SR/PCT;Explaining with PCT;Attractors;etc.

[From Bill Powers (930514.0700)]

Bob Clark (9305 --)

I guess the passage of yours that confused me was this (930509):

In any situation where the Reference Level is essentially
constant, the system can be described in S-R terms. Each
portion of the system operates as a cause-effect unit (S-R).
When the units are combined, the loop can still be described as
in S-R terms. (I believe Powers has pointed this out
somewhere.) S-R type experiments are designed to keep any
relevant Reference Levels fixed, unchanging, during the
experiment. This can be done at any level of the Hierarchy.
OF COURSE the results fit the S-R description! And, naturally,
all studies that are inconsistent with this form of
experimental design are rejected.

The S-R appearance of behavior does, in its purest form, require
a constant reference level as you say. When the reference level
varies normally, the S-R appearance become statistical, because
only the _average_ reference level and therefore only the
_average_ relationship between S and R can be seen. This is the
normal state of affairs in S-R (or S-O-R) psychology, probably
accounting for much of the apparent variability in observed S-R

However, even if the reference level is constant, the appearance
of a stimulus-response relationship still involves a
misidentification: seeing a disturbing variable as the stimulus
that the organism is sensing. The actual variable being sensed is
not allowed to change, or if it does change it is restored
immediately to its former state. The disturbing variable changes
and the action of the organism changes in a commensurate way, but
the controlled variable actually being sensed is _prevented_ from
changing. The action prevents the apparent stimulus from actually

I felt that your paragraph above gave the impression that keeping
reference levels constant is the only requirement for
interpreting behavior as a response to a stimulus. In addition,
one must mistake a disturbing variable for a sensory input

I presume that you didn't make that mistake, but your comment
didn't make that clear to the reader because it seemed to make
constancy of the reference level the only condition under which
the S-R mistake could be made -- and implied that with a constant
reference level, the S-R explanation worked. Your example of the
salt shaker is off target, because here you're picking an example
in which the controlled variable -- perceived saltiness of the
food -- is initially different from the desired amount, and you
"reach for the salt shaker and use it." There is no disturbance
as such in this example, and so there is nothing to be mislabeled
as a stimulus. The S-R explanation would have to go far afield to
find a causal stimulus, because the act of reaching for the salt
shaker and using it is apparently spontaneous. The desired amount
of saltiness can't be observed as an external variable.

A better example would be that of adding sugar to your lemonade
after someone surreptitiously adds more lemon extract to it. The
appearance is that the adding of the extract is the stimulus, and
your adding sugar is a response to it. The response, however,
keeps the net sweetness of the drink from being changed more than
momentarily. The sweetness is the actual controlled perception,
not the sight and sound of lemon extract being added.


Oded Maler (930513.1630 -ET) --

I said:

This is quite unlike the "distributed" concept of perception
that others have offered (Martin Taylor, for example). So I
suppose that some day we will find out which model is right.

You said:

I don't know why you both keep on insisting on "one signal for
one controlled variables" - it depends any way on the scale
where you look. Even the elementary sensation is distributed in
space and time.

You're just contrasting your model with mine. If you can make a
distributed-perception model work, fine, go ahead. Then we can
compare the two models as explanations and predictors of specific
behavioral phenomena. We aren't "insisting" on one signal for one
controlled variable. We're proposing it as a model, an
alternative to the distributed-perception model that many other
people have proposed. So far our model seems adequate for
explaining the kinds of processes we observe experimentally. Some
day we may find a phenomenon that we can't explain that way, just
as you may find a phenomenon you can't explain with a distributed

I don't deny the plausibility of the general principle. I just
say that suppose I knew how to design a system that controls
a 14df arm, I would not know how to design a such a high-level
system - of course I could give qualitative description, but
I could not generate this kind of behavior in a model.

I think there is a problem here with what is meant by "explaining
behavior with a control system model." We seem to be having a
similar problem with Ken Hacker. Your remark above seems to show
that you demand a control-system model to be able already to
explain any behavior at any level, "out of the box" (that is, as
you buy it from the store). Ken Hacker, similarly, seems to
object that the PCT model doesn't explain all the interactive
phenomena he's interested in.

When I say that I wish people would start using PCT to explain
behavior, I don't mean that they should just start using the
language of PCT and arbitrarily assume that all the behavior they
see is control behavior. Doing that would be just as bad as what
people do now with other theories.

I claim, or I propose, that PCT can be used to explain all
behavior at all levels. But when I say it CAN be used that way, I
don't mean that it ALREADY DOES explain behavior at all levels. I
mean that it provides principles that can be used to construct
specific models to see whether they work properly as explanations
of what is observed.

Those specific models are not going to spring into existence all
by themselves. Somebody actually has to do the work. What I'm
saying is like saying that finite-element analysis can be used to
explain the structural properties of a truss bridge. The methods
and principles of finite element analysis that you find in a book
will not do any explaining by themselves. Somebody has to make
measurements and fit the analysis to a specific truss bridge,
then calculate the variables of interest such as deflections
under loads. Only when that is done is there any prediction to
compare with the actual behavior of the real truss bridge.

It's the same with PCT. To try out the PCT model as an
explanation of any behavior, you must first analyze the behavior
in terms of possible controlled variables, possible output
processes, and all the rest of the elements of the canonical
diagram. You have to make specific proposals, then convert them
into definitions of each block in the diagram. Then you have to
make this diagram runnable as a simulation, or calculate/deduce
its behavior from the properties you've given it. Only after all
that is done can the model predict a specific behavior to compare
with what the real person or group of persons does. Only when the
model's behavior finally does approximate that of the real system
or systems can you try varying the conditions to predict how
people will behave outside the original circumstances used to fit
the model to behavior.

When I suggest a PCT explanation for a behavior, I'm just going
through the first tentative steps of this process, sketching in
some ideas for how the model might be constructed for a specific
situation. From the reactions to these suggestions, I get the
impression that these preliminary sketches are being taken as the
final product -- "the PCT explanation." This, I suspect, is how
most theories in the behavioral sciences are used. If the theory
can lead to a plausible description of behavior, that is all that
is required of it.

But that isn't how we do research in PCT. I think that much of
the critical reaction to the experimental work of Marken and
Bourbon and me comes from not understanding what we're trying to
do. We have to use simple experiments because we're still trying
to verify that the model predicts correctly in all situations
that we can imagine AND IMPLEMENT AS EXPERIMENTS. This is all
ground-breaking work. It doesn't lead to making big important
announcements about matters of international social importance.
It doesn't touch on whatever trendy fads are occupying the
attention of others. All it does is to establish one case after
another in which PCT leads to models that predict human behaviors
in great detail and with high reliability. Sure, they're simple
behaviors. But is there any other science of behavior that can
predict even such simple behaviors with comparable accuracy? Not
within an order of magnitude.

So far we haven't found any failures of the model. Every time we
see how to set up a new control experiment, we find a PCT model
that works, that behaves just like the real people do. I can't
think of any reason to abandon this path, or any reason why it
will not lead naturally, step by step, to similar results with
behaviors of increasing complexity.

As far as I'm concerned, the rest of the scientific world can go
right ahead with trying to guess how it's all going to come out,
hoping to find big abstract principles and generalizations that
will reveal the truth of it all. That has nothing to do with me,
or PCT, or as far as I'm concerned, science. I want to attract
people into this work who are willing to go step by step,
building up a picture of behavior that we can PROVE fits the way
people actually work. I'm not interested in doing big flashy
bellyflops into the sea of knowledge, hoping to scare up a few
fish. I'm weaving a net that we can pull systematically along,
from which those fish can't escape.

It is not the point that I speak about "correct/incorrect"
which imply some kind of reference, value-judgement etc. Take
some neutral adjective that you want to associate with the
current state of the US economy/society - can you define it as
a "perceptual" function (viewed by some omnipotent observer) of
the elecrodes in the pockets, stomachs, prisons, etc. all over
the US?

The state of the US economy/society is a perception, period. It
has no state of its own. You can make models of it and speak of
the state of the model. You can say the model is behaving the way
you want it to, or you can see something wrong with its behavior
and try to fix it. But when you look at the actual US economy,
all anyone can see is a lot of people doing what they're doing.
There isn't any "economy" there. That's a perception constructed
out of all that you see going on. Different people see different
societies, different economies. None of them are right; none of
them are wrong. They're just reporting their own perceptions. The
manufacturer sees a recovery under way; the working man sees a
recession getting worse or at least not getting better. They are
both right. That's what they see.

Ideally, for an opinion to be worth something, there should be
first an hypothetical function that calculates the objective

But how can you know that the function calculates the objective
state-of-affairs? You have no way other than through perception
to know anything of the objective state-of-affairs. So you can't
tell whether the function fits or not. The function is a
perception of one kind; your knowledge of the state-of-affairs
comes from perceptions of other kinds. You are just comparing one
kind of perception with another. If I ask you to describe the
objective state-of-affairs, how will you answer me? You will tell
me about the world as you experience it. If we're talking about
the density of lead, it's perhaps easy to find agreement among
people's experiences. But when we talk about the state of an
economy or a society, agreement is very hard to achieve,
sometimes impossible. There is nobody to whom you can turn, and
say "Tell me, what is ACTUALLY going on out there?"

Then there would be a perceptual function, constructed only
from the limited sensing and computing capabilities of the
individual, but still somehow yielding some correlation with
the CEV.

But how are you going to find out about the actual CEV without
using your perceptions, or somebody's? To calculate the
correlation, you must know both the perception and the thing
that's causing it. But you can't peer around the edges of your
visual field to see what is causing your visual impressions. You
can't get outside your skin to see what is creating the
sensations of touching things. You can't intercept sound waves
before they get into your inner ear, to see what vibrations they
actually carry. You and every person alive is stuck inside the
sensory interface, looking only at what the sensory signals can
tell you. The CEV is what you perceive; it is a construct of the
brain. It's inside.

In order for this to be of any relevance to the world, not only
should your poor perceptual function correspond to the real
thing, but also your imagined abstract action X could be
realizable in the world.

Your perceptual functions may be poor, but they're all you have.
There is no a priori way of knowing what action will be effective
in controlling any perception. Each person has to learn that
alone. We know that an external world exists, because it can
affect our perceptions, and we can affect our own perceptions by
acting on that external -- but unseen -- world. The relevance to
the world is contained in the fact that we can systematically
alter our perceptions by learning what actions to take. Try
starting from that basis, and see how your problem looks then.

All people control for their perceptions, but will you ignore
the different types of hierarchies that might develop in
different groups, cultures and geographical regions?

No, I will try to model them if I should live so long. These
phenomena develop out of the properties of the individual person
and the environment as it exists at any time. It's because people
control their own perceptions that there are groups and cultures.
PCT may some day help us understand why the phenomena of which
you speak exist. They don't exist just because they exist. They
exist because of the properties of the people involved. And the
properties of the people involved come to be as they are because
each individual is a self-organizing hierarchy of control systems
living in a physical and social environment -- or that is how I
expect the story to turn out.

What attracts the ball is the combination of gravity and
geometrical constrains. These are local conditions that imply
global convergence. I (personal taste) think it is a very
powerful and useful metaphor (well, apparantly also abuseful).

I don't think it's expecially powerful. Nothing actually
"attracts" the ball. "Attraction" is a term that comes from
prescientific times, along with the "impulse" that keeps objects
moving and the "abhorrence of a vacuum" that makes air rush into
an evacuated container. I much prefer an actual physical analysis
of the situation, which explains the behavior completely without
invoking unwanted images. With a physical analysis, we can see
fundamental differences between physical systems that happen to
be described by similar mathematical forms and abstractions. I
don't object to the forms and abstractions; only to reifying them
and endowing them with mysterious powers of explanation that
don't explain anything.
Chuck Tucker (930513) --

Subj: After a glance I say

My name is NOT Chuck Tucker it is Charles Wright Tucker, Junior
and your name is NOT Bill Powers but rather William T. Powers.

Better read again. My point was not what your name is, but that
you can think any thought and know without a doubt that you are
thinking it. This is direct contact with part of Boss Reality.

... there is no difference epistemologically or ontologically
between PCT and S-R.

So what? There is no such difference between the phlogiston and
oxygen theories of combustion -- but one explains observations
that the other does not.

I think my scheme avoids that problem. I am sure you disagree
on that point.

I do, because you are making factual statements about things you
can't know for a fact. My view is agnostic. When you put physical
models together with neural models of behavior, you have to
conclude that the brain can't know the relationship of its
experienced reality to whatever lies outside it. For all we can
tell now, the brain's picture is perfectly accurate. Or
systematically distorted. But there is evidence that can be taken
as pointing to existence of an external reality, and that is what
I go on -- pragmatically.

By the way, re: you second post today, I really would like to
hear your answer to a question I raised yesterday. When you give
instructions to yourself, who is the giver, and where does the
giver get those instructions? And in what form are those
Bruce Nevin (930510.0923) --

RE: Joint paper.

Good suggestions. An outline is called for, as you suggest.

Right now I'm waiting to collect examples of misstatements about
control theory. It may be that this material will be enough to
form the basis of the whole paper. I agree that reviewers'
comments as well as contents of published papers would be
approrpiate. One format that appeals is to organize the
misstatements so they can be paired with a running development of
what control theory actually says, thus weaving the two themes
together in a compact way.

Bob Clark (930513.1630) --

To me, it is pretty straightforward and familiar to perceive,
from memory, past experiences of bicycles and bumps -- followed
by using the memory as a guide in selecting a combination of
actions to avoid repeating an unpleasant experience. I think
that most people would accept this view as consistent with
their own experiences -- although they would not ordinarily
bother about reporting these internal details.

Whether you perceive these things in present time or in memory,
the perceptions or the memories can't "guide" anything -- they
are simply reports of what does appear to exist, or what did
appear to exist. The memories must still be interpreted and
compared with reference levels, and the error must be converted
into an action of a kind that will reduce the error. Unless we
can draw a diagram showing how memory has any effects other than
playing back previous perceptions, I don't think we can explain
anything by using memory. It's just too easy to throw everything
we can't explain into memory and assume that the problem is

Bill P.