[From Rick Marken (960912.2200)]
Martin Taylor 960912 11:00 --
In fact, when the human controls accurately, almost ANY control >system
that controls accurately will provide a good fit to the >human data. It is
only when the human has problems that different >simulated control systems
will have appreciably different fit values.
Bruce Abbott (960912.1640 EST)]
Martin, now that you mention it, this is one of those "interesting
points" I was going (am going?) to introduce that Simon raises >about
artifacts in _The sciences of the artificial_, the book Rick >doubts has
anything to say of relevance to PCT. Rick, what do you >think: is this
point relevant?
The idea that "ANY control system that controls accurately will provide a
good fit to the human data" seems to be based on the notion that good
control is simple and that poor control is complex; good control can be
simulated with ANY high gain controller (controlling the same perceptual
variable as the human, of course) but poor control really separates the men
from the boys. The output of a good controller (human and model) matches
the disturbance; big deal. The output of a poor controller is all over the
place; it's a good model, indeed, that can explain all the human
controller's apparently random output variations.
One problem with this idea is this: suppose you see a person who is
apprently controlling poorly; the controlled variable is moving all over
the place and outputs are only roughly (negatively) correlated with the
disturbance. The problem is that you have no idea whether the variance in
output (and input) is a result of poor control or good control relative to
a changing reference for the controlled variable. I gave a demonstration of
the equivalence of these two possibilities during my talk in Wales. I
showed a plot of cursor and handle position for a low gain control system
and for a high gain control system controlling relative to a variable
reference (varying as the cursor had varied for the low gain control
system). The plots for the two systems were, of course, identical. There
was no way to distinguish the behavior of the low gain from the high gain
controller.
This demonstration was part of a talk about why manual control theorists --
psychologists in the 50s and 60s who used control theory to model behavior
-- never understood what was actually going on in these control tasks
(control of input); a manual control theorist, thinking that input guides
output, would have seen the apparently poor control in only one way -- as
poor control; the idea that they might have been looking at good control
relative to a changing reference would have never occurred to them.
I suggest that the way to compare control models is not to make behavior
noisy but to create situations where the models make different predictions.
Avery Andrews (960913:1051 Eastern Oz Time)
One of the cases that Sacks describes provides, I think, quite nice
evidence for a certain amount of modelling in the human perceptual
system... She found that she couldn't deal with social situations >orindeed almost any aspect of the outside world because people, >cars, etc.
kept popping up in utterly unexpected places.
Why is this evidence for modeling in the perceptual system? I think Sacks
had it right: the patient lost a perceptual function that computed a
perception of motion or transitions; so she couldn't control perceptions
that depended on the ability to detect change. I would find it very
difficult to control my social distance from people if a person who had
recently been "over there" was suddenly "right here".
Bruce Abbott (960912.2025 EST) --
Is is your hypothesis that, where we disagree on an issue, it
must be because I don't really understand circular causation, or >must not
think it's important?
No. My hypothesis is that you are controlling for the idea that PCT can
exist comfortably in the context of conventional psychology. So far,
everything I have said that would be a disturbance to that idea -- eg.
"Simon has nothing to contribute to our understanding of purposeful
behavior", "psychologists who study the causes of behavior are on the wrong
track" -- has met with strong resistance.
If the "causes of behavior" reside in the organism's purposes, and >one
wishes to identify what those purposes are, then I don't know >of any
other way to do so than to apply disturbances and observe >how the
organism responds to them. Do you?
You bet. Applying disturbance and observing how the organism responds tells
you nothing about an organism's purposes. This is just standard IV-DV
methodology. What is missing -- the essential step -- is applying these
disturbances to a _hypothetical controlled variable_. The important thing
to watch for is _lack_ of effect of the disturbance.
Any disturbance is bound to have an effect on many controlled variables; so
there will be many systems pushing back, to varying degrees, against any
arbitrarily selected disturbance Similarly, any arbitrarily selected
response measure is bound to reflect, to varying degrees, the "pushes" of
the many system's acting to resist the disturbance in different ways. So
there is bound to be at least a statistical relationship between any
arbitrarily selected disturbance and any arbitrarily selected response.
This is what psychologists have been looking for when they search out the
"causes of behavior". Such relationships tell us nothing at all about any
of the organism's purposes.
>I would call that "studying the causes of behavior."
If you used The Test properly then I suppose you could say that you are
looking for the "causes of behavior". But then you are conflating what you
are doing -- looking for the purposes of behavior -- with what
psychologists have been doing for the last 100+ years -- looking for the
external variables that cause behavior via the organism. I think it would
be much clearer if, when you do experiments to determine an organism's
purposes, you call it what it is: studying the purposes of behavior.
I think I am leaping at the word "cause" in lineal and circular >causality
because you said that behavior is not caused -- a >demonstrable fallacy.
Behavior -- controlled perceptual variables -- is not caused: it is
controlled. There's a difference.
Now you're trying to weasle out of it by suggesting that I'm >somehow to
blame for this error.
I'm not trying to weasel out or to "blame" you for anything; I'll say it
again if you like (though I don't recall saying it the first time).
Behavior is _not_ caused. Behavior (controlled perceptions) is controlled.
Best
Rick