit
behavior once you have it right. I think this sort of experience was
pivotal for Rick and Tom; I know it was for me.

The nearest I can come to giving non-modelers an inkling of these
experiences is contained in Demo1. Anybody who has yet to run Demo1 is
missing a huge chunk of understanding of PCT, and should remedy that lack
at once. Aside from the detailed demonstrations of different controlled
variables, Demo1 teaches some very important things about PCT that you
could miss if they weren't pointed out.

The most important one is that Demo1 invites the participant -- ANY
PARTICIPANT -- to engage in some control behaviors, and EVERY PARTICIPANT
who isn't brain-dead can in fact do them all with very little practice.
Furthermore, every step in the demo is accompanied by an automatic writeup
(shown when the results of each experimental run are complete) that
describes the results that will be observed, again for ANY PARTICIPANT, and
when the participant controls the variable in question, the plot of the
results conforms exactly to the description in the writeup. In many steps,
the participant is asked to select either one or two random disturbance
patterns, out of a choice of 10, before the experimental run starts. When
the demo program is first installed, these random disturbance tables are
generated from scratch; there is no way the program can know what they will
be. There are even blanks in the writeup which are filled in by
calculations from the data after the participant does an experimental run:
a high correlation between disturbance and action, and a low correlation
between the controlled variable and the action (described as such). Every
single run of every part of Demo1 contains a prediction of how a totally
unknown individual participant will perform, describing in advance the data
and even the results of some calculations that are done after the data are
obtained. So every single step of Demo1 is a test of control theory as a
model of behavior -- a test of a kind that no other theory of human
behavior that I know about could come close to passing, even for the
simplest possible experiments.

If you're done Demo1 but didn't understand these underlying features, I
suggest running it again with all this in mind. If you haven't done it,
what are you waiting for?

Best,

Bill P.

···

I wasn't doing this consciously, of course. But I remember thinking
that this was a very unfamiliar way of looking at behavior. In
fact, I had no existing organizational framework in terms of
which to see behavior. People walked, talked and (if I was lucky)
gave me a kiss. I didn't see behavior in terms of a particular
organizational scheme. But, by the end of the course I was able
to see behavior through S-R glasses: people "emitted" walking,
talked in response to questions and kissed in response to begging.

So I learned to perceive behavior in terms of the prevailing
organizational framework of psychology: S-R. When I started to
learn Powers' control theory (now PCT), I had to learn to see
behavior in terms of a completely new organizational framework;
control. I didn't realize this right away. At first I approached
PCT as "just another theory of behavior". I had no idea that I
was looking at _behavior_ itself from the point of view of the
wrong organizational framework.

I think I didn't really understand PCT until I learned to look
at behavior from a control perspective: the organizational
framework of PCT. My ability to see behavior in this way
came _after_ I understood the model itself; I knew how to write
a program to implement a simple control model before I was
really able to see behavior itself as a control phenomenon.

When I talk about "applying PCT" I am talking about applying
the organizational framework of control to our perception of
real, everyday behavior; I am talking about learning to _see_
behavior through PCT glasses.

Seeing behavior as control is a skill that must be learned, just
like the skill of seeing behavior in S-R terms. I don't believe
we come into the world with a tendency to see behavior in terms
of S-R or control. At least, I didn't.

What I want to discuss in this thread about "applying PCT" is not
which organizational framework, S-R or control, is the right one.
If I wanted to know that, I would, indeed, have to collect data. But
I think my concerns come before data collection. I don't think one
would even be interested in getting the data to decide the question
of S-R vs control unless they had learned to see behavior from the
point of view of _both_ organizational frameworks.

I think the reason the data doesn't exist to decide between the
organizational schemes of S-R and control is because people have
_not_ learned to look at behavior through PCT glasses. People
won't do research to determine whether, say, "classical conditioning"
is S-R or control if they can't see "classical conditioning"
through PCT glasses.

Also, I think there are plenty of people who don't (and will never
do) research who would, nevertheless, benefit from learning how
to see behavior through PCT glasses.

Best regards

Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
MindReadings.com mailto: marken@mindreadings.com
www.mindreadings.com

42

[From Bruce Abbott (2000.09.24.1340 EST)]

Bill Powers (2000.09.24.0233 MDT) --

I know of only three people connected with PCT who have formulated new
control experiments, devised a control-system model to explain them, and
done the experiments to test the model against real human behavior: Rick
Marken, Tom Bourbon, and I. Bruce Abbott has done some modeling but only as
thought-experiments; though he is perfectly capable of doing it, he has yet
to devise a new control-system experiment with a human being, write the
model as a simulation, and commit the model to a test against reality.

Point well taken, but I would like to point out that I have done a bit more
than "thought experiments." One example is the study I ran _every day_ for
_300_ days in an effort to determine whether lab rats would demonstrate good
control over their body weights (and/or food intake) as I disturbed the
amount of food available to them in their home cages. (The rats could make
up for any shortages there by earning extra food in daily one-hour sessions
in an operant chamber.) Although we (Bill and I) did develop a control
model to fit to the data, the actual match of prediction to data was
disappointing. In my judgement there were at least two reasons for this
poor fit. First, the rats may have been altering their metabloic rates
along with changing their earned food-intake as means of control; I was not
equipped to monitor the rats' metabolic rates, and if these were changing,
they needed to be included in the model. Second, I believe that I ran into
short-term regulatory effects that shut down food consumption in the operant
chamber short of the quantity needed to maintain food intake and body weight
in the long term. I could do better, but it will require equipment (and the
money to buy it) that I do not currently have.

A second study tested the prediction of one of Bill's models. For some
time, Bill had been suggesting that a simple control-system model would
account for changes in lever-press rates on schedules of reinforcement that
occur as the reinforcement rate of the schedule is varied. In this model, as
the rate of reinforcement (food pellet delivery) experienced at a given rate
of lever-pressing is lowered, the rats should increase their rates of
lever-pressing in an attempt to restore the rate of reinforcement to its
former level. The prediction was tested using both ratio and variable
interval schedules of reinforcement. On the ratio schedules, reponse rate
remained constant at all ratios tested. On the variable interval schedules,
response rate declined as the programmed reinforcement rate declined.
Neither result agrees with the predictions of Bill's simple control-system
proposal. Obviously, a better model is needed, and I feel confident that the
better model will be another control-system model. But the system is
evidently more complex than originally thought; in its final form I suspect
that the model will need to include a number of different controlled
perceptions, some in conflict with others.

With respect to human experiments, while learning how to write
control-system simulations, I developed a novel test situation involving the
"inverted T" illusion. Two alternative control-system models were
developed, the objective being to determine which of two alternative
controlled variables produced a better fit to the data collected --
difference in length or ratio of lenghts. Not earth-shaking, I know, but it
does show that I've been doing more than "thought experiments," although, it
must be noted, thought experiments can be useful, too.

I have also done several simulations that serve as "proofs of concept."
They show that a proposed system does indeed perform as expected, and that
such a system might capture essential elements of some real system. Rick's
initial fielder simulation provides an example, and so do my several
simulations of hypothetical e. coli.

I know that I need to do more in the way of formulating and testing new
control models (where does the time go?); that is why I stated at the outset
of this post that Bill's point is well taken. But I've certainly done more
work on PCT than just conducting a few "thought experiments."

Bruce A.

43

[From Bill Powers (2000.09.24.1317 MDT)]

[From Bruce Abbott (2000.09.24.1340 EST)--

Although we (Bill and I) did develop a control
model to fit to the data, the actual match of prediction to data was
disappointing.

Actually the fit of the model to the data was good enough to produce
correlations of well over .9 between the model's weight changes and real
weight changes, with eating rate as the means of control. I was puzzled at
the time when you seemed to be declaring that the model didn't work. We had
to go to a simpler hypothesis, since weight was not well controlled, but we
did, I thought, come up with at least one reasonable model. One of the
problems with fitting a model was the 19 (if I remember right) major
changes in experimental conditions that took place during the runs. I was
able to incorporate some of them in the model, but not all. We never did do
an entire run (6 months or so) with the same set of conditions in effect
from beginning to end (for the models I was involved with).

A second study tested the prediction of one of Bill's models. For some
time, Bill had been suggesting that a simple control-system model would
account for changes in lever-press rates on schedules of reinforcement that
occur as the reinforcement rate of the schedule is varied. In this model, as
the rate of reinforcement (food pellet delivery) experienced at a given rate
of lever-pressing is lowered, the rats should increase their rates of
lever-pressing in an attempt to restore the rate of reinforcement to its
former level. The prediction was tested using both ratio and variable
interval schedules of reinforcement. On the ratio schedules, reponse rate
remained constant at all ratios tested. On the variable interval schedules,
response rate declined as the programmed reinforcement rate declined.
Neither result agrees with the predictions of Bill's simple control-system
proposal. Obviously, a better model is needed ...

Correct. But we also need better data. I was modeling on the basis of
reports that rats could vary their rate of bar-pressing as schedules or
other parameters of the experiment were changed. What we found is that the
rats can't alter their rates of _contiguous_ pressing at all (despite what
is reported in the literature): they are either pressing as fast as they
can, or they aren't pressing at all (and in many cases have left the
vicinity of the lever). So I had a mistaken idea of what the phenomenon was
that was to be modeled. You remember this as a failure of the model; I
remember it as a case of having been misled by what people said in print
about variable rates of bar pressing.

With respect to human experiments, while learning how to write
control-system simulations, I developed a novel test situation involving the
"inverted T" illusion. Two alternative control-system models were
developed, the objective being to determine which of two alternative
controlled variables produced a better fit to the data collected --
difference in length or ratio of lenghts. Not earth-shaking, I know, but it
does show that I've been doing more than "thought experiments," although, it
must be noted, thought experiments can be useful, too.

My apologies, I had forgotten that one. So you are to be included in the
ranks of modelers, and that makes a (pitiful) total of four. Richard
Kennaway, of course, is the other major modeler I forgot to mention!!! So
it's five.

Hell, that ought to be plenty.

I have also done several simulations that serve as "proofs of concept."
They show that a proposed system does indeed perform as expected, and that
such a system might capture essential elements of some real system. Rick's
initial fielder simulation provides an example, and so do my several
simulations of hypothetical e. coli.

The E. coli simulations were interesting, and I thought the learning
aspects were useful. But they're still thought experiments.

I know that I need to do more in the way of formulating and testing new
control models (where does the time go?); that is why I stated at the outset
of this post that Bill's point is well taken. But I've certainly done more
work on PCT than just conducting a few "thought experiments."

Yes, I'll agree to that, and I'm sorry to have downplayed your
accomplishments. I have hoped, however, that you would try some of the
simpler experiments -- checking out some possible controlled variables, for
example, just to get the experience of doing it, without some complicated
hypothesis in the background. This isn't to prove anything to you; it's
just that the experience is worth having.

And actually, I think you could make a big splash if you used that A/D
converter I sent you to get rats controlling some continuous variables by
using continuous outputs instead of the repetitive actions they are so poor
at using. Of course maybe they'd have problems with continuous actions, too.

Best,

Bill P.

44

[From Bjoern Simonsen (2000.09.24.21.50 GMT+1)]

from Rick Marken (2000.09.22.1100)

To quote Mickey Rooney from nearly every Andy Hardy movie
ever made "Hey, I've got an idea!" Why don't you talk about
applying PCT. I don't mean applying PCT to a particular
human endeavor, like management or education. I mean
applying the PCT model to _any_ observable phenomenon.

I fall for this idea (if I understand you correct).

Why don't we (all of us) try to apply the PCT model to one special
phenomenon from Wednesday 2000.09.27. and closure Friday 2000.10.27.
We write a log every day and we present the log Friday 2000.10.27.
in a letter like this under the theme/subject "My log for applying
the PCT model" I will remind you for mailing the log Wednesday
2000.10.25. And I will remind you for writing the log the first days
and some days now and again.

I will start for myself next Wednesday and separately everyone come
into line or neglect it as we _choose_.
I'll try to find time for generating an limited Excel model for this
happening.

It shouldn't be necessary to name _an observable phenomenon_ we can
apply the PCT model on. But I will mention tree phenomenon where I have
applied the PCT model earlier.

1. I stopped smoking. (This I did before knowing PCT, but today
        I see I used the PCT model in coarse features.)
2. I decided to get out of bed at 04.44 every working day.
3. I decided to carry out the same 20 minutes early morning exercises
        every morning (also Saturday and Sunday some later than 04.50)

For myself I will start with the observable phenomenon "to have a 30 minutes
walking tour every day at 18:00 - 18:30 from 2000.09.27 until 200.12.31."
I guess I will get a chance to control different perceptions trying this.

I will mention in detail and substantiate how I changed my behavior to get
out of
bed at 04.44. Here I accept feedback (in the popular meaning of the concept)
of all kinds.

a. At the day N I chose to start my day 04.44 from day N+1.
        I didn't have a reference for getting out of bed at 04.44. (I used to
        get up at 05.15.)
b. At day N I used muscle relaxing technical before I fell asleep. In the
relaxing
        situation I formulated the personal thesis. "To morrow I will awake at
04.44.
        I introduced for my inner sight the clock radio showing 04.44. And I
ordered
        the clock radio to switch on at 04.44
c. I did this for a fortnight (if I remember correct). The first mornings I
        controlled the perception of the clock radio switching on by getting out of
bed.
        The reference for perceiving 04.44 was a.o. "don't stay in bed".
d. After a fortnight I ordered the clock radio to switch on at 04.45 and I
        awaked before the clock radio turned on. Later I put it back to 04.44 and
        most often I awake before the radio switches on. But that doesn't' matter
        I have got out of bed 04.44 five days a week and most often I awake at the
        same time Saturdays and Sundays, but these days I seldom get out of bed
before
        between 06.00 and 07.00.

I am reluctant for the daily walk between 18.00 and 18.30. Most because it
shall happen
every day. But later I'll change behavior and look forward to at the walk.
This I'll do
by establishing a reference like " a walking trip between 18.00 and 18.30 is
wonderful".
This I'll establish with reorganization.

Best

Bjoern

45

Bill Powers wrote:

My apologies, I had forgotten that one. So you are to be included in the
ranks of modelers, and that makes a (pitiful) total of four. Richard
Kennaway, of course, is the other major modeler I forgot to mention!!! So
it's five.

Hell, that ought to be plenty.

And you forgot me: I also feel like an active modeler! So we have six!

Wolfgang

46

[From Wolfgang Zocher (2000.09.25.0930 MESZ)]

Just an addendum to my last posting:

I'm doing _all_ my modeling on the base of measured data. And I try to
fit my models with more than one source of data available. This makes
all my modeling _very_ time consuming but I think this is the only way
to prove that the ideas behind PCT are right. Thought experiments are
something like Hesse's "Glasperlenspiel" ...

Best,
Wolfgang

47

[From Bill Powers (2000.09.25.0733 MDT)]

Wolfgang Zocher (2000.09.25) --

And you forgot me: I also feel like an active modeler! So we have six!

Yi yi yi! My brain is turning to mush. Who else is there? It's getting
crowded in here -- too many modelers.

Best,

Bill P.

48

[From Bill Powers (2000.09.25.0736 MDT)]

Wolfgang Zocher (2000.09.25.0930 MESZ)--

Just an addendum to my last posting:

I'm doing _all_ my modeling on the base of measured data. And I try to
fit my models with more than one source of data available. This makes
all my modeling _very_ time consuming but I think this is the only way
to prove that the ideas behind PCT are right.

... or wrong. I agree completely with your approach.

There is a use for thought-experiment modeling, which is simply to
understand how a particular kind of model works before trying to apply it
to real data. My inverted pendulum model is such a model, and so is the
Little Man. The E. coli reorganization models are other examples. This is
not an attempt to see whether a model is "right," but simply to see what is
required to make it work, what it can do, and what its properties are. It
could be that the model has no real application at all. Or it could be that
some serious modifications have to be made before the model can be used
with the real behavior. Isaac Kurtzer is helping me modify the Little Man
to fit the latest of what is known about the muscles and spinal reflexes,
so for the first time it could be considered a model of the real human
system, in the sense that your models are real models of the oculomotor
systems. But it has been interesting just to see what is needed to make
such a model work, and what it can and can't do.

Thought experiments are something like Hesse's "Glasperlenspiel" ...

I guess that's what is translated in English as the "Glass Bead Game." Now
I am going to have to read it again to figure out what you mean, unless
some kind person will tell me what you mean, in baby talk.

Best,

Bill P.

49

[From Mike Acree (2001.09.28.0926 PDT)]

Thanks to Bill P. for several magnificent posts on the attack. Especially since we have have disagreed specifically on political issues, it is very gratifying to see one statement after another that I wholeheartedly (emphasis on the heart) endorse. All beautifully said.

David Goldstein (09.28.2001.1200noon)--

If we wanted to know what was in the mind of the people who carried out these acts, we >would have to interview them. We would probably want to do something like the MOL.

I have always been amazed that the clinical psychology students I teach believe that the way to understand why people do the things they do is simply to ask them. One would think psychotherapists, of all people, would recognize how often we don't know, or have difficulty articulating, our motives. I think what people say about their motives is relevant, but only one source to be considered among several. The MOL might well get at what we want, but it requires a foundation of trust, which would surely be a challenge to establish in this case. (David: I recognize that you may not disagree with any of this, but the general points seemed worth making.)

I've been out painting my house for most of the past month, and am leaving for a week's vacation tomorrow, so I won't be participating in much on the Net for awhile; but I did want to acknowledge how much I appreciated these posts.

All best,
Mike

50

[From Bruce Abbott (2001.09.28.1320 EST)]

Mike Acree (2001.09.28.0926 PDT) --

Thanks to Bill P. for several magnificent posts on the attack. Especially

since

we have have disagreed specifically on political issues, it is very gratifying
to see one statement after another that I wholeheartedly (emphasis on the
heart) endorse. All beautifully said.

Yes indeed. But perhaps I can add a few additional thoughts to the
discussion. According to PCT, behavior is the control of perception, but
perception involves more than mere sensory input. Beliefs, too, are forms
of perception. For example, Bin Laden and many of his followers may believe
(perceive) the United States as the Great Satan, a great diabolical force in
the world which has to be neutralized, by whatever available means, for the
glory of Allah and the good of Islamic peoples. If so, that perception
emerged from a complex weave of factual information, misinformation, and
interpretation and is now being reinforced by explicit training in these
beliefs. What matters is not whether these beliefs are true but that a
large number of individuals have come to believe that they are true, and
therefore that any means are justified in order to bring our civilization,
or at least its influence on the Islamic world, to an end. People act to
control their perceptions, whether those perceptions accurately reflect or
wildly distort the underlying reality.

I agree that we as a country have done things which have contributed to
these perceptions, but now that they are established in some, almost
anything the United States does will be interpreted by them in the light of
those perceptions, i.e., in the worst possible way. And those who are not
quite sure will be looking to our behavior to decide whether we are in fact
the beasts the terrorists take us to be.

Given the current dynamic it is unlikely that we will be able to do anything
to convince the terrorists that we are not who they think we are, and we
will not be able to eradicate them by mere exercise of military force. But
there are probably many things we as a country and a government can do to
influence in a positive way the perceptions of less radical-minded people in
the arab/muslim world, and these we must do if this clear and present danger
is to be contained, for in the end only the impassioned efforts other
muslims/arabs to counter those radical beliefs among their own, and to root
out those who cannot be reformed, will put an end to terrorist attacks
against us.

Bruce A.