Is PCT falsifiable (was Why we do that)

[Martin Taylor 2010.04.18.15.31]

[From Martin Lewitt (2010.04.17.1525 MDT)]

  I wonder if PCT is falsifiable.

I suppose it depends on what you mean by "PCT". The phrase "Perceptual Control Theory" can be, and is, used in quite a few different ways.

At one extreme is the Rick Marken usage, in which "PCT" means precisely the 11-level purely hierarchic organization of scalar control units described by Bill Powers, along with the assertion that all behaviour is the control of perception. This version would be falsified if it were to be determined that the control structure is not strictly a pure hierarchy of scalar control units, or if one even one occasion some action could be determined not to be the output of a functioning control system.

Bill P has a slightly less rigid usage, since he frequently mentions that the nature and ordering of the levels of the hierarchy are not cast in stone, and in his modelling he sometimes inverts the ordering of some levels. But Bill's usage still asserts a strict hierarchy of scalar control systems, and would be falsified if it could be shown that even one feedback loop existed within a level (as would be the case, for example, in a flip-flop decision circuit at the category level -- a quite plausible possibility). Both Rick's and Bill's versions would be falsified if it were ever determined that some controlled perception was not scalar (the analogy to the studies of perception when I was in school would be to "integrable" and "non-integrable" perceptual dimensions).

I try to use the term "HPCT" or "Hierarchic PCT" to refer to the theory when it includes a strictly hierarchic organization of scalar control systems (and I suppose I would continue to do so if the control units were hypothesized to control vector perceptions rather than scalars). In my usage, "PCT" is more general. For me, it would still be PCT if it were found that control systems existed that sensed and controlled the actions of other control systems within the same structure, such as altering their gain or changing the parameters of their sensory input functions. Such "interior" control systems cannot exist in HPCT, and the discovery of one would falsify HPCT. Likewise it would still be PCT if the value of one controlled perception influenced the value of another perception (controlled or uncontrolled). Such an interaction is disallowed in HPCT, at least in the pure version.

Going to the other extreme, at the most general level, PCT can be seen as a corollary of the thermodynamic law that entropy always increases to a maximum in a closed system, since the influence of the environment in reducing the Gibbs free energy of the environment-lifeform system is unavoidable. The structure of the lifeform can be sustained only by either isolating the structure from the environment (skin and shell) or by sensing and countering the environmental effects that would destroy the structure. This second possibility is exactly "PCT" in its most general usage. In this sense, PCT is falsifiable only by a falsification of the second law of thermodynamics.

Between the essentially unfalsifiable "general PCT" and the easily falsifiable HPCT, there is a range of possibility. For example, in some usage, does PCT claim _all_ behaviour is the output of actively functioning control system(s), or does it claim that _some_ behaviour is the effect of control, while other behaviour may be spontaneously generated, or may be intentionally output without any corresponding sensory input that must be brought near a reference value (that's two different variants, but they might be combined). The "all behaviour" version could be falsified if it could be shown that even one action was not intended to influence a controlled perception. The "some behaviour" version would be proven correct if even one action could be shown to have been performed so as to bring some perception closer to a desired value. Having proved that, the "some behaviour" variant would be unfalsifiable in the same sense that the claim "some days are sunny" is demonstrably true and is therefore unfalsifiable.

There's no value in going through all the different possible meanings of "PCT", at least not for determining whether PCT is falsifiable. If you want to test a theory, you have to specify what that theory claims. Once you have done that, you can assess whether it is falsifiable. But "PCT" is a moving target, and if you falsify one version, there are lots of others awaiting your arrows. It is much easier to assess the falsifiability of a version of PCT that makes a specific claim about the organization of control units than it is to assess the falsifiability of a version that just says there must be some organization without specifying what that organization might be.

So, is PCT falsifiable? Let me be perfectly clear: "Yes and No, and maybe".

Martin

[From Rick Marken (2010.04.18.2210)]

Martin Taylor (2010.04.18.15.31)--

>Martin Lewitt (2010.04.17.1525 MDT)--

�I wonder if PCT is falsifiable.

I suppose it depends on what you mean by "PCT".

As I said in my earlier post [Rick Marken (2010.04.18.1000)] I think
it actually depends more importantly on what one means by
"falsifiable".

At one extreme is the Rick Marken usage, in which "PCT" means precisely the
11-level purely hierarchic organization of scalar control units described by
Bill Powers, along with the assertion that all behaviour is the control of
perception.

PCT means neither of these things to me when I'm talking about it as a
falsifiable theory. I have always treated the proposed 11 level
hierarchy as a heuristic device. I've never tried to make the levels
in the hierarchical models I've built (such as my Coordination model--
Bimanual Coordination) correspond to the
11-level hierarchy. And the assertion that all behavior is the
control of perception is just a way of saying that psychologists study
the purposeful behavior of living systems, not non-purposeful behavior
(such as a person's acceleration as they fall from the garage roof).
But whether or not some behavior is control of not is a matter for
test.

What I so mean by "falsifiable PCT" is the process of comparing the
behavior of working models of control built on PCT principles (ie.
models that control perceptual variables) to that of actual living
systems. Models are falsified to the extent that another model (which
might also be a model built on PCT principles) handles the same data
better.

I think a nice example of "falsification" of a PCT model is shown in
my "Degrees of Freedom in Behavior" paper (1991) published in
Psychological Science, 2, 92 - 100 and reprinted in _Mind Readings_.
In the section of that paper entitled "Test for the controlled
variable" I describe comparing two different models of the behavior in
a two dimensional tracking task. Both are PCT models in the sense that
both are "control of perception" models. The difference between the
models is that one controls a Cartesian representation of the two
dimensional position of the cursor and the other controls a Polar
representation. Both models control equally well and both fit the
basic tracking behavior equally well. The difference between the
models shows up only in their response to a brief disturbance in one
dimension. The Cartesian model (like the human subject) responds to
the disturbance in only that one dimension; the Polar model (unlike
the human) responds to the disturbance in both dimensions. So the
Polar model of two dimensional control has been "falsified" and the
Cartesian model is (provisionally) accepted.

If you could develop a control model that works according to
principles that are inconsistent with PCT -- for example, a model that
controls using inverse kinematic calculations -- and show that this
model fits all the data that can be handled by a PCT-based model as
well as some new data that cannot be handled by a PCT- based model
then I think you could say that you have "falsified" PCT.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Martin Taylor 2010.04.19.10.40]

[From Rick Marken (2010.04.18.2210)]
   
Thanks for your clarifications, especially as to how you think of the specific hierarchy. In most of your writings, you give the impression of being much more rigidly tied to the specific structure than you do in this explicit posting. Here, you match the way I described Bill P's position -- I hope not in too cartoonish a fashion.

Martin Taylor (2010.04.18.15.31)--

Martin Lewitt (2010.04.17.1525 MDT)--

  I wonder if PCT is falsifiable.
       

I suppose it depends on what you mean by "PCT".
     

As I said in my earlier post [Rick Marken (2010.04.18.1000)] I think
it actually depends more importantly on what one means by
"falsifiable".

At one extreme is the Rick Marken usage, in which "PCT" means precisely the
11-level purely hierarchic organization of scalar control units described by
Bill Powers, along with the assertion that all behaviour is the control of
perception.
     

PCT means neither of these things to me when I'm talking about it as a
falsifiable theory. I have always treated the proposed 11 level
hierarchy as a heuristic device. I've never tried to make the levels
in the hierarchical models I've built (such as my Coordination model--
Bimanual Coordination) correspond to the
11-level hierarchy. And the assertion that all behavior is the
control of perception is just a way of saying that psychologists study
the purposeful behavior of living systems, not non-purposeful behavior
(such as a person's acceleration as they fall from the garage roof).
But whether or not some behavior is control of not is a matter for
test.
   
Falling off a garage roof is not the kind of possibly uncontrolled behaviour I mentioned as a potential route to falsifying the "all behaviour is the control of perception" class of "PCT" descriptions. I considered muscular spasms as one kind of uncontrolled behaviour, as well as behaviour for which the result might be hoped for, but not sensed (e.g. making a disk of human voices for extra-solar civilizations to discover in the indefinite future, so that they might know they were not alone in the Universe). We would say that the latter involves control in imagination, but to accommodate it in the definition of "PCT" we must demand that any viable definition include some version of unobservable action that influences only internal perceptions and not perceptions influenced via the external environment.

What I so mean by "falsifiable PCT" is the process of comparing the
behavior of working models of control built on PCT principles (ie.
models that control perceptual variables) to that of actual living
systems. Models are falsified to the extent that another model (which
might also be a model built on PCT principles) handles the same data
better.
   
You deal with an area that I completely omitted namely the fitting of models. I didn't think it was relevant because finding that a model doesn't fit the data very well (or worse than a non-PCT model) would in no way falsify any of the forms of "PCT" that I described. It would always leave open the possibility that a different PCT-based model might provide a better fit. Martin L asked whether PCT itself could be falsified, and failures of model fit cannot address that question. Such failures can reduce the credibility of the specific model (controlled variable(s), structural organization, parameter ranges tested ...), and a whole series of model failures could reduce the credibility of the modeller as a skillful simulator or the credibility of the types of structure modelled, but only persistent failure of PCT models to work better than the best attempts at non-PCT modelling could seriously reduce the credibility of "PCT" as such. And we know that PCT-based models do not persistently fail in those cases where they have been tried.

If you could develop a control model that works according to
principles that are inconsistent with PCT -- for example, a model that
controls using inverse kinematic calculations -- and show that this
model fits all the data that can be handled by a PCT-based model as
well as some new data that cannot be handled by a PCT- based model
then I think you could say that you have "falsified" PCT.
   
You still have to specify what you mean when you say "PCT" would have been falsified. No matter how you define "PCT" (beyond saying it is explicitly and exclusively defined by the models you used), there always remains the possibility you haven't hit on the right structure or parameter values or controlled variables in your modelling, and if you did find those elusive properties, your resulting model could fit better than the non-PCT model.

Personally, I don't think "falsifiability" in the sense of a black-and-white "cannot be true" is an appropriate consideration for most theories. I prefer Watanabe's formulation of credibility (S. Watanabe, Knowing and Guessing: a formal and quantitative Study, Wiley, 1969), according to which the accumulation of information (in this case quality and number of independent kinds of data fits using theory-based models) almost always leads good theories to a vastly higher level of creditation than it does bad theories. In that sense, your approach improves the creditation of some versions of "PCT" while possibly lowering the creditation of others. "Falsifiability" in this sense just means the possibility of reducing the creditation of a theory to very low levels. It's not a binary decision in the classical sense of falsifiability.

Martin

[From Rick Marken (2010.04.19.1120)]

Martin Taylor (2010.04.19.10.40)--

Thanks for your clarifications, especially as to how you think of the
specific hierarchy. In most of your writings, you give the impression of
being much more rigidly tied to the specific structure than you do in this
explicit posting.

Maybe in my writings on the net but certainly not in my published
research work.

Falling off a garage roof is not the kind of possibly uncontrolled behaviour
I mentioned as a potential route to falsifying the "all behaviour is the
control of perception" class of "PCT" descriptions. I considered muscular
spasms as one kind of uncontrolled behaviour, as well as behaviour for which
the result might be hoped for, but not sensed

Muscular spasms are almost certainly output oscillations resulting
from an improperly tuned (in terms of gain and/or slowing) control
loop. "Hoping" is obviously a control process. But so what? PCT
doesn't say that all behavior is control of perception; PCT says (if
anything) that only control (the behavioral phenomenon that control
theory explains) involves control of perception.

You deal with an area that I completely omitted namely the fitting of
models. I didn't think it was relevant because finding that a model doesn't
fit the data very well (or worse than a non-PCT model) would in no way
falsify any of the forms of "PCT" that I described.

That's why "falsify" seems like a poor term for what is done in
science. I think the only thing we can do in science is see which
model fits _all the relevant data_ better than another. A good example
is the Bourbon and Powers research described in their "Models and
their worlds" paper (I think it's on the net somewhere but I couldn't
find it). What they showed is that an S-R, Cognitive and Control model
make equally good predictions of behavior until you change the nature
of the "world" in which behavior takes place (by adding disturbances).
When you add disturbances the S-R and Cognitive models do much more
poorly at predicting human behavior than the control model. I would
say that this demonstration suggests that the Control model is the
only successful model of the purposeful behavior under study. I
suppose you could say the results "falsify" the S-R and Cognitive
models, but only relative to the performance of the Control model. I
think that comparing the performance of models (in terms of how well
they account for what is actually observed) is really the only way to
"falsify" a model.

Personally, I don't think "falsifiability" in the sense of a black-and-white
"cannot be true" is an appropriate consideration for �most theories

There you go!

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Bruce Gregory (2010.04.19.1440EDT)]

[From Rick Marken (2010.04.19.1120)]

PCT doesn’t say that all behavior is control of perception; PCT says (if
anything) that only control (the behavioral phenomenon that control
theory explains) involves control of perception.

Wow! Who could possibly argue with that? I’m convinced.

Bruce

[From Rick Marken (2010.04.19.1215)]

Bruce Gregory (2010.04.19.1440EDT)--

Rick Marken (2010.04.19.1120)]

PCT�doesn't say that all behavior is control of perception; PCT says (if
anything) that only control (the behavioral phenomenon that control
theory explains) involves control of perception.

Wow! Who could possibly argue with that?

Quite a few psychologists (of the "manual control" persuasion: see,
for example, "Control theory for humans: Quantitative approaches to
modeling performance" by R. Jagacinski and J. Flach, NJ: Erlbaum,
2002) and control engineers, all of whom think that control involves
the control of output.

I'm convinced.

Just because I said so? Your approach to science seems remarkably "faith-based".

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Martin Taylor 2010.04.19.14.31]

[From Rick Marken (2010.04.19.1120)]

Martin Taylor (2010.04.19.10.40)--
     
Maybe in my writings on the net but certainly not in my published
research work.

I considered muscular
spasms as one kind of uncontrolled behaviour, as well as behaviour for which
the result might be hoped for, but not sensed
     

Muscular spasms are almost certainly output oscillations resulting
from an improperly tuned (in terms of gain and/or slowing) control
loop.

Well, that's certainly a theoretical possibility. I can't see how it could fit the observed properties of spasms, such as a sudden one-way movement with only a controlled corrective follow-up, but it is possible. Do you consider your hypothesis to be a part of "PCT", such that if it turned out not to be true it would falsify "PCT"?

  "Hoping" is obviously a control process.

This sounds circular, but it doesn't fit with your following statement.

But so what? PCT
doesn't say that all behavior is control of perception; PCT says (if
anything) that only control (the behavioral phenomenon that control
theory explains) involves control of perception.
   
So, in your view, PCT is a tautology? Only control involves control of perception? All that says is the obvious, that if a property is controlled, the value that is controlled must be the value that is sensed? That's true whether the control mechanism is a thermostat or a bacterium, or anything else that controls some variable quantity. It just says "this is a property of control". I think there's a lot more to PCT than that, but if you want to reduce "PCT" to just that, I suppose your version can't be falsifiable.

I do grant you that a lot of people think that control is control of output, but even then, the output has to be sensed and controlled according to the value of what is sensed.

For you to say "PCT says (if anything)" makes it extremely hard to see that your version of PCT could possibly be a theory with any power at all. In my view, PCT is a very powerful theory, but the many uses of the phrase make it difficult to determine what about the theory is a core claim, the falsification of which would destroy the whole structure.

   

You deal with an area that I completely omitted namely the fitting of
models. I didn't think it was relevant because finding that a model doesn't
fit the data very well (or worse than a non-PCT model) would in no way
falsify any of the forms of "PCT" that I described.
     

That's why "falsify" seems like a poor term for what is done in
science. I think the only thing we can do in science is see which
model fits _all the relevant data_ better than another.

Well, at least we agree on that!

As for Bourbon and Powers "Models and their worlds", it was in the Special Issue on PCT that I edited (June 1999) of the International Journal of Human-Computer Studies, 50/6, 445-461. I couldn't find a downloadable copy on the Web, at least not without paying for it. I have the text, which I could make available, but not the figures, so far as I can discover. Maybe Bill has the full thing and could make it available.

Martin

···

On 2010/04/19 2:21 PM, Richard Marken wrote:

[From Bruce Gregory (2010.04.19.1525 EDT)]

[From Rick Marken (2010.04.19.1215)]

Bruce Gregory (2010.04.19.1440EDT)–

I’m convinced.

Just because I said so? Your approach to science seems remarkably “faith-based”.

BG: Not in the least. If all that PCT claims, if it claims anything, is that control is the control of perception, then anyone who disagrees with PCT must have something other than control in mind. If you can’t perceive the outcome of your actions, all you can do is flail about aimlessly. Inverse kinematics is perfectly adequate for flailing.

Bruce

[From Rick Marken (2010.04.19.1300)]

Martin Taylor (2010.04.19.14.31)--

Rick Marken (2010.04.19.1120)--

Well, that's certainly a theoretical possibility. I can't see how it could
fit the observed properties of spasms, such as a sudden one-way movement
with only a controlled corrective follow-up, but it is possible. Do you
consider your hypothesis to be a part of "PCT", such that if it turned out
not to be true it would falsify "PCT"?

If the spasm were a controlled variable and its behavior were better
explained by a non-PCT based model then, yes, that would cause a
problem for PCT.

�"Hoping" is obviously a control process.

This sounds circular, but it doesn't fit with your following statement.

It's obviously a control process because hoping involves bringing the
state of an imagined perception to a desired (reference) state.

But so what? PCT
doesn't say that all behavior is control of perception; PCT says (if
anything) that only control (the behavioral phenomenon that control
theory explains) involves control of perception.

So, in your view, PCT is a tautology?

Not at all. There are alternatives to the PCT model of control, as you
know. "World model"-based control models are an example. Saying that
PCT is a model of control behavior is no more tautological than saying
that Newtonian physics is a model of causal behavior.

Only control involves control of perception?

True (according to PCT) but control of perception is not the only
model of control.

All that says is the obvious, that if a property is controlled,
the value that is controlled must be the value that is sensed?

Well, there are control models that do not clearly incorporate a
perception of the controlled variable as a central feature of the
model. That's why non-PCT control theorists don't understand the
importance of defining the controlled variable in their models of
control. I explain what I see as the difference between PCT and
non-PCT approaches to understanding of control in my review of the
"Control Theory for Humans" book
(BookReview2002).

For you to say "PCT says (if anything)" makes it extremely hard to see that
your version of PCT could possibly be a theory with any power at all.

And yet, it is;-)

Indeed, one of the more powerful (and important) aspects of PCT is the
recognition that the theory applies only to the phenomenon of control
(purposeful behavior). Psychologists assume that the behavior of
living systems is a causal (S-R), not a control process. If they are
right, then PCT simply doesn't apply to the study of living systems.
PCT says that, to the extent that the behavior of living systems _is_
a control process, then a control model (like PCT) is required to
explain it. It looks a lot like the behavior of living systems _is_ a
control process (that it is purposeful) but PCT doesn't declare in
advance that all behavior is definitely control. PCT just makes us
aware of the (very good) possibility that any particular example of
behavior could be an example of control. So before saying that PCT
applies it's good to have some evidence -- such as that provided by
testing for controlled variables -- that the behavior involves
control.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Bill Powers (2010.04.19.1529 MDT)]

Rick Marken (2010.04.19.1120) --

RM: A good example is the Bourbon and Powers research described in
their "Models and their worlds" paper (I think it's on the net
somewhere but I couldn't
find it).

BP: Found a pdf file. Attached. I think use of this file or reprints
of it in any publication will require a release from the
International Journal for Human-Computer Studies. Martin T., do you
think I could get a release of the rights from Academic Press?

None of either author's address information is current.

The original article after Bourbon's initial draft was considerably
more critical of existing theories than this final version was. It
was initially rejected on this account. I undertook to tone it down,
but unfortunately I could not get a reply from Tom in time to meet
the publication deadline (this was a special issue of the journal
edited by Martin Taylor). I sent the revised version and it was
accepted, but Tom was considerably upset that I had changed the paper
without his consent. I wish I had not offered it without his explicit
agreement. I don't have Tom's original version, but if anyone does
have it, it would be good to make it available.

Tom and I actually tried to publish this paper in 1992; we submitted
it to "Theory and Psychology" and it was rejected. My answer to the
editor, Stam, after the rejection, is attached, too. As you can see,
I was boiling with anger at the stupid reviewers. It didn't do any
good, as I can now see was inevitable.

Best,

Bill P.

WORLDS2letter.doc (5.49 KB)

Models&Worlds-HC980263.PDF (2.3 MB)

[From Rick Marken (2010.04.19.1600)]

Bruce Gregory (2010.04.19.1525 EDT)--

Rick Marken (2010.04.19.1215)]

Just because I said so? Your approach to science seems remarkably

"faith-based".

BG: Not in the least. If all that PCT claims, if it claims anything, is that
control is the control of perception, then anyone who disagrees with PCT
must have something other than control in mind.

Yes, that turns out to be true, I think. Manual control theorists, who
model behavior as the control of output, have a problem because they
are seeing the phenomenon they are trying to explain (such as manual
tracking) as an S-R (causal) rather than as a control phenomenon. I
actually think the most important "claim" made by PCT is a factual
rather than a theoretical one. PCT claims (a better way of saying it
would be "PCT points out") that there is phenomenon of control. This
phenomenon is seen when a variable is maintained in fixed or variable
states, protected from the effects of variables (disturbances) that
should move it away from these states. The phenomenon of control is
seen in the behavior of living systems as purpose. I think that this
is why Bill titled his most recent book: The Fact of Control.
Behavior:

If you can't perceive the
outcome of your actions, all you can do is flail about aimlessly. Inverse
kinematics is perfectly adequate for flailing.

This is not necessarily true. Inverse kinematic models can produce
consistent results (that is, they can look like they are controlling)
as long as the disturbances present are very small or transient. In
other words, non-control models of control can appear to work as long
as they are tested in an environment (world) in which control is
unnecessary.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[Martin Taylor 2010.04.19.19.03]

[From Bill Powers (2010.04.19.1529 MDT)]

Rick Marken (2010.04.19.1120) --

RM: A good example is the Bourbon and Powers research described in their "Models and their worlds" paper (I think it's on the net somewhere but I couldn't
find it).

BP: Found a pdf file. Attached. I think use of this file or reprints of it in any publication will require a release from the International Journal for Human-Computer Studies. Martin T., do you think I could get a release of the rights from Academic Press?

I have no connection with Academic Press or any more with the journal (I used to know the general editor at the time). As an author, you might be able to get Academic Press to allow you redistribution rights, or permission to send to this mailing list. I really don't know. My own experience with Academic Press is not happy, but they did return to us the copyright of our Psychology of Reading, after having done everything they could to prevent its sale, including explicitly refusing to sell it to at least one teacher who wanted to use it in a course.

Martin

[From Dag Forssell (2010.04.19.1810 PDT)]

[From Bill Powers (2010.04.19.1529 MDT)]

See Closed Loop, 3(1), 47-72. Also reprinted as Chapter 12 in Runkel (2003)

Tom was upset that you cut out many of the key references he had labored hard to put together without asking him. That is what he told me when I put Runkel's book together. Closed Loop and Chapter 12 is complete.

Closed Loop is available at

ftp.pctresources.com
user name pctstudent
password re5earch!

Best, Dag

[From Rick Marken (2010.04.20.0820)]

Bill Powers (2010.04.19.1529 MDT)--

Tom and I actually tried to publish this paper in 1992; we submitted it to
"Theory and Psychology" and it was rejected. My answer to the editor, Stam,
after the rejection, is attached, too. As you can see, I was boiling with
anger at the stupid reviewers. It didn't do any good, as I can now see was
inevitable.

But you were so much older then; you're younger than that now.

I notice that you describe one reviewer's criticisms as "sophistry".
Sounds like that reviewer is someone who is now on CSGNet;-)

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com
www.mindreadings.com

[From Bruce gregory (2010.04.20.1413 EDT)]

[From Rick Marken (2010.04.20.0820)]

I notice that you describe one reviewer’s criticisms as “sophistry”.
Sounds like that reviewer is someone who is now on CSGNet;-)

Don’t be so hard on yourself.

Bruce