The phenomenon of control

[From Rick Marken (950114.1600)]

Chuck Tucker and Martin Taylor have recently posted data that seemed
(to them) to pose "problems" for the basic PCT model of behavior. As a
big fan of research on living control systems, I applaud every effort
to experimentally test PCT; nothing could be more important to the
development of a science of living control systems. But I think it
is important for researchers to remember what PCT is about -- the
phenomenon of control. PCT does not explain everything -- it doesn't
even explain everything what is called "behavior". PCT explains the
fact that certain consequences of an organism's actions are maintained
in a constant of varying reference state: it explains control. If there
is no control, then there is nothing to explain. The fact that a subject
in an experiment is moving a handle and producing (as a result) effects
on a computer screen does not mean PCT can predict the relationship
between handle movements and screen effects. It can only do this if
the subject is controlling the relevant variables.

The apparent "problems" for PCT in Chuck and Martin's data are only
problems if we know (for a fact) that subjects are skillfully controlling
a particular variable. If the subject's are unquestionably controlling
a variable and they are known to be able to control it well (skillfully)
then, if this control process cannot be duplicated by the PCT model, the
PCT model has problems. Neither Chuck nor Martin presented data on how well
subjects were able to control (under the best circumstances) so we really
have no idea whether a skillful PCT model controlling those particular
variables is even appropriate.

For those about to embark on PCT research, I reiterate my yearly cry:
phenomena first! Make sure that you are dealing with the phenomenon
control before you start trying to mimic it with control models. There
is, for example, a relationship between my daily cries for sunshine and
the daily amount of rainfall in LA LA Land. You could try to model my
"control of sunshine" with a PCT model but you would do very poorly
because, obviously, I'm not controlling it.

I think we PCTers make a strategic error when we emphasize the THEORY of
perceptual control on this net. The basic theory of perceptual control is
EXTREMELY simple. What complexity exists is in the perceptual functions
(and we rarely use perceptual functions that are any more complex than
constants) or in hierarchical relationships between control systems (and
we rarely study hierarchical phenomena). The simplicity of PC THEORY can
lead theory freaks to try to complexitize it -- just to make things more
interesting. To a theory freak, the measure of a good theory is how many
people, other than the theorist, can understand it; the more people there
are who understand it, the worse the theory. People who are into PCT for
the theory -- theory freaks -- are often disappointed by its simplicity.

What I think we should always emphasize is that PCT is about a phenomenon.
The important (and astounding) aspect of PCT is that it points to the existence
of a phenomenon that has been going on in front of us forever, that is
the most pervasive aspect of the behavior of living organisms; a phenomenon
that has been completely MISSED. PCT shows that organisms control -- every-
where, all the time -- and that this fact has been overlooked because control
can LOOK LIKE S-R, control by contingency, planned output, etc. PCT
is about a phenomenon that 99.9% of those in the behavioral and social
sciences are completely unaware of. PCT shows how to bring that phenomenon
into the lab so that we can look at it using the mircoscope of control theory.

Theory is great; but phenomena come FIRST.

Best

Rick

[Martin Taylor 950116 12:10]

Rick Marken (950114.1600)

It's becoming a tedious refrain: "Rick, where did you get the idea I
said anything like that?" Tedious, and the more tedious because of the
frequency with which it becomes necessary. Today's case in point...

Chuck Tucker and Martin Taylor have recently posted data that seemed
(to them) to pose "problems" for the basic PCT model of behavior.

I posted real data on classical tracking studies programmed by Bill Powers.
I showed the results of an analysis using a model consisting of unit
functions all the way around the loop except for a pure integrator with
delay as the output function. I commented that these results did not
provide the 95% plus correlation with the data that is often cited as a
requirement for believing in the validity of a model. I asked the "old
hands," specifically mentioning Rick, for suggestions as to what
modification to the model would, in their practiced opinion, be most
profitable to explore so as to find a model that would fit all or most of
the "non-sleepy" data. I suggested the possibility that Bill P had earlier
said he was exploring--a kind of nonlinearity that emulated a dead-zone.
(But I haven't heard from Bill as to his success in that exploration.)

Where in all of this is there the slightest suggestion that I think the
data pose problems for the basic PCT model of behaviour? Does the basic
PCT model of behaviour insist that EVERY control system has only a unitary
perceptual function, an output function that is a pure linear integrator,
and a fixed, uniform loop transport delay? It is THAT model for which the
data DO pose problems.

If I thought that the data posed problems for PCT as such, would I bother
asking the "old hands" for the benefit of their experience in successful
modelling?

I have no idea why Rick thinks that a past president of CSG and I both
have problems with the basic model of PCT.

···

=====================

The apparent "problems" for PCT in Chuck and Martin's data are only
problems if we know (for a fact) that subjects are skillfully controlling
a particular variable. If the subject's are unquestionably controlling
a variable and they are known to be able to control it well (skillfully)
then, if this control process cannot be duplicated by the PCT model, the
PCT model has problems. Neither Chuck nor Martin presented data on how well
subjects were able to control (under the best circumstances) so we really
have no idea whether a skillful PCT model controlling those particular
variables is even appropriate.

It has often and forcefully been stated that the PCT modelling works not
only when the subjects control well, but also when they control poorly.
The model, or at least a correct model, controls poorly under the same
conditions, but it at least matches the subject's data.

It has seemed to me also that once we accept that the correct model of
all behaviour is the control system, then it is incumbent on us to find
out what kind of control system(s) is(are) acting in any particular situation.
Since any effective control system will correlate extremely highly with
any other equally effective control system, the key statistic should not
be the correlation as such, but the relation between the rms error (or
some such statistic) of the real control and the rms error of the difference
between model and real. Correlation is guaranteed to approach 1.00 between
two equally effective control systems acting against the same disturbance.
The ratio of prediction error to tracking error is not guaranteed to approach
any particular value, but if the model is good, it should be near zero
no matter how good the tracking is.

The fact that a subject
in an experiment is moving a handle and producing (as a result) effects
on a computer screen does not mean PCT can predict the relationship
between handle movements and screen effects. It can only do this if
the subject is controlling the relevant variables.

Yes. This is why I initially mentioned data only from those sessions in
which there was good reason to believe that most of the subjects were trying
to do what they were asked to do. Later, when they were sleep deprived,
it was clear that at least some of the time they were not controlling the
kind of variable we were interested in (if they were even moving). But to
see what might have been happening then, we need (or at least I need) some
assurance that the model being applied to the (should be) good data is
a valid model. And I DON'T mean "that the basic PCT model is valid".

Martin

[From Dag Forssell (920903-1)]

Bill Powers (920903.0800) in response to Penni Sibun (920902.1400)

This is why we don't start with the math, but with the phenomenon of
control. Once you have experienced and understood control as something
happening, the PCT explanation becomes the only feasible one; it's clear
that no other explanation can even come close to handling what you see
happening. It would, in fact, take an enormous act of faith to believe
that the control-theoretic explanation is wrong.

Penni, if you give me your snail mail address, I will be pleased to send
you my information packet which includes a do-it-yourself demonstration
of the phenomenon of control. The demo includes a narrative adapted from
Bill's work and several illustrations. You perform the demo with another
person (preferred) or by just reading and visualizing.

By getting some personal, physical experience with control and what it
looks like in action, you may gain a fresh perspective on the discussions
here.

Dag Forssell
23903 Via Flamenco
Valencia, Ca 91355-2808
Phone (805) 254-1195 Fax (805) 254-7956
Internet: 0004742580@MCIMAIL.COM

Dag- could you send me the info & demo too? I need all the
          insight I can get.
          Thomas B. Baines
          ARGONNE NATIONAL LABORATORY
          EID900
          9700 South Cass Avenue
          Argonne, IL 60439

Dag,
You probably saw my 'intro' on the net a few days back. Anything along the
lines of PCT in industry that you could throw my way would be much
appreciated. By the way, I found your performance review proposal informative
and helpful.

I've gotten much material from Ed Ford and Tom Hancock -- in fact, Tom gave me
copies of your 2/4/90 & 2/11/90 drafts concerning Deming and control systems.
I'm really struggling to make PCT a real mindset to some people (including
myself), and eventually, to an entire organization here.

I would appreciate it if you could put me on your 'snail mail' list; please
let me know what postage and handling fees would be appropriate.

Regards,
Ray

···

*******************************************************************
Ray L. Jackson
3613 W. Saragosa St.
Chandler, Az 85226
attmail.com!rljackson
602-963-6474