[From Rick Marken (920604 10:00)]
Gary Cziko (920604.0915) says:
I've have been greatly enjoying reading through your collection of papers
in the recently published _Mind Readings_ book.
I love you.
One study that I find especially intriguing is "The Cause of Control
Movements in a Tracking Task." In this experiment you show that you can
get subjects to to respond very similarly (r > .99) in the task by having
them do it again with the same disturbance pattern, but that variations in
the position of the cursor between the two runs are not similar (r usually
< .20). You conclude that "This result seems to rule out stimulus
variations as the cause of responses which control (stabilize) the cursor."
I wanted to call the paper "The cause of control in a tracking task" but
they didn't understand what that meant; so the title is really wrong.
The paper is not about how movements (of the arm or handle) are controlled;
it's about how control (stabilization against disturbances) occurs. This
was beyond the control theory experts who reviewed for the journal. But,
at least they let it get published -- with that stupid change in the title.
This is indeed a very ingenious demonstration. But even after about three
years now of studying PCT, what you've demonstrated still looks a bit like
magic to me.
It is -- just like the path of light being bent by a mass. But once
you have the right model it all makes sense.
My brain keep saying to me (or at least part of it): "Surely
there must be SOME aspect of the cursor which determines the response.
It's the part of your brain that can't think in circles (though you can
think in circles around me). There IS, indeed, some aspect of the
cursor that determines the response -- it is the position of the
cursor. But AT THE SAME TIME the position of the cursor is being determined
by the response. The significance of this fact is hard to "think through"
using our usual "lineal" approach to thinking -- a causes b causes c causes...
You just have to trust the math on this one.
If
it isn't the simultaneous position of the cursor, then perhaps it is the
cursor's position some milliseconds before or the speed or acceleration of
the cursor or SOMETHING which determines response."
That's why I did the experiment -- ALL these possibilities are ruled out
by the lack of correlation between cursor traces. If, for example, the
ACTUAL cause of the response is the value of the cursor 100 msec before
the current display then this would be true during both runs since the
response is EXACTLY the same both times. So the cursor traces would be the
same (high correlation) on both runs -- but they are not. The fact that
the cursor traces are different means that NO aspect of the cursor trace can
account for the identity of the responses on both runs. This just falls
out of the feedback equations; remember, output depends on disturbance,
NOT sensory (controlled) input. This demonstartion just shows that what
the equations say is true; the equations imply a result that is just as
magical as the results in this paper. Bill Powers discovered this fact
about control and showed that it really happens; I got to show that it
really happens in another way -- idential responses produced by different
inputs.
And if what you say is true, how on earth do all the human factors types
continue to see the human operator as a transfer function between stimulus
and response?
Why, for that matter, has psychology and the life sciences in general
continued to see organisms as transfer functions between inputs (or
"commands" or "plans") and outputs while completely ignoring the fact
that control systems DON'T WORK THIS WAY? I think you know the answer;
did YOU really want to understand PCT at first? The life sciences are made
out of people with careers, reputations, etc to protect and families to
support. It seems quite understandable to me that they would be reluctant
to find out that there is a fundemental flaw in the idea that supports their
career, reputations, etc; a flaw that says "everything you have been saying
about behavior is completely wrong". Why would ANYONE want to go to the
troble to find THAT out? Just nut cases like you and me.
What type of function do they find? Indeed how can they
find any function at all if in fact there just ain't none, as your research
suggests?
Their models work because stimulus response models work, with the proper
time damping and gain, in a closed loop. We have already gone over this
in considerable grisly detail. Basically, they build models that say
o = f(i) output is function of input
and ignore the fact that
i = g(o) as well,
though they have to hook things up properly so that this second equation
holds true.
This "conventional approach" to control ignores the fact that the reference
level of sensory input is determined by the organism (a VERY important
omission; in fact, just about the whole ball game).
But you might wonder why the conventional control theorists never noticed
the fact demonstrated in my experiment (no apparent relationship between
controlled input and response). Two reasons: 1) they never looked and
2) if they had looked, they would have probably found far more of
a relationship than I found because they usually do their tracking
experiments with the subject operating near the limits of control; they
use high frequency, high amplitude disturbances so control is POOR. When
this is true, more of the variance in the cursor (controlled variable)
is caused by the disturbance. So, to the extent that there is any
control (and since responses correlate with disturbances, not with the
controlled variable) there will be a correlation between response and input.
The results reported in my paper depend on the subject being IN CONTROL of
the cursor; most of the disturbance-caused variance in the cursor is
removed by the repsonses of the subject.
I can assure you, Gary, that there is no trick involved in the experiment
you read about; you can go out and demo this to yourself anytime. The
magic, I'm afraid, is real. The result can only be understood in terms
of the simultaneous equations that describe a negative feedback relationship
between an organism and its environment -- the situation that exists for
all organisms, all the time. You can't understand it in terms of lineal
cause and effect because control does not work that way. The results are
only magic from the lineal cause effect point of view -- they are a
yawner from the PCT point of view. But if one has an investment in the
linear cause effect point of view then these results will evoke responses
of disbelief or disinterest; and these have, indeed, been the most common
responses to this study.
I think that you DO understand the results, Gary. They
are only surprising when you look at them from your old point of view --
and the one that is easier for everyone to use, myself most emphatically
included -- the point of view of the lineal cause effect model of behavior.
Best regards
Rick
···
**************************************************************
Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
E-mail: marken@aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)