Retrofaction, Testing theories, Quality of Control

[From Rick Marken (960129.1100)]

Hans Blom (960129b) --

If the outside world (partly) determines reference levels, it -- by
implication -- controls what the organism controls for. Agreed thus far?

No. The outside world _determines_ reference levels but it doesn't _control_


I think we (PCTers and quasi-PCTers) all agree that the environment
(or our experience thereof) does _not_ control our behavior.


What is the definition of "control" that is both consistent with this
remark and with the fact that in the PCT hierarchy (lower level)
references have values that depend upon (perceptions of) the world?

Bill Powers (960129.0800 MST) already answered this. The definition of
"control" that is consistent with this remark is the one we have been using
all along. Control occurs when, for every disturbance that tends to move
a variable from its current state there is an action that tends strongly to
restore that variable to a particular state. Control systems restore
environmental variables to particular states if they are disturbed; the
environment does not restore reference signals to particular states if they
are disturbed.

I think we have more than a definitional problem here. I don't think anyone
would have trouble understanding why the environment does not _control_
(or _retrofact_) if they had a feeling for the nature of the phenomenon to
which these words (control, retrofact) refer. Anyone who has ever "grabbed" a
variable that is being controlled by control system has literally _felt_ the
difference between a system that controls ("retrofacts") and one that
determines or influences. If the inanimate environment (rocks, pavement,
wind, etc) started responding to my actions the way a control system responds
to disturbances to a controlled variable, I'd start offering up sacrifices of
appeasment faster than you could say "but that's your first born".

Bruce Abbott (960129.1055 EST) --

Creation "scientists" argue that variation and natural selection cannot
explain the origin of species, and prove their case with a little simple

I'm sorry you think that the research and modelling we've done on the nature
of human controlling is comparable to that done by creation scientists. But
that explains a lot.

I have recognized occasions on which certain spokespersons for PCT have made
claims about those opposing views that are not warranted by the facts as I
see them.

We do experiments to show how control works and how controlling differs from
responding to input. It seems to me that the only time we are told that we
have misrepresented an opposing view is when we have clearly demonstrated an
incorrect prediction of that view. For example, we have developed (and you
have presumably done) experiments that show 1) no correlation between input
and output in a control task 2) production of non-random results despite
random consequences of action 3) variations in press rate to compensate for
variations in "schedule" to control a cursor 4) disappearance of "stimulus
control" when a disturbance is added to the response variable.

In every case we have been told that the results of these experiments are not
inconsistent with opposing views (S-R, reinforcement, cognitive); that if we
had not "misrtepresented" those views we would see that the results we found
were exactly what was predicted. Yet we are never shown a non-control model
that can 1) control the cursor by generating outputs based on the inputs
2) produce consistent results in the random consequences experiment 3) vary
response rate to control a cursor or 4) maintain a stimulus-response
relationship when a disturbance is added to response variable.

It seems to me that the only "creation scientists" in your analogy between
PCT and conventional psychology are the conventional psychologsts themselves.
Apparently, their theories (like those of creation science) can account for
_any_ observation that has been or might be obtained. If we do experiments to
reject their theories, and the theories are rejected, then we have
misrepresented the theories. Cute.

If you are really "controlling for promulgation of a view of organisms as
purposive systems and for the development of a science of such systems," you
should be applauding rather that disparaging my efforts.

I'll applaud your efforts when you stop working so hard to bail out models
that are clearly at variance with the facts at hand and start coming up with
suggestions for experiments that will expose the rather substantial
difference between the PCT view of organsisms as purposeful systems and the
conventional view of the organisms as non- purposeful, cause-effect systems.

Martin Taylor (960129 01:40) --

Why do you keep bringing up what we all know you believe (as do many others,
including, I think, Bruce), that the mechanistic basis of Killeen's work is
built on quicksand. That's totally irrelevant to the question of whether it
actually describes the results of experiments.

You guys apparently learned your science at the same monastary.

At my monastary, they taught us that the question of whether a theory
actually describes the results of experiments is the _only thing relevant_ to
our beliefs about the "mechanistic basis of the theory". As long as the
theory predicts the results of experiements, then we accept the basis of the
theory. We only reject a theory if it fails to predict the results of

Bruce seems to think that Killeen's theory predicts the results of adding
non-contingent incentives in an operant task. If this is so, then there is
nothing wrong with the "mechanistic basis" of Killeen's theory.

Based on my understanding of Killeen's theory, I'm surprised to find that
there is apparently no simple observation we can make that will discriminate
it from control theory. If there is no such observation that can be made,
then Killeen's theory and control theory are not fundamentally different
(which is what I'm pretty sure Bruce has always believed); control theory
just might provide a more economical explanation of all the things Killeen's
theory explains (which is what I think Bruce has always wanted from
control theory - - a more elegant explanation of how organisms respond
to incentives). Since other experiments suggest to me that organisms
control incentives -- they don't respond to them -- I feel that Bruce is
(unintentionally) obfuscating the difference between reinforcement theory
(currently represented by Killeen's theory) and PCT. But there's not much I
can do about it; we've got a control system on the loose and he's controlling
for using control theory to explain anything but control;-)

Hans Blom (960129e) --

There is no absolute for the quality of control.

You're right about the measure of control of temperature being dependant
on the scale of measurement; good catch. If the reference, T', and resulting,
T, temperature are 20 and 18 in centigrade and 560 and 550 Kelvin (I made up
the Kelvin equivalents) then obviously the quality of control, measured as
T/T' will seem quite different (.9 in the first case; .98 in the second).
But there are other measures of control that are absolute. One is the
stability measure, S:

S = 1- (V/V')^1/2

where V is actual and V' expected variance of the controlled variable. This
measure of quality of control comes out the same regardless of the scale of
measure of the controlled variable. The problem is that, in order to obtain
this measure, you have to know how all variables (including the output of
the controller) WOULD have influenced the controlled variable if there were
_no_ control system p[resent. We can do this in our tracking experiments; but
its not a great general-purpose measure of control.

I think that, in practice, we are stuck with relative measures of control;
ie. RMS error, ratio of reference to actual values of controlled variables,
etc. But that's OK.