[From Rick Marken (941218.1800)]
Bruce Abbott (941218.1650 EST)
Me:
The failure of Bruce's selection by consequences model under this
"fairly simple change of conditions" is very important, not only
because the control model did not fail under these conditions but also
becuase people did not fail under these conditions either.
Bruce:
I provided a fairly detailed reply as to WHY this happened, which was
ignored. Basically it came down to two points: (a) people use other
variables to assess the consequences of their actions than the limited
one provided to my little learning e. coli
This is just conjecture and there is reason to believe that it is not
true. Bill and I explained that people can control the spot in the E. coli
demo when given only the information available to your (and my)
model.
and (b) Rick's model performs well in this situation because it is not a
learning model.
This is true if, by "learning", you mean the change in control
parameters that results from the consequences (rewards and punishments)
of bar pressing (tunbles). But this is precisely my point! Your model
doesn't work BECAUSE these parameters are selected by the consequences
of actions.
This whole "learning" thing just obscures the PRINCIPLE that is being
illustrated in this demo. The principle is that "selection by
consequences" eliminates control. Your model "controlled" in only
one circumstance: when the consequence of a tumble is more likely to
be a reward when going away from the target and a punishment when
going toward it. Your "learning" system destroys control in all other
circumstances -- control that the control model and people retain.
You say:
You may recall that what you were asserting at the time was that a
"reinforcement" based model COULD NOT control in this situation. I
demonstrated that one could, and that is all.
But this is like saying that a pendulum, which returns to the same state
after being displaced (by a "push" disturbance), is a "Newton's law"
model of control. Your model (like the pendulum) only controls (or, in
the case of the pendulum, appears to control) under special environmetnal
circumstances (and it does so because it IS actually structured as a
control model). The "reinforcement" aspect of the model DEPRIVES your
model of the ability to control in all but one special environmental
circumstance.
If you apply only transient disturbances to the pendulum, the bob always
returns to plumb and it looks like the pendulum is "controlling". If,
however, you apply a constant disturbance, the bob doesn't return to plumb
and it is clear that the pendulum is not controlling. Similarly, if you use
only the "regular" feedback function in the E. coli situation, it looks
like (and it's true) that the reinforcement model is controlling. If,
however, you change the feedback function, you see that the "reinforcement"
model is not controlling any longer: the spot does not move to the target.
Me:
The result of this set of demos should have been the realization that
selection by consequences is incompatible with the control exhibited
by people in the E. coli demo.
Bruce:
I fully understand--and agree with--this. (Why you don't think I do is
an endless source of mystery to me.)
Because you keep saying things that confure me, like "a reinforcement
model can control". Reinforcements are consequences that presumably
select the actions that produced them. Your model shows that if
consequences actually worked this way, they would PREVENT control under
most circumstances.
But let's talk about learning. How do your human participants learn
to react in such a way as to bring e. coli under control?
I have no idea.
How about producing an actual MODEL of this, rather than a verbal
description of how such a model might work?
Well, I don't have any learning data so I don't know what to model.
Your current e. coli model does not learn, and thus does not embody
the full capability of your humans participants to adapt.
See my previous post from today. It seems premature to start studying
learning before 1) we thoroughly understand the controlling done in a
particular situation and 2) we have some learning data that needs to be
explained.
Also, from your own description, it took you quite a while to discover
a set of conditions under which my learning e. coli would fail. How
about listing all those variant contingencies you tried in which the
model stubbornly refused to fail?
The only thing that took a long time was for my dufus brain to figure
out a simple way to do what I knew must be done -- make "reward"
and "punishment" (as you computed them) equally likely after a press
when the spot is moving toward or away from the target.
I was afraid you still felt this way. Rick, I accepted feedback-regulated
control as a general principle of behavior (and rejected the
reinforcement model) before I ever heard of Bill Powers and his book.
This isn't a religion. I don't care if you "accept" it or not. I care if you
UNDERSTAND it. I may be a little (a lot;-)) high gain on this but I've
just run into too many people who accept the hell out of PCT and then
proceed to theorize and do research as though they had never heard of it
at all. I'm so paranoid about this that even a seemingly harmless
statement like "I accepted feedback-regulated control..." gets me all
nervous. I can't think of any sense in which control could be
described as "feedback regulated"; if anything, control REGULATES
FEEDBACK, where feedback is the perceptual consequence of the
controller's own output. Conventional psychologists think of feedback
as something that the environment "provides"; so a conventional
psychologist would certainly agree that "feedback-regulated control" is
a general principle of behavior. But that is precisely the misconception
that PCT aimes to remedy. The general PCT principle of behavior is
regulation (control) of feedback (perception).
other related sources on feedback-regulated control had prepared me
to recognize the value of Bill's contributions as soon as I encountered
them.
The problem with conventional treatments of control theory is that
they DO make it seem like the controller is "regulated" by feedback
(self-produced input). The genius of Powers was to realize that this is
precisely backwards; the controller regulates (controls) its feedback
input relative to its own, internally specified references.
If PCT can be shown to handle a phenomenon (e.g., the empirical law
of effect), great, I've learned something.
But what if it's not a phenomenon? That's the problem for PCT.
Because psychologists have worked from the beginning with no
understanding of the nature of behavior as control, they have described
many "facts" of behavior that are not facts at all.
I also have in mind the ultimate acceptance of PCT within my field of
research. The failure of PCT to take the field by storm is due, in my
opinion, to more than simple reluctance to adopt a new way of
thinking about behavior.
Ah, I admire your idealism!
To gain acceptance, PCT must demonstrate an unequivocal ability to
account for the known phenomena of the field, and to do so in a
more natural and parsimonious way.
I suggest that you look at what happened to Bill Powers' and Greg
Williams' paper on alternatives to "motor control" theories of
behavior before you get your hopes up. But I do think there is some
chance that if YOU do the research and the modelling, there might be a
chance (see, I haven't lost my optimism completely).
Rick, you said that you wish you could find just ONE more research
psychologist who, like Tom Bourbon, would take up research from a
PCT perspective. You've got that guy on the end of your fishing line,
and you don't even know it.
I'm a lousy fisherman (as Bill Powers will be the first to point out). I am
really thrilled that you plan to do the PCT research on operant behavior and
I'm 1000% on your side. I'm just real high gain on "PCT correctness"; to me,
having more Carver/Scheier type PCT friends (that is, people who are "for"
what they call PCT but are really just using PCT words to describe
S-R/Cognitive psychology) is worse than having enemies.
I don't assume that you are a Carver/Scheier type; but I'm going to do my best
to make sure that you don't end up like them.
Now, off to re-bait my hook;-)
Best
Rick