[From Bill Powers (2000.09.25.0823 MDT)]
Rick Marken (2000.09.24.1740)--
When they called some things stimuli and other things responses,
did they explain what the underlying model was? And did they try
to _test_ that model in any way, to see if it was correct?
Of course not. But we can! We know that seeing behavior in S-R
and PCT terms implies a particular underlying mechanism (model)
that has particular properties. And we know how to test to
determine whether we are really looking at an S-R or PCT-based
phenomenon.
I think the Test is specifically designed to see whether behavior fits the
definition of a control process. But perhaps there could be some similar
Test that would reveal whether behavior fits the definition of an S-R
process. Could it be possible that the same behavior might pass _both_
kinds of tests? If that were true, then which way to interpret behavior
would be optional, and we would have to search for some new way to choose
between these views (unless someone could show that they are completely
compatible).
We have developed a formal model of a control process, a model that seems
fairly complete in concept. What we need is an equally-well-developed model
of an S-R process, so we can see what it would imply that we could test
experimentally. This is one reason I'm not urgently interested in teaching
people how to interpret the world in PCT terms. That's for believers, and I
don't want to help produce a generation of blind believers in PCT. What I
want is an honest comparison of PCT with, in this case, S-R theory, so we
can apply normal scientific procedures to choosing between them. By that I
mean procedures that do not involve persuasion, emotional pressure, appeals
to authority, or blind faith. Just "if this is true, then that has to work
that way." That is the only basis for choosing between theories that has
any chance of convincing the scientific community in the long run.
My teachers would never have thought of _testing_
the S-R view itself; they thought (and all current experimental
psychologists still think) that S-R is just the way behavior _is_.
Then that's enough reason to throw down a challenge, or an invitation, and
open up a debate in which the questions can be settled by scientific means.
One of the people who can help us greatly with this is Bruce Abbott. I
don't see eye-to-eye with Bruce on many matters, but I respect his
knowledge of behaviorism and his ability to pose tough questions.
This is not a simple matter we're talking about. If it were, somebody just
as smart as we are would have settled it long ago. There are terrible
ambiguities between interpretations and experimental designs: if an
analysis proves false, was the analysis wrong or was the data, in some
sense, wrong? Bruce and I have just such a situation going on now: did the
control model fail to explain the behavior of the rats, or was the
experiment so full of changes in conditions that the very basis for the
behavior changed during the experiment? Are rats, in fact, capable of
varying their rate of responding in the manner assumed in the original
control model? The evidence now seems to be that they cannot do so, but
perhaps in some other kind of experiment they can. As it happens, we had to
give up on the "rate of responding" as the output variable, and substituted
the "rate of ingestion" which was variable and could be directly measured,
more or less (by subtracting food that fell through the cracks, literally,
from food delivered to the rats). Water intake and excretion were measured,
also. That left oxygen intake and water vapor and other gas exhalations
unmeasured, but still allowed modeling a reasonable control relation
between food ingestion rate and weight.
In my opinion, however, this experiment was far too complicated to help us
make any basic decisions about S-R and PCT. The more we investigated, the
more variables there were to keep track of, and the more loopholes in any
explanation. The forest was definitely not a forest: it was full of trees.
Bruce has suggested that maybe the lower levels of behavior are organized
as control systems, but the higher are organized as S-R systems. This idea
can't be dismissed summarily just because we have a proposed hierarchy that
is control systems all the way up. Why couldn't there be high-level (for
example, symbol-handling) systems that are S-R systems? Tom Bourbon, in
fact, wrote a paper the title of which escapes me at the moment showing how
a control hierarchy could actually be set up to behave like an S-R system
(as well as a cognitive command-driven system, which is essentially the
same thing).
The question, of course, is not why high-level systems couldn't be set up
as S-R systems, but whether in fact they _are_ set up that way rather than
as control systems. And this brings us back to the need for a formal
statement of an S-R model, including a statement about how it would
probably work.
There is actually one such statement: it's known in some quarters as Modern
Control Theory. This is the model in which a goal of producing some
specific effect outside the system is met by using a detailed model of the
system's own organization and that of the outside world to compute the
inverse kinematics and dynamics of the whole system, so that the required
command signals can be calculated and then issued. The inverse kinematic
and dynamic calculations are the answer to the question of _how_ the S-R
system must be organized to work: I agree completely that this is the
correct answer. And this has always been my basis for rejecting the S-R
model. I claim that the required "endo-model" as Cliff Joslyn calls it
can't be constructed with anything like the required accuracy or completeness.
In the present debate, however, this is only a theoretical objection
without, as yet, any recourse to experimental data. The question is not
whether the inverse calculations could in principle be carried out, but
whether the brain does in fact carry out any such calculations. If it
doesn't, that settles the matter, and the plausibility of the theory is
irrelevant. The same goes for the control model: if the brain doesn't
contain the equivalent of input functions, output functions, and
comparators, and all the rest, then the plausibility or explanatory power
of the control model is irrelevant.
At present we are in good shape with regard to the lower levels of control,
say the first two or three. I think there is ample evidence down to the
level of wiring that there are input functions, comparators, and output
functions connected just as we require for the HPCT model. In fact, the
idea for the HPCT model came largely from learning about neural connections
that were known to exist (30 or more years ago) in the spinal cord,
brainstem, and midbrain. By the same token, the S-R model is in very bad
shape on the basis of the neurological evidence, at these lowest levels of
organization. There is no "stimulus" at these levels that is not determined
as much by the system's ongoing "responses" as by external events.
At the higher levels, neurological evidence is lacking. Sad but true. So
Bruce's proposal can't be conveniently ruled out on that basis. If there is
to be any basis at all for deciding between the models, it has to be
evidence from behavioral experiments.
The two models are sufficiently different that it would seem there must be
some critical experiment that would relieve us of the responsibility of
deciding between them. But look at the case of the driver of a
turbine-engine racing car that Bruce A. brought up recently. The driver
learns to mash down on the accelerator very early in exiting from a turn,
so the power will be there by the time the turn is ended. Is the mashing
down a response to cues encountered during the turn, or is it control of a
relationship between perceptions of pressing the accelerator and
perceptions of being at a certain place in a curve? Either side can come up
with a plausible explanation. The data do NOT decide between the theories.
As matters stand right now, if you like S-R theory there is no reason not
to use it in this example. Ditto for control theory.
But if S-R theory is right, the brain has one organization, while if PCT is
right, the brain has a very different organization. It's simply not
possible for both theories to be right, as far as current knowledge goes.
Therefore it must be possible to devise a way to determine which theory is
right. Whether this can be determined by behavioral experiments is, of
course, the most relevant question. Maybe only investigations of brain
architecture will do the trick, but that would be disappointing to me.
We'll have a long wait before that happens.
I would really like to see a lot of people thinking about these issues and
trying to resolve them without any prior committment to either side. The
attempt to find a scientific resolution is, to me, the only interesting
issue here. If we defend PCT while other people defend S-R or cognitive
theories, we end up with attackers and defenders and a lot of wasted
effort. I want to get onto higher ground, where the pursuit of truth is
what matters. I know, I know for certain, that there are lots of people who
agree with me. So why not just do it? And why have commercials taken all
the best lines?
Best,
Bill P.