The Curse of Control Theory

[Martin Taylor 920309 11:00]
(Gary Cziko 920309.0745)

Gary says that the key to seeing control is the finding of zero or near zero
correlation. Inasmuch as the correlation between almost any two variables
in the universe is very near zero, that test would lead to the conclusion
that almost everything is linked by control.

(Bill Powers and Rick Marken 920308)

Bill and Rick both assert that control is a directly observable phenomenon.
I had said I thought it was a theoretical construct rather than a direct
observation, by which I meant something like "blueness" or "tastiness."
I think we are likely to get into an unnecessary war of words, here. To
Bill I would suggest that we have long agreed that the discovery of a possibly
controlled percept is often difficult and requires insight before the Test
can be applied. Even then, it is usually not easy to determine what
disturbances are occurring, and it is only for very low levels that the
outside observer (experimenter) can determine with precision that the
putative controlled variable in the subject's perception is close to the
one "observed" by the experimenter.

To Rick, I would point out that William James lived some time before Bill
Powers. One may credit Bill with a lot, but not with the discovery that
people achieve one purpose by variable means. And I doubt that James would
have taken it as an original observation, when he said: "Provided the same
conclusion be reached, the means may be as mutable as we like, for the
"meaning" of the stream of thought will be the same. What difference does
it make what the means are? "Qu'importe le flacon, pourvu qu'on ait l'ivresse.""

If, as an experimenter, one can presume some pattern in the mutually observable
environment represents a perceptual variable being controlled by the subject,
then one can attempt to disturb that pattern and see whether the subject
acts so that the pattern is restored or maintained. The pattern will show
little correlation with the experimenter's disturbances or with what the
experimenter observes of the subject's actions. If the experimenter happened
to be correct that what she did would have disturbed the pattern if the
subject had not been there, then there is evidence that the subject is
controlling. The presumption that the experimenter would have disturbed the
pattern is just that, a presumption. It is not an observation, because it
didn't happen. Explaining why things do not happen is trickier than providing
rationales for why they do happen. The failure of a presumed "cause" is
easier to justify as that it was not a cause than as that an exactly
countervailing cause was applied at the same time. I think this is at the root
of the communication difficulty with cause-effect psychologists. Causes have
effects, and PCT is supported when what should be causes are observed to
have little or no effects.

In PCT terms, one can model a subject and an experimenter seeking the
subject's perceptually controlled variables as being two control systems
with conflicting references. The experimenter's reference is that the
presumed percept of the subject should be altered, and the subject's
reference is that it should not be. This is an interesting contrast to
the description of communication, in which the two parties are assumed to
have common goals (that each should be satisfied that the communication has
happened as the other desired). And that raises interesting questions about
ethics in psychological experimentation.

Martin

[From Gary Cziko 920309.0745]

Bill Powers (920308.1800) noted concerning the phenomenon of control:

The controlled outcome will show a low correlation with both the actions of
the system and the independent disturbance. The better the control, the
lower the correlation. Therefore the observed outcome or controlled
variable will be discarded by a statistical analysis because it shows no
significant relationship to either "stimuli" or "responses." It will not be
recognized as an outcome of behavior.

This, I believe, is the hardest part of understanding control and getting
others to see it happening in living organisms. A high correlation (either
positive or negative) is one which we have been taught (even since
psychology became "scientific") is important. Low ones (close to zero)
mean that two variables are unrelated--and yet it's amazing how low the
correlations can be and still be considered "significant" and important if
they are between a type of stimulus (independent variable) and response
(dependent variable).

I wonder if control theory will ever gain widespread understanding because
of this absolutely foreign perspective on correlation. I'm not even sure
that *I* really understand how the "correct inference" involves taking the
path of the close-to-zero correlation. If I didn't keep going back to
DEMO2 I think I would be lost to CT as well. And the fact that in no other
science (that I know of) uses low correlations to find out what is
happening doesn't help matters. Bill, is it the case that even the
engineers who create artificial control systems don't need The Test since
they know what they want to control and it becomes quickly obvious if they
have succeeded or not without calculating the low correlations?

I therefore strongly encourage Bill and any others who have insights on how
this can be made understandable to post them to the net. Perhaps the CT
old-timers have forgotten how difficult this radical idea is that the
important relationship is shown by a near-zero correlation. I wonder how
many people on this network have even a basic understanding of this. I
think that this is really the CURSE OF CONTROL THEORY (which is, of course,
also it's most important insight).--Gary

···

------------------------------------------------------------------
Gary A. Cziko Telephone: (217) 333-4382
Educational Psychology FAX: (217) 244-40538
University of Illinois Internet: g-cziko@uiuc.edu
1310 S. Sixth Street Radio: N9MJZ
210 Education Building
Champaign, Illinois 61820-6990
USA
------------------------------------------------------------------

[from Gary Cziko 920309.1220]

Martin Taylor (920309 11:00) responds:

Gary says that the key to seeing control is the finding of zero or near zero
correlation. Inasmuch as the correlation between almost any two variables
in the universe is very near zero, that test would lead to the conclusion
that almost everything is linked by control.

You see, I was right--the curse strikes again.

What Martin's comment points out seems to be that the low correlation
between the controlled outcome and the actions of the system is a special
type of low correlation. It is not a low correlation given by an amorphous
blob of points on a scattergram indicating that any value of x can occur
with any variable of y. Instead, it is a low correlation given by lots of
different values of x (action of the system and/or independent
disturbances) occuring with only a small range of values of y (controlled
outcome).

So a low correlation is not enough. It is a special type of low
correlation. And it would disappear of the perceptual link between
controlled outcome and behaving system were eliminated.

I hope Bill and/or Rick will have something to add to this.--Gary

···

------------------------------------------------------------------
Gary A. Cziko Telephone: (217) 333-4382
Educational Psychology FAX: (217) 244-40538
University of Illinois Internet: g-cziko@uiuc.edu
1310 S. Sixth Street Radio: N9MJZ
210 Education Building
Champaign, Illinois 61820-6990
USA
------------------------------------------------------------------

[From Rick Marken (920903)]

Martin Taylor (920309 11:00) says:

To Rick, I would point out that William James lived some time before Bill
Powers. One may credit Bill with a lot, but not with the discovery that
people achieve one purpose by variable means. And I doubt that James would
have taken it as an original observation

Yes, James knew purpose. I might add that Tolman did too, And McDougall
apparently talked about it knowledgeably. So I will refine my claim. Powers'
contribution was to quantify the phenomenon of control and show that it
applied at all levels of behavioral organization -- from muscle tensions on
up. He also presented a quantitative model of the phenomenon.

I do think that James did understand the nature of the phenomenon
of control. But because he lacked a model and a way to quantitatively
demonstrate purposeful phenomena he never made much of an impact on the
study of purpose. James wrote eloquently about purpose in the first chapter
of his "Principles.." but the concept is basically abandoned in the rest of
the text, and in psycholgy from that time on, I might add. Quite a different
story in Powers' test BCP.

Gary Cziko (920309.1220) says:

So a low correlation is not enough. It is a special type of low
correlation. And it would disappear if the perceptual link between
controlled outcome and behaving system were eliminated.

I hope Bill and/or Rick will have something to add to this.--Gary

Yes, a low correlation is not enough. The special type you mention is
what you would expect if the reference level of the controlled variable
is fixed. But if it is not then you might see something other than a
horizontal line on an scatter plot -- it could even be a
circular cloud. What is needed besides a low correlation is knowledge
that there WOULD be a HIGH correlation between disturbance and controlled
variable if the variable is NOT controlled. What has been forgotten in this
discussion of testing for controlled variables is that the researcher
chooses the disturbance because s/he knows what its effect will be if there
is NO control. In my mindreading demo I know exactly what the effect of
the disturbance would be if a number on the screen is NOT controlled. So
a different relationship (such as a low correlation) suggests control -
NOT because the correlation is low (it will be fairly low for all the numbers)
but because it is not what is EXPECTED. So, in order to test for control
you must have a pretty good model of how variables interrelate when there is
NO control. And even when you do get results that are unexpected, based on
the non-control models, you can make yourself more certain that control is
occuring by watching it break down ( as Gary noted) when you prevent
the system identified as the controller from perceiving the hypothetical
controlled variable. This can also be shown with the mindreading demo --
when the subject shuts his/her eyes the effect of the disturbance to
the controlled number is once again just what is EXPECTED; control has
disappeared.

Does anyone have a spare million bucks for me so I can do this all the time?

Ah well, back to work.

Rick M.

···

**************************************************************

Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
Internet:marken@aerospace.aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)