PCT vs Conventional Research

[From Rick Marken (941013.1000)]

Martin said:

More generally, Rick often pleads that the only legitimate aim of PCT
research is to find "the controlled variable." If one is to believe
the premises of PCT, this search is doomed to fail, because which
variables are controlled, and the "insistence" on each (to use a word
introduced to the discussion long ago as a generalization of "gain")
changes from moment to moment. What is a controlled perception one
moment is not one the next.

I replied:

Well, this is an interesting revelation. You (along with Oded and
Hans) apparently think that PCT research is impossible.

And Martin (941012 15:40) flabbers back, aghast:

Did you get THAT out of my argument? Well, well. I am not surprised--

In case you lost it, the argument was twofold:

[More flabber deleted]

I'm sure the point of your remarks, Martin, is well beyond my powers of
understanding. I can only note that the understanding of your statement
that was gleaned by other Powers seems to have been the same as mine.
Specifically, Bill Powers (941008.1650 MDT) says:

That sounds like a wonderful excuse for not doing any research at all.

CHUCK TUCKER (941011) --

I believe there are more studies using THE TEST than we realize. I think
the Milgram studies (including his "crowd" studies), the studies by Sherif
(including the autokinetic one), a study by McPhail where the students
"walked- out" of his class in the middle of a lecture and he continued to
lecture to two student sitting in the front of the room, and the Asch study
(with derivations) all used THE TEST

If all these people are using The Test, then all conventional behavioral
researchers are using The Test. Unfortunately, The Test involves a tad more
than just applying "disturbances". The first, and I think the most
important, step in doing The Test is to identify the hypothetical controlled
variable; the variable that you think a person is controlling. This is the
step that is skipped in all your examples. The next step is to say what
would happen to that variable as a result of applying the disturbance if
it were NOT under control. The next step is to describe what actually happens
to the variable during disturbance. Since your first guess about the identity
of the controlled variable is not likely to be right, you have to be
willing to iterate this process until you get it right.

Bill Powers' model of classical conditioning (just the UR part is enough --
then its a model of a "reflex") shows why just applying a "disturbance" and
monitoring some noticable response to that disturbance is not the way to do
The Test. Psychologists have known about all kinds of "reflexes" for years;
they know that certain "stimuli" lead to certain noticable "responses". So
you could say that psychologists been doing The Test for years when they
apply a "stimulus" (disturbance in PCT lingo), like food in the mouth, and
observe a response, like "salivation"? But I don't think there is any
Testing going on here at all. The problem is that psychologists have not seen
a reflex as a control process so they haven't tried to identify the
controlled variables.

A very likely candidate for a controlled variable in the "salivary reflex",
for exmaple, is viscosity of the food in the mouth. This is tested, not just
by applying a disturbance, but also by monitoring the effect of the
disturbance on the hypothsized controlled variable. There are no tests of
this aspect of reflexes in the conventional literature -- that I know of,
anyway; no "Tests for the Controlled Variable" (you can't do the Test by
accident). All we get from conventional studies of reflexes are fairly noisy
relationships between distrubance and response; the same disturbance (such as
food in the mouth) does not always produce the same response (level of
salivation). In fact, the variations in the "response" to disturbance in a
simple reflex can be rather substantial. The reason for these variations
should be obvious from Bill's PCT model of the reflex (UR); the organism's
"response" is just one influence on the controlled variable; other influences
include your disturbance as well as other "natural" disturbances, such as
the amount of saliva or water already in the mouth. If you just look at the
relationship between your disturbance and a particular response measure, you
will see a lot of puzzling variation (welcome to the world of conventional
behavioral science). If, however, you monitor the variable that is actually
under control, you will see great reliabiliy (as we do in the various
versions of our tracking tasks, where we know which variables under control -
- by doing versions of The Test).

Conventional behavioral science research is not designed to study organisms as
though they were living control systems; in particular, it is not aimed at
discovering controlled variables or monitoring the ways in which the
reference states of these variables are varied in order to control other
variables. But this doesn't mean that conventional behavioral science won't
find SOMETHING when it is applied to the study of what turn out to be living
control systems. All the research results that Bruce Abbott mentioned,
although often statistical, are actually found when we use conventional
research methods to study living control systems. We do find serial position
curves, conditioned responses, schedule effects in operant conditioning,
effects of "experimenter pressure" on compliance (Milgrim), etc. But all we
are finding is that various measures of the "response" of a complex
hierarchical control system (measures that we call "behavior") will vary
in sometimes rather systematic ways when one "pushes" on that system. But
just because we have "results" from conventional research doesn't mean that
these results tell us anything about the nature of living control systems.

Those of you who continue to harbor some faith in the worth of conventional
behavioral science should get a hold of a copy the Lotus or Excel version of
my spreadsheet model of a hierarchical control system. This is the PCT model
of a living systems, operating in a VERY simple environment but controlling
many perceptual aspects of this environment simultaneously. It is not a model
of a real organism but it is a demonstration, in principle, of how a
hierarchical control system works, dynamically. You can do all kinds of
things to this model -- the kinds of things that conventional behavioral
researchers do to organisms in conventional research. You do these things by
changing the states of independent variables (disturbances) in the model's
environment and measuring asoects of the behavior (outputs) of the model. If
you plot out the results of this kind of analysis (much like the IV-DV
research in conventional psychology, but here done on one --simulated -
- organism at a time) you will find "results"; and you could name these
results in a way that makes them seem interesting. The model, for example,
might respond to sudden changes in two of the six independent environmental
variables rather strongly; these variables could be called the "aggression
triggers" and the response the "agression response". These are great names,
but in this case you know that you have obtained these "aggresion" results
from a hierarchical control model that has no "agression" detector or
"aggression response" systems. The model is simply controlling intensities,
linear combinations of intensities and relatinships between these linear
combinations. The model shows how the results of conventional research could
be simply one's own interpretation of variations in variables that one
happens to consider "interesting". Such results, however, reveal nothing
coherent about what an organism is really doing (controlling various types of
hierarchically related perceptual variables) if the organisms is a living
control system.

The closer conventional psychology comes to looking at the variables that are
actually involved in control -- and doing so with individual control systems -
- the less noisy the data and the closer the results come to revealing
something about the nature of control. This is why many studies of operant
behavior can be the start of good studies of living control systems. In
operant studies, an organism is given control of some variable (such as food
delivery) and it is clear that the organism can control the variable because
it maintains the variable in a reference state despite changes
in (disturbances to) the "schedule" that is actually the feedback function
relating outputs to controlled input. The problem with operant studies (from
the point of PCT) is that the people who do them refuse to abandon their
comittment to the notion that the environment "controls" organsms; so these
researchers don't go on to use The Test to determine the aspects of the
environment that are actually being controlled by the organism. This could
be done easily in operant studies by determing whether a variable like food
delivery state is maintained against disturbances like the occasional
insertion or deletion of food pellets.

If you want to do PCT, you have to start doing PCT; and that means, you have
to stop doing that "old time behavioral science". If organisms are control
systems (and I think there is ample evidence that they are) then old time
behavioral science just gives us some misguided, side-ways glances at how
organisms actually work.

For those who like astronomical analogies, I would say that using conventinal
behavioral science data as a basis for testing PCT is like using the
ancients' description of the constellations as a basis for testing Newton's
laws of planetary motion.