PCT versus reinforcement theory

[From Bruce Abbott (960128.1610 EST)]

Rick Marken (960127.1700) --

Bruce Abbott (960127.1720 EST)

You are wrong about my not being interested in rejecting Killeen's
theory. I am just not interested in rejecting it on spurious grounds.

You still seem to think that there is something subtle about the
difference between a model that controls its input (like PCT) and
one that doesn't (like Killeen's). So far as I can recall, every
difference between reinforcement theory and PCT to which we have
pointed over the last several months (and these differences have
_not_ been subtle) has been declared "spurious" by you.

I call 'em as I see 'em.

I am impressed by the fact that after well over a year of
studying PCT (you were presumably already familiar with
reinforcement theory) you haven't been able to find one glaring,
non-spurious way in which PCT differs from reinforcement theory;
you can think of no non-spurious basis on which to compare these
two theories. Doesn't this make _you_ wonder what you are
controlling for?

No, I know what I am controlling for: truth. When I perceive untruth I try
to correct it. Sometimes the truth is not what we want to hear. What are
_you_ controlling for?

Regards,

Bruce

[Hans Blom, 951023]

(Bruce Abbott (951022.1405 EST))

In my understanding of reinforcement theory, people would be under-
stood according to what reinforces their behavior, what behavior has
been reinforced in the past, and in the presence of what discrimina-
tive stimuli. Reinforcement theory provides an explanation for why
people act as they do; PCT provides a different explanation. Both
attend to what the organism perceives about its environment, but
offer a different view of how those perceptions relate to behavior.
Both attempt to understand people (and animals) "according to what
drives them to act," as you put it.

"PCT provides a different explanation" from reinforcement theory
(RT), you say. For once, I disagree with you. Or maybe not. Depends
upon what you mean by "explanation". Both PCT and RT study the same
domain of knowledge, so their conclusions, if valid, must agree.
Overwhelmingly, they do indeed. Being the devil's advocate (or
heretic) that I am, I tend to read RT publications from my control
perspective, and frequently the only problem I have is in the attempt
to find equivalences between notions, just like one can express the
same ideas in different languages. Just like in languages, there is
seldomly a one-to-one translation from word to word. Yet, different
languages can "explain" the same things. And much of the RT litera-
ture is very understandable once you know its idiom.

Where differences arise, is that PCT mostly studies control after
learning has been finished (i.e with a steady state, fully converged
model), whereas RT mostly studies learning (i.e. convergence of the
model) -- necessarily while control goes on. Maybe I am biased, but
from the perspective of adaptive / model-based control theory, most
of RT is perfectly understandable. And mostly offers valid conclu-
sions. Often superfluous, maybe, because they are confirmations of
the basic theoretical findings, but affirmations, nonetheless, that
the basic theory applies as well in the new cases described.

If by "explanations" you mean "models" (i.e. subjective -- i.e. from
a certain perspective -- simplifications of reality), I agree. The
models have different form (depend, partly, on different underlying
concepts) and different content (because they are descriptions of
different things -- mostly control versus mostly learning).

So, in my opinion, the disagreements are minor. It sometimes seems to
me that a well-known social process takes place in many PCT comments
on RT, as also happens in e.g. religion: the closer sects are to each
other within the religious spectrum, the more necessary it seems for
followers to stress distinctions rather than similarities.

So basically I agree with you that much of RT provides a fine basis
for more PCT research. You say something similar:

By the way, PCT offers its own definition of a reinforcer: that
which when produced by an action, reduces the error signal in the
system producing the action.

Greetings,

Hans

[From Bruce Abbott (951023.1235 EST)]

[Hans Blom, 951023]

(Bruce Abbott (951022.1405 EST))

"PCT provides a different explanation" from reinforcement theory
(RT), you say. For once, I disagree with you. Or maybe not. Depends
upon what you mean by "explanation".

Disagree with me? Now who would do that? (;-> What I meant was that
reinforcement theory and PCT offer different mechanisms to explain the
observations. RT says that certain consequences of behavior "strengthen"
that behavior, making it more probable in the presence of environmental
stimuli present during conditioning. These stimuli then "control" the
probablility of the reinforced response. This is, as you are well aware, a
different mechanism from the one proposed by PCT.

Both PCT and RT study the same
domain of knowledge, so their conclusions, if valid, must agree.
Overwhelmingly, they do indeed. Being the devil's advocate (or
heretic) that I am, I tend to read RT publications from my control
perspective, and frequently the only problem I have is in the attempt
to find equivalences between notions, just like one can express the
same ideas in different languages. Just like in languages, there is
seldomly a one-to-one translation from word to word. Yet, different
languages can "explain" the same things. And much of the RT litera-
ture is very understandable once you know its idiom.

Yes. At the surface level the two theories often describe the same apparent
relationships, as when certain actions are observed to increase when those
actions produce certain consequences. One can call such consequences
"reinforcers" because of their apparent effect on the behaviors that produce
them, and demonstrate that such consequences may have a similar apparent
effect on other behaviors if those behaviors are allowed to produce them.
Control theory provides the mechanism through which these surface phenomena
appear. In the context of steady-state performance, reinforcement, the
observable phenomenon, is thereby explained in terms of control-system
action under certain specific conditions (presence of error).

Where differences arise, is that PCT mostly studies control after
learning has been finished (i.e with a steady state, fully converged
model), whereas RT mostly studies learning (i.e. convergence of the
model) -- necessarily while control goes on. Maybe I am biased, but
from the perspective of adaptive / model-based control theory, most
of RT is perfectly understandable. And mostly offers valid conclu-
sions. Often superfluous, maybe, because they are confirmations of
the basic theoretical findings, but affirmations, nonetheless, that
the basic theory applies as well in the new cases described.

As I pointed out some time ago, the reinforcement principle has been applied
to two distinct cases: acquisition of behavior and maintenance of that
behavior. Control systems theory explains the observations labeled
"reinforcement" in the steady-state, but does not yet offer an adequate
explanation for acquisition, in my opinion. "Reorganization" as presently
developed only states that persistant error triggers changes in
control-system parameters, which is only to state that learning takes place
when learning is required. It has been suggested that this reorganization
might involve a random, trial and error process. Reinforcement as an
acquisition principle may yet have something to add here, although that
remains to be demonstrated.

Because the _phenomenon_ of reinforcement is an objective fact, PCT does not
and cannot disprove it; on the contrary, PCT _predicts_ it and _explains_
it, at least in the case of maintained behavior. PCT tells when a given
consequence of behavior will serve (or will not serve) as a reinforcer and
why this particular consequence will have the effect on that behavior it is
observed to have.

So basically I agree with you that much of RT provides a fine basis
for more PCT research.

Thanks, Hans, your check is in the mail. (;->

Regards,

Bruce