Abbott's project

[From Bill Powers (2001.05.24.1406 MDT)]

Bruce Abbott is generating a new summary of PCT and has asked for
suggestions about which experiments and demos to include.

I've been reviewing Rick Marken's "Mind Readings." It contains a number of
important basic demonstrations showing various principles of control
behavior, which are summarized nicely in the Introduction. I realized
during this re-reading, however, that most people will need a much more
detailed explanation of the Figures that show the results of experiments
and demonstrations. Also, the Figures themselves are often cryptic and hard
to understand. It would be helpful to re-do these experiments with modern
computers and better plotting routines, with attention to making
independent disturbances much more independent. And the captions on the
figures ought to contain much more detailed descriptions, to tell the
reader _what to notice_ about the Figures.

The papers in the "Mind Reading" section are of crucial importance to an
understanding of PCT, and especially of the difference between PCT and
conventional theories of behavior. One of the main reasons for rejecting
the idea of intentional or purposive behavior has been that there is no way
to distinguish an intentional effect of a "response" from an accidental
effect. The "Mind Reading" papers focus on the proof that there is a
difference, and on techniques for making the distinction. In my Demo1, the
section called "intentional vs. accidental effects" is another proof, using
the special case in which one of three potential controlled quantities is
held in a known reference state. In Marken's more general demo, the
controller is allowed to move one of the objects in _any_ consistent
pattern, not known to the experimenter. Since these demos contradict all
previous reasons for rejecting purpose or intention, they are crucially
important.

Another extremely important section is "The Causal Circle" (p. 61 ff). The
first article is "The Cause of Control Movements in a Tracking Task." The
essence of the article is shown in Fig. 1, p. 64. Here two runs of a
tracking task using the same disturbance are plotted, showing cursor
position and handle position. In other discussions, several people have
noted that even though there is a very low correlation between cursor and
handle position, there would be a much higher correlation between cursor
position and handle _velocity_, so the "disproof" of S-R explanations is
not convincing. Here, however, we see that the two handle-position plots
are essentially identical, while the two cursor plots are about as
different as they can be. So the supposed integrals of two _different_
stimulus behaviors turn out to generate the _same_ response.

In another experiment, which both Rick and I have done independently, the
cursor position from one tracking run is played back during a second run
(in the second run, the handle position does not affect the cursor).
According to the S-R explanation, this should produce an identical
"response pattern." Rick showed the computed cursor position that would
have occurred if control had actually been possible, given the handle
positions that were observed in the second run and the original disturbance
pattern. The cursor behavior was very different in the second half. I have
done longer runs, and shown that in the no-control mode, the handle
position deviates farther and farther from its behavior during the
closed-loop portion of the experiment. I think I first showed this demo at
Haimowoods. In this sort of demo, it is shown that the _same_ stimulus
leads to _different_ responses, the converse of the case above.

Reading over Rick's book, I became discouragedly aware of how far we have
strayed off the path on which we started. In the early days, we were trying
to think of proofs of principle, with experiments and demos that both
tested and challenged the proposed properties of human control systems in
ways that were as objective as possible. When someone thought of a
difficulty, we devised more demos and experiments that took care of the
supposed exception. I think we were gradually converging on an irrefutable
set of arguments in favor of PCT. So what happened?

I think what happened was that we got involved with a lot of people who
didn't care about irrefutable proofs and only wanted to apply the proposed
theory to real behavior, right away. Applications are the whole point of a
theory, but to start worrying about applications before we had achieved a
broad consensus about the correctness of PCT was a serious mistake. As long
as any substantial number of scientists doubted the validity of PCT, we
could expect to have any proposal shot down, any publications rejected, and
every opposing theory stoutly defended, however feeble its scientific
underpinnings. And that is exactly what we got.

I think we need to get back on that original path. We have to consider
every opposing view, and think of experiments that will allow us to compare
the PCT predictions with predictions based on the other view. Where we are
uncertain of what an opposing view would predict, we have to try to find an
adherent of that view to tell us. Gradually, we must built up that array of
irrefutable proofs -- or if we can't do that, go earn an honest living
doing something else.

Best,

Bill P.

[Martin Taylor 2001.06.02. 16:34]

[From Bill Powers (2001.05.24.1406 MDT)]

As long
as any substantial number of scientists doubted the validity of PCT, we
could expect to have any proposal shot down, any publications rejected, and
every opposing theory stoutly defended, however feeble its scientific
underpinnings. And that is exactly what we got.

I think there's more to it than that. What I tend to get from local
colleagues is not that PCT is wrong, but that it is too
all-encompassing, and needs too many extra assumptions or prescribed
conditions to apply to the situation in which _they_ are interested.
They seem to "accept" the validity of PCT (possibly without
understanding what it is they claim to accept) but not to want to use
it.

I can't even get them to eliminate significance testing from their
arsenal of techniques, which one would think would be a
self-evidently necessary change they could make easily. How, then
should they be expected to make the bigger break from what they were
trained to do, and reinvent their part of science in the light of PCT?

ยทยทยท

----------------------------------

In case you were wondering, I haven't been "lost" as Bill P. put it
one day when I phoned him. I've been away a lot, and busy with a
variety of things, and although I have saved all the last year's CSG
messages, I've dipped into only a few of them.

And also in case you were wondering, the experiment on the effect of
sleep on tracking at different levels of perception was run last year
with 32 subjects, but I encountered computational artifactual
problems in the analysis, which I believe I have run to ground. So
you may get to hear the results some day. I hope that day won't be
too distant.

Don't expect too many new postings from me in the near future--I'm
still busy and expecting to be away a lot. But I'm as concerned with
PCT as ever, even when silent on CSGNet.

Martin