Everyone's a theorist

[From Rick Marken (960110.2200)]

Bill Powers (960110.1200 MST) --

My point was that identifying _any_ variable in a closed loop as "the
cause" is arbitrary, and wrong.

A fact which Bruce can demonstrate to himself by replicating the
"Cause of control movements" experiment described in chapter 3
of "Mind Readings".

Bruce Abbott:

There is a simple test that will distinguish which is correct: apply a
load to the crankshaft.

Bill:

The PCT model would fail in this case...This is not a control system.

And we are back to the fact that PCT is a model of _control_; a control
model is applied when we know that control is happening; that is, we
know that a variable is under control and that it is the organism that
is doing this. Reinforcement theory -- including Killeen's version
thereof -- is not a model of control because reinforcement theorists
don't know what control is; therefore, they have not done any research
aimed at determining whether organisms control and, if so, what they
control.

Bruce:

The difference between these views (control and reinforcement] has
little to do with what is actually producing the power to turn the
crankshaft. The difference is in the nature of the systems, whether
equilibrium or control.

Bill:

It is not a difference in views that makes the difference: it is the
organization of the system. A control-system model would be rejected
by the Test...no control system.

And we are back to phenomena again. If there is no controlling going
on (as revealed by The Test) then there is no need to consider control
theory as a model of the observed behavior. Since reinforcement
theorists have never done The Test to determine whether organisms
control, they have never seen the need to develop a control model.
The methods used in EAB (the research arm of reinforcement theory)
are designed to determine how organisms respond to causa inputs; so
models of the "behavior" observed in these experiments are attempts
to explain how organisms respond to inputs. Since organisms behaving
in a loop don't respond to inputs (inputs respond to the organism
_while_ the organism responds to the same inputs) reinforcement
models are irrelevant to the behavior of organisms.

It all comes back to the fact that the phenomenon of control has simply
not been studied by conventional psychologists; reinforcement
theorists don't even know that organisms (besides themselves) control.
This is why the first order of business for students of living control
systems should be to determine whether organisms in operant situations
are controlling and, if so, what they are controlling.

But nobody seems to want to do experiments anymore. Everyone's a
theorist. Doesn't anyone care anymore about what they're developing
a theory _of_?

Bill:

The argument between reinforcement theory and PCT is not a
dispute over which part of the loop is really the cause of behavior. It is
an argument about whether ANY variable in the loop should be
considered as a cause of the loop's behavior.

And, again, this is an argument about phenomena, not theory. If organisms
actually exist in a loop (if a closed loop relationship between organism
and environment is a real phenomenon) and if the behavior of this loop is
stable (another phenomenon) then these organisms ARE controlling some
aspect of their sensory input. Organisms pretty obviously do exist in
a closed loop (we can see that what occurs at their sense organs depends
continuously on what actions they are taking (how they are oriented, how
they are moving) and what actions they are taking depends on what occurs
at their sense organs.

So it's a pretty sure bet that PCT does apply to behavior; it's also a
sure bet that any theory that tries to identify one variable in this
closed loop (such as reinforcement) as a cause or "selector" of behavior
is wrong -- and will make the wrong predictions of the behavior of the
loop.

If Killeen's model actually is a closed loop model then it is either a
positive feedback model (which is stable only when the loop gain is less
than 1.0) or a negative feedback model. If it is the latter, then it's a
control of perception model, and it doesn't compete with PCT; it _is_
PCT. If, however, Killeen's model actually is PCT, then it is beyond
me why EAB researchers haven't been busy testing to see whether the
variables controlled by the model are what organisms actually do
control. Since there is no research aimed at determining controlled
variables in EAB, I'm pretty sure that no reinforcement model --
including Killeen's -- is a control model.

I wonder if everyone out there is a theorist? Aren't there any folks
who like to set up experiments that let us see if what our theories
say happens really does happen?

If not, then OK. Here's my contribution to the reinforcement/control
theory debate. Here is my derivation of Kileen's reinforcement model:

N = trunc(exp(-jr/b)*cos(Pi*F/s))

where r is reinforcement rate, b is behavior rate, F is the size of
the reinforcements (mg), s is seconds and N is the number of angels
that can dance on the head of a pin.

Best

Rick

[From Bruce Abbott (960111.1145 EST)]

Rick Marken (960110.2200) --

If Killeen's model actually is a closed loop model then it is either a
positive feedback model (which is stable only when the loop gain is less
than 1.0) or a negative feedback model. If it is the latter, then it's a
control of perception model, and it doesn't compete with PCT; it _is_
PCT. If, however, Killeen's model actually is PCT, then it is beyond
me why EAB researchers haven't been busy testing to see whether the
variables controlled by the model are what organisms actually do
control. Since there is no research aimed at determining controlled
variables in EAB, I'm pretty sure that no reinforcement model --
including Killeen's -- is a control model.

Well, Rick, don't let a mathematical proof get in the way of your
assertions; if it proves you wrong, just ignore it.

Killeen's 1995 JEAB model is a control system in the outer loop. If you
wish to assert that it is not, I wish you would provide the mathematical
justification for your assertion.

Killeen's 1994 BBS model is a more complete description of the inner
(reinforcement) loop of the 1995 model. It is NOT a control system, but it
IS an equilibrium system. An equilibrium model has negative feedback but is
not a control model. When you said

If Killeen's model actually is a closed loop model then it is either a
positive feedback model (which is stable only when the loop gain is less
than 1.0) or a negative feedback model. If it is the latter, then it's a
control of perception model, and it doesn't compete with PCT; it _is_
PCT.

you overlooked this third possibility.

I wonder if everyone out there is a theorist? Aren't there any folks
who like to set up experiments that let us see if what our theories
say happens really does happen?

First things first. Not too long ago your complaint was that there was no
reinforcement model with which to compare PCT. Here it is.

Regards,

Bruce

[From Rick Marken (960111.1000)]

Me:

If Killeen's model actually is a closed loop model then it is either a
positive feedback model (which is stable only when the loop gain is less
than 1.0) or a negative feedback model.

Bruce Abbott (960111.1145 EST) --

Well, Rick, don't let a mathematical proof get in the way of your
assertions; if it proves you wrong, just ignore it.

I was basing my assertion on Bill Powers' mathematical proof regarding the
properties of closed loop systems (P-systems and N-systems) described on
pp. 141-147 of "Living Control Systems".

Killeen's 1995 JEAB model is a control system in the outer loop. If you
wish to assert that it is not, I wish you would provide the mathematical
justification for your assertion.

I didn't assert that Killeen's model is _not_ a control system; I said that
_if it is_ a control system (and I'm happy to accept that it is) then I find
it strange that there has been absolutely no research done to test whether
the variable controlled by Killeen's model (F -- amount of food in the gut --
I believe it was in your implementation) actually IS controlled by real
organisms.

Killeen's 1994 BBS model is a more complete description of the inner
(reinforcement) loop of the 1995 model. It is NOT a control system, but it
IS an equilibrium system. An equilibrium model has negative feedback but is
not a control model.

If Killeen's model has negative feedback then it is a model of an N-system
(as described in LCS). An N-system is a control system (though the degree of
control exerted by the system depends on the loop gain). A control system
keeps its perceptual signal approximately equal to its reference signal, the
closeness of the approximation depending on the loop gain.

You seem to be saying that some N-systems are not control systems but,
rather, equilibrium systems. Could you please explain the difference
between an N-system that is a control system and an N-system that is an
equilibrium system.

An ideal N-system (infinite loop gain) is described by the following two
equations (adapted from Powers, LCS, 1989, p. 145):

o = g-1(r- h(d))

p = r

These are the equations for an _ideal_ N-system; output (o) is negatively
related to the effect of disturbance (h(d)) on input (p); and input (p)
precisely tracks the reference value (r). Could you please present the
equivalent equations for an equilibrium system.

Me:

Aren't there any folks who like to set up experiments that let us see if
what our theories say happens really does happen?

Bruce:

First things first.

Exactly. The first thing to do is collect data on control. There is _no_ data
on control available in the operant literature (at least, none has been
presented yet). Why start trying to develop models of control before you know
that control is even happening? You yourself showed that it's a good bet that
the animals in food "scheduling" experiments are _not_ in control of their
food input at all; they are just responding and eating as fast as they can.
If this is the case, then there is not much basis for building a control
model of this behavior.

Not too long ago your complaint was that there was no reinforcement model
with which to compare PCT. Here it is.

Yes. I wanted a reinforcement model so that we could finally go ahead and do
an _experiment_ to compare the predictions of the PCT and the reinforcement
model. You posted a Pascal version of Killeen's model; unfortunately, I
didn't save it. But I'm willing to take this model as "ground truth" for a
reinforcent model. Perhaps you could post it to me again. I will try to map
the variables in the model into variables that can be used in a computer
experiment using human subjects. Then we can get some data and compare it to
the predictions of Killeen's model and PCT. Whaddaya say?

Best

Rick

[Martin Taylor 960111 14:20]

Rick Marken (960110.2200)

I'm intrigued and a little puzzled.

If there is no controlling going
on (as revealed by The Test) then there is no need to consider control
theory as a model of the observed behavior.

Then you assert that there is another theory that applies to some
observed behaviour for which PCT fails?

I thought you were pretty adamantly opposed to that idea. It seems a bit
odd to me, but if you have some kind of a proposal for what this other
theory might be, I guess we would have to consider it seriously. Then
we would have to discover how to merge it with PCT in a theory that applies
to all behaviour.

Lots of fun till the millennium!

Martin

[From Rick Marken (960111.1245)]

Me:

If there is no controlling going on (as revealed by The Test) then there is
no need to consider control theory as a model of the observed behavior.

Martin Taylor (960111 14:20) --

Then you assert that there is another theory that applies to some observed
behaviour for which PCT fails?

Yes, indeed. For behavior of the sort we deal with, I'd say the theory that
applies when PCT fails (that is, the theory that applies when The Test
reveals that no control is involved in a behavior) is the best current
physical theory of the behavior.

If you do The Test and find that a variable (like food delivery rate) is not
under control, then you know that PCT is not relevant (not that it has
failed; PCT can't be used to explain the behavior of Mars as it orbits
the Sun but this is not a failure of PCT; it just means that Mars' behavior
does not involve control). The theory that explains food delivery behavior
(if food delivery is not controlled) is Newtonian mechanics combined with
some circuit theory.

There are many behaviors -- like falling off a garage at 32 ft/sec^2 -- that
don't involve control. We usually don't call these events "behaviors", I
think, because we typically use this word, especially in the context of
discussions of living organisms, to refer only to "intentionally produced
results of action". People (notably behavioristic people) might not care for
this restrictive definition of "behavior" but I think this is the way the
word is used in practice.

Best

Rick