Varying Lever Pressure

[bnhpct]

[From Bruce Nevin (2000.11.18 1234 EDT)]

Bill Powers (2000.11.18.0404 MST)--

So your contention that control theory investigates behavior while
behaviorism investigates learning does not bear up under close examination.
It is impossible to study learning except in the context of a model of
behavior.

My purpose was not to defend behaviorism but to figure out why no one has proposed a crucial experiment (as somebody called it, Popper?) of the sort Rick was asking about, and how to go about doing so. The point was not that one studies one and the other the other. Probably both would regard a theory of learning as secondary to (derivative of, in a sense) a theory of behavior. The point is rather that there is a striking difference as to what is in the foreground and what is put in the background, either because it is taken for granted (EAB) or because it is deferred for future investigation (PCT).

What I was trying to suggest is first that a direct attack on their bankrupt theory of behavior is ineffectual because in their minds it scarcely needs defending, it is taken for granted; and second that it could be more effective to develop experimental studies of learning that more adequately cover the ground where their attention *is* focused.

How might the case be made that the PCT account in terms of variables and functions covers the same ground (learning) better than the operant conditioning account in terms of frequencies of discrete events?

This sounds like the contrast Chomsky drew very effectively in the 1950s and 1960s between descriptive adequacy and explanatory adequacy. We know that the behaviorist account is not an adequate explanation, but convincing them of that is tough. A crucial experiment must make the case that their account is not even an adequate *description* of their own data, and that a description of the data that is informed by PCT includes essential information that they leave out. The critical thing is that they have to care about the stuff that they're leaving out.

Generalization is one field of data that could be a good candidate. I don't control the issues well enough to suggest anything specific, but perhaps Bruce Abbott could?

Bill Powers (2000.11.18.0404 MST)--

All that is required is to introduce disturbances of the
consequence that behaviorists call reinforcement.

This must be done in such a way that they won't say "Oh, well, it's not the same consequence now. You've changed the consequence." Disturb the subject organism's control without disturbing the EAB observer's control.

The clicking movement of a latched lever is a discrete event, and the rat applies whatever pressure is necessary to bring it about. By measuring muscle tension (or force or intensity signals) and lever resistance in the midst of an experiment that measures frequency of lever presses, you are setting up a PCT experiment upon an EAB experiment. What is a side effect for the observer of experiment 1 is a variable that the observer controls in experiment 2.

Using the example of lever pressure, it sounds like the EAB claim is that when the resistance of the lever increases there must be a learning process before the rat applies sufficient pressure, and the PCT claim is that the rat varies its responses to counter the disturbance without a learning process -- "immediately and without practice". That as the resistance of the lever is slowly varied the rat will always push just hard enough to unlatch it. Importantly, as resistance goes down, the rat stops pushing immediately when the lever clicks, with less and less effort; it is much more obvious that the reverse is true as resistance goes up, and that effort stops immediately when the lever clicks. (Why is it more obvious?)

This experiment wouldn't do because the rat is not learning a new control system for pushing on a lever until it moves, the rat is learning to push a lever repeatedly until food appears.

So why should an event-counter care about this continuous variation? Are you removing something from his field of vision (the field he calls "learning") to the field called control?

If resistance is abruptly changed, of course, we would see the equivalent of falling through the door that you thought was stuck or bumping against the door that you thought swung freely. One of perhaps many wiggle-worms to be extracted from the can labelled "learning" and given its own account, a making of relevant distinctions that PCT can do and, I suspect, EAB cannot. Such distinctions of different phenomena all called "learning" (distinguished by their explanatory mechanisms in PCT, absent in the theory of behavior underlying EAB) can be the basis for designing experiments that show contrasts between the two theories of behavior, doing that in terms of the phenomena of learning that are of concern to EAB researchers.

         Bruce Nevin

···

At 05:36 AM 11/18/2000 -0700, Bill Powers wrote:

[Richard_Marken2]

[From Rick Marken (2000.11.18.1710)]

Me:

What I would like to see us do is develop an experiment that would
convince _us_ -- those of use who are involved in developing the
experiment -- that the PCT view of behavior is (or is not)
preferable to the reinforcement view.

Bruce Gregory (2000.1118.1736)

Great idea. However, it has already been done by Bourbon and
Powers. Nevertheless, those who cannot remember the past are
welcome to repeat it.

Yes. The Bourbon/Powers research is precisely the kind I'm
thinking of. In their "Models and their Worlds" paper they
used environmental disturbances (the "worlds" of the title)
to show that the output generation (cognitive) and stimulus guidance
(S-R) views of behavior (the "models" of the title) can't account
for the observed behavior in a simple tracking task.

I think a similarly _simple_ experimental situation can be
designed to show that the reinforcement view of behavior can't
account for the behavior that occurs in some sort of learning
task. Since you are familiar with the Bourbon/Powers research,
perhaps you could start the ball rolling by describing the
outlines of a simple experimental test that pits reinforcement
theory against control theory.

Best

Rick

···

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: marken@mindreadings.com
mindreadings.com

[Abbott_Bruce]

[From Bruce Abbott (2000.11.21.1700 EST)]

Bill Powers (2000.11.15.0819 MST)]

Bruce Abbott (2000.11.14.1330 EST)--

Those who subscribe to this view believe that the selecting is being done by
mechanisms within the organism; the notion that the environment is doing the
selecting is simply a shorthand way of noting that the environmental
consequences of behavior, as perceived by the organism, result in a change
in what the organism is likely to do in the future under similar conditions.
Certain consequences have this effect because of the way the organism is
organized.

Why, if they believe that the organism initiates and does the selecting, do
they use terms which suggest (strongly) that it is the environment that
does the initiating and selecting? I am reluctant to believe that this is
merely a sloppy habit of speech. I have never seen it said in a JEAB
article that an organism determined its own behavior and controlled the
consequences of that behavior. Everything I have seen suggests the exact
opposite. Are you really speaking for the behaviorist community when you
say they mean that the organism is the controller and the environment only
a passive responder to the organism's actions?

That's not what I said at all -- I said that the behaviorist community would
acknowledge that the observed changes in behavior are mediated by some
mechanism within the organism. There must be some mechanism that alters the
future probability of a given behavioral act when that act is followed by
certain consequences. I don't think that saying this is the same as saying
that the organism is actively choosing to engage in certain behavioral acts
for the purpose of reproducing the consequence.

What that mechanism is and how it might work is another matter. I do think
that many of those same behaviorists would take the view that the resulting
changes _appear_ to be purposive without really being so, in the same way
that Darwinian selection results in changes that appear to be purposive when
in fact all that is happening is blind variation and differential survival.

For the time being, please direct any replies to this post both to CSGnet
_and_ to my address (abbott@ipfw.edu). At present I'm getting posts from
the CSGnet archives and don't want to download the current 600+ K file if I
can avoid it.

Bruce A.

[Richard_Marken2]

[From Rick Marken (2000.11.21.1530)]

Bruce Abbott (2000.11.14.1330 EST)--

Those who subscribe to this view believe that the selecting is
being done by mechanisms within the organism...

Bill Powers (2000.11.15.0819 MST)--

Why, if they believe that the organism initiates and does the
selecting, do they use terms which suggest (strongly) that it
is the environment that does the initiating and selecting?

Bruce Abbott (2000.11.21.1700 EST)

That's not what I said at all -- I said that the behaviorist
community would acknowledge that the observed changes in
behavior are mediated by some mechanism within the organism.

Actually, what you said (see above) is exactly what Bill thought
you said: that behaviorists believe that the mechanisms that
_select_ consequences are in the organism. Now you agree that
the behaviorists don't really believe this at all. All they
believe is that mechanisms in the organism _mediate_ the
selection process. And, of course, this is true. What "mediates"
the selection process, according to behaviorists, is some mech-
anism in the organism that forms associations between stimuli
and responses. This association mechanism is _operated on_ by
environmental consequences (reinforcements) so that some
associations are strengthened and others weakened by these
consequences. This is quite a different view of the selection
process than the PCT view. In PCT, mechanisms within the organism
mediate the selection of environmental consequences _by behavior_.
In behaviorism, mechanisms within the organism mediate the
selection of behavior (responses) by environmental consequences.

I think these views are so clearly different that it should
not be difficult to design experiments that clearly reveal the
difference. I would really like to hear your ideas for such
an experiment.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
MindReadings.com mailto: marken@mindreadings.com
www.mindreadings.com

[Bill_Powers1]

[From Bill Powers (2000.11.21.1148 MST)]

Bruce Abbott (2000.11.21.1700 EST)--

That's not what I said at all -- I said that the behaviorist community would
acknowledge that the observed changes in behavior are mediated by some
mechanism within the organism. There must be some mechanism that alters the
future probability of a given behavioral act when that act is followed by
certain consequences. I don't think that saying this is the same as saying
that the organism is actively choosing to engage in certain behavioral acts
for the purpose of reproducing the consequence.

Only a very small change would turn this into active choosing. If you said
that this mechanism resides in the organism, and that it alters the
probabilities of many behaviors until it finds one that will reproduce the
desired consequence, you would have the PCT view.

Your recasting doesn't really say, yet, what the difference with PCT is.
Under PCT, organisms do not choose "to engage in certain behavioral acts"
either unless those acts themselves are specifically intended. An example
would be a dancer executing an arabesque. More generally, organisms choose
to reproduce consequences of acts, using whatever behavioral acts are
needed at the moment to reproduce them (this will change, of course,
according to what disturbances happen to be acting from moment to moment,
as well as on environmental contingencies).

I have been trying, lately, to bring up the subject of disturbances again
to see what you will say. When the occurrance of the consequence is being
disturbed, a _different_ behavioral action will be needed to reproduce the
_same_ consequence. A control system can do this immediately and
automatically because of the negative feedback loop involved in control. If
the consequence is altering the future probability of a specific behavioral
act, on the other hand, it will have no way to take disturbances into
account (especially those not accompanied by any perceptible signs that a
disturbance is present, other than the change in the consequence itself).

What that mechanism is and how it might work is another matter.

I don't think that "altering the future probability of a given behavioral
act" will work to reproduce a consequence of that act when other influences
are having unpredictable effects on the same consequence. Yet I would claim
that organisms can reproduce specific consequences under conditions where
actions must vary in order to reproduce them. Would this not refute the
behaviorist model?

I do think
that many of those same behaviorists would take the view that the resulting
changes _appear_ to be purposive without really being so, in the same way
that Darwinian selection results in changes that appear to be purposive when
in fact all that is happening is blind variation and differential survival.

I can't tell from this whether you are siding with them. In PCT, of course,
we have a model of purposive behavior (including changes of behavior via
reorganization) under which the appearance of purposiveness can be taken as
factual. We have also demonstrated that even with random variation being
involved, there can be purposive selection of specific consequences from
among a set of randomly-produced consequences, where the purpose is a
built-in preference of the organism. Differential survival does exist, but
it is not the only possible selection mechanism, and it is far from the
most efficient possible mechanism.

Do behaviorists have a concept of purposive behavior, even if only to deny
that it exists?

Best,

Bill P.

[Abbott_Bruce]

[From Bruce Abbott (2000.11.22.1220 EST)]

Be forwarned that our net access from home keeps failing lately (it did
again last night). If it happens again, they're not likely to fix it
promptly over our holiday break.

[From Bill Powers (2000.11.21.1148 MST)]

Bruce Abbott (2000.11.21.1700 EST)--

That's not what I said at all -- I said that the behaviorist community would
acknowledge that the observed changes in behavior are mediated by some
mechanism within the organism. There must be some mechanism that alters the
future probability of a given behavioral act when that act is followed by
certain consequences. I don't think that saying this is the same as saying
that the organism is actively choosing to engage in certain behavioral acts
for the purpose of reproducing the consequence.

Only a very small change would turn this into active choosing. If you said
that this mechanism resides in the organism, and that it alters the
probabilities of many behaviors until it finds one that will reproduce the
desired consequence, you would have the PCT view.

I did say, before, that behaviorists would agree that it resides in the
organism. They wouldn't be so happy with "desired" or with the idea that it
is intentionally searching for a behavior that will reproduce the
consequence, because these are from their perspective mentalism and mental
stuff is not viewed as having any causal role to play in behavior. However,
cognitive types wouldn't find much if anything to object to in that statement!

Your recasting doesn't really say, yet, what the difference with PCT is.
Under PCT, organisms do not choose "to engage in certain behavioral acts"
either unless those acts themselves are specifically intended. An example
would be a dancer executing an arabesque. More generally, organisms choose
to reproduce consequences of acts, using whatever behavioral acts are
needed at the moment to reproduce them (this will change, of course,
according to what disturbances happen to be acting from moment to moment,
as well as on environmental contingencies).

In the strictly Skinnerian view, no physical mechanism is proposed to carry
out the process. There is simply description: Behavior varies. When
certain events occur while the organism is doing something, the organism
tends to do that something again, especially those portions of the act that
repeatedly are ongoing when the event occurs. If certain stimuli are
present when these acts are followed by those events, then the acts that are
often followed by those events tend to occur with greater frequency when
those stimuli are present and not necessarily at other times.

Others have from time to time proposed mechanisms of various sorts (remember
Killeen?), but none have become generally accepted, to my knowledge.

I have been trying, lately, to bring up the subject of disturbances again
to see what you will say. When the occurrance of the consequence is being
disturbed, a _different_ behavioral action will be needed to reproduce the
_same_ consequence. A control system can do this immediately and
automatically because of the negative feedback loop involved in control. If
the consequence is altering the future probability of a specific behavioral
act, on the other hand, it will have no way to take disturbances into
account (especially those not accompanied by any perceptible signs that a
disturbance is present, other than the change in the consequence itself).

We had this discussion before, but to refresh everyone's memory, we are
talking about observable behavioral acts (like crossing from one side of a
shuttlebox to the other), parts of the behavorial stream, that the organism
can repeat with reasonable fidelity. Those portions of those acts that are
consistently followed by certain events tend to be repeated with increasing
frequency over time. At bottom there is presumed to be some sort of
associative process at work. At this level of analysis, what is important
is that "shuttling to the other compartment" is a bit of behavior that the
rat can repeat. When this behavior is followed by some consequence of
importance to the animal (like termination of shock), the rat tends to
repeat it (in this case, when the shock comes on again).

The rat may be doing this because it "wants to escape from the shock," but
this takes us into the mind of the rat and makes it sound as though spooky
nonphysical mental forces can exert effects on physical matter, never mind
the fact that one cannot get into a rat's mind and record what, if anything,
it is thinking. The strategy the behaviorists adopted to avoid this trap
was to stick to observables.

What that mechanism is and how it might work is another matter.

I don't think that "altering the future probability of a given behavioral
act" will work to reproduce a consequence of that act when other influences
are having unpredictable effects on the same consequence. Yet I would claim
that organisms can reproduce specific consequences under conditions where
actions must vary in order to reproduce them. Would this not refute the
behaviorist model?

The question, then, is why the rat is doing what it is doing. Is it trying
to produce the consequence? Or is it that, because of the nature of that
consequence and the rat's history of repeatedly experiencing that
consequence while doing the act, the rat is driven to repeat the act? I
think that many behaviorists think it is the latter, but behaviorists are a
diverse bunch and I wouldn't want to paint them all with that same brush.

I do think
that many of those same behaviorists would take the view that the resulting
changes _appear_ to be purposive without really being so, in the same way
that Darwinian selection results in changes that appear to be purposive when
in fact all that is happening is blind variation and differential survival.

I can't tell from this whether you are siding with them.

No, I'm not siding with them. I'm trying to separate a purely descriptive
account (which Skinner aspired to but didn't always achieve) and one that
proposes some specific mechanism. At the descriptive level, behavior
varies. When certain behavioral acts are fairly consistently followed by
certain events, we observe that those acts begin to occur more frequently.
We have the three elements of Darwinian evolution: variation, selection, and
resulting change in the observed population (in this case, the population of
behavioral acts). This is Skinner's "selection by consequences."

If we now ask about mechanism, many behaviorists will begin to talk about
associations being formed or strengthened and other such stuff that
represent processes inferred from the observations. At the level of
inferred processes, behaviorists and cognitivists begin sounding much alike
if you ignore the differences in vocabulary.

In PCT, of course,
we have a model of purposive behavior (including changes of behavior via
reorganization) under which the appearance of purposiveness can be taken as
factual. We have also demonstrated that even with random variation being
involved, there can be purposive selection of specific consequences from
among a set of randomly-produced consequences, where the purpose is a
built-in preference of the organism. Differential survival does exist, but
it is not the only possible selection mechanism, and it is far from the
most efficient possible mechanism.

This is where I think misunderstanding arises about the position I have been
taking. The selecting referred to in "selection by consequences" refers to
the flagging of certain behavioral acts for repetition -- those that
consistently have been in progress when certain consequences occurred. The
organism wants (has a reference for) the consequence but does not know how
to get it. When what it wants occasionally appears while the organism is
engaged in behavior, it attempts to reproduce (sets a reference for) those
conditions that were present when the event occurred in the past. It is as
if the rat is developing hypotheses about what behavioral act is producing
the consequence, and testing them.

What determines which behavioral act the rat will reproduce is the
coincidence between behavioral act and appearance of the desired
consequence. This is "selection by consequences."

This is a different process entirely from the one described as "selection
_of_ consequences," which assumes that the organism already knows what it
needs to do to produce a given consequence, and selects which consequence to
produce based on higher-order needs.

Do behaviorists have a concept of purposive behavior, even if only to deny
that it exists?

Behaviorists would note that mechanisms can generate behavior that is
adaptive and apparently purposive (e.g., Darwinian evolution) without the
organism in question having the slightest intention to produce those
results. They would view the reinforcement process in this same way -- as a
mechanism that produces adaptive, apparently purposive behavior without the
organism in question having intentions. Many would view intensions as
nonphysical, mental things that can have no causal influence on physical
mechanisms or behavior.

Bruce A.

[Bruce_Gregory8]

[From Bruce Gregory (2000.1122.1251)]

Bruce Abbott (2000.11.22.1220 EST)

The strategy the behaviorists adopted to avoid this trap
was to stick to observables.

Isn't it fortunate that Newton didn't take that route.

BG

[Richard_Marken2]

[From Rick Marken (2000.11.22.1030)]

Bill Powers (2000.11.21.1148 MST) to Bruce Abbott --

Yet I would claim that organisms can reproduce specific
consequences under conditions where actions must vary in
order to reproduce them. Would this not refute the behaviorist
model?

Bruce Abbott (2000.11.22.1220 EST)

The question, then, is why the rat is doing what it is doing.

No. The question is "Would this [the fact that organisms can learn
to reproduce specific consequences under conditions where actions
must vary in order to reproduce them] not refute the behaviorist
model?". I would really like to hear your answer to this question.
If your answer is "yes" then that's all you need to say; I already
know why it would refute the behaviorist model. If, however, your
answer is "no" then I think you've got some 'splainin' to do.

Best

Rick

···

---
Richard S. Marken Phone or Fax: 310 474-0313
MindReadings.com mailto: marken@mindreadings.com
www.mindreadings.com

[Abbott_Bruce]

[From Bruce Abbott (2000.11.22.1525 EST)]

Rick Marken (2000.11.22.1030) --

Bill Powers (2000.11.21.1148 MST) to Bruce Abbott

Yet I would claim that organisms can reproduce specific
consequences under conditions where actions must vary in
order to reproduce them. Would this not refute the behaviorist
model?

Bruce Abbott (2000.11.22.1220 EST)

The question, then, is why the rat is doing what it is doing.

No. The question is "Would this [the fact that organisms can learn
to reproduce specific consequences under conditions where actions
must vary in order to reproduce them] not refute the behaviorist
model?". I would really like to hear your answer to this question.
If your answer is "yes" then that's all you need to say; I already
know why it would refute the behaviorist model. If, however, your
answer is "no" then I think you've got some 'splainin' to do.

The answer is "no." If actions must vary to reproduce specific
consequences, then there must have been some way the organism learned what
actions to vary and in what direction. The behaviorist view would be that
such appropriate behaviors occur now because in the past such behaviors have
been reinforced under these conditions. So, when conditions are such as to
require different actions, what actions occur depends on what actions have
been successful in producing those consequences in the past under those same
conditions or conditions like them.

You're still confusing the cases of learning what to do and doing what you
already know how to do. How do you get from one state to the other? If you
start by assuming that the system is already organized to perform the
correct actions as needed to bring about the consequence, then it is no
surprise that it works as required. To put reinforcement and PCT models on
equal footing, both must begin with a naive subject! In that situation it
is not at all clear to me that PCT's reorganization process is up to the job.

Bruce A.

[Abbott_Bruce]

[From Bruce Abbott (2000.11.22.1535 EST)]

Bruce Gregory (2000.1122.1251) --

Bruce Abbott (2000.11.22.1220 EST)

The strategy the behaviorists adopted to avoid this trap
was to stick to observables.

Isn't it fortunate that Newton didn't take that route.

Yes, instead he unhappily settled for "spooky action at a distance," very
much against his will. Behaviorism represented an attempt to keep at bay
the accepted theory of behavior in those days (and still fashionable today
in some quarters!): spooky action from inside (nonphysical "mental" causation).

I appreciate why they chose to go this route, but agree with you that they
ended up throwing the baby out with the bath water.

Bruce A.

[Richard_Marken2]

[From Rick Marken (2000.11.22.1330)]

Me:

The question is "Would this [the fact that organisms can learn
to reproduce specific consequences under conditions where actions
must vary in order to reproduce them] not refute the behaviorist
model?". I would really like to hear your answer to this question.

Bruce Abbott (2000.11.22.1525 EST)

The answer is "no." If actions must vary to reproduce specific
consequences, then there must have been some way the organism
learned what actions to vary and in what direction. The behaviorist
view would be that such appropriate behaviors occur now because
in the past such behaviors have been reinforced under these
conditions. So, when conditions are such as to require different
actions, what actions occur depends on what actions have been
successful in producing those consequences in the past under
those same conditions or conditions like them.

Thanks. Nice, clear answer. And I think it also clearly suggests
the experiment that would distinguish the PCT and behaviorist
views of learning. As you say, the behaviorist would explain
the fact that "such appropriate behaviors occur now [under new
conditions]" as being a result of the fact that "in the past
such behaviors have been reinforced under these conditions". So
I submit that the following experiment would distinguish the
behaviorist from the PCT view of learning:

The organism learns to reproduce some specific consequence under
conditions where a specific disturbance requires specific action
variations in order to reproduce it. Once the organism has learned
the action variations that reproduce the consequence, a new
disturbance is introduced so that _different_ action variations
are required to reproduce the consequence. For example, the
organism could be required to learn to move a cursor to a target
(cursor under target is the "specific consequence") while a
sinusoidal disturbance is applied to the position of the target.
Once the organism has learned to reproduce this consequence
(consistently move the cursor to the target) the disturbance
to the target could be changed from a sine wave to a square wave.

The behaviorist prediction is that the organism will have to learn
anew the behavior (square rather than sine wave action variations)
that reproduce the consequence because the newly appropriate behavior
(square wave action variations) was not previously reinforced. The
PCT prediction, on the other hand, is that the organism will continue
to reproduce the consequence, using a never before performed (and,
thus, never before reinforced) set of action variations, without
difficulty.

What do you think?

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
MindReadings.com mailto: marken@mindreadings.com
www.mindreadings.com

[Bill_Powers1]

[From Bill Powers (2000.11.22.1832 MST)]

Bruce Abbott (2000.11.22.1220 EST)--

I did say, before, that behaviorists would agree that it resides in the
organism. They wouldn't be so happy with "desired" or with the idea that it
is intentionally searching for a behavior that will reproduce the
consequence, because these are from their perspective mentalism and mental
stuff is not viewed as having any causal role to play in behavior. However,
cognitive types wouldn't find much if anything to object to in that

statement!

"Mentalism" is a pretty old-fashioned concept. I think most scientists
today view mental activity as neural activity in the brain (as do I), which
certainly can't be considered "spooky."

If certain stimuli are
present when these acts are followed by those events, then the acts that are
often followed by those events tend to occur with greater frequency when
those stimuli are present and not necessarily at other times.

The term "stimuli" does not refer to something observable, because it
presumes that the observable event or object referred to is causing neural
activities inside the organism, which can't be observed (without invasive
procedures). To make this way of reporting the process objective, you'd
have to describe the entire environment around the behaving organism,
because there is no _a priori_ way to determine what is or is not a
stimulus at a given time.

I have been trying, lately, to bring up the subject of disturbances again
to see what you will say.

We had this discussion before, but to refresh everyone's memory, we are
talking about observable behavioral acts (like crossing from one side of a
shuttlebox to the other), parts of the behavorial stream, that the organism
can repeat with reasonable fidelity.

The fidelity of repeating the behavior is not what is in question. The
problem is that in real environments (that is, natural environments outside
the laboratory), repeating the same behavior will _not_ cause the same
consequence to be repeated. In fact, in order to produce the same
consequence twice in a row, the most general case is that the behavior will
have to change just enough to offset changes in disturbances that are
acting. So an organism cannot learn to produce the same consequence
repeatedly by learning to produce the same action repeatedly. In general, a
_different_ amount and direction of behavior must be generated in order to
produce the same consequence as before.

In the behaviorist's laboratory, of course, things are arranged so
repeating the same behavior _will_ cause the same consequence to repeat.
Thus it may appear that what the organism learns is to repeat the same
behavior that was associated with the consequence. Only by introducing
disturbances can we distinguish between learning to repeat a behavior and
learning to _vary_ behavior when that is required to keep producing the
same consequence. The latter, of course, is what a control system does.
Note that if an organism learns in the latter way, it will naturally repeat
the same behavior to produce the same consequence when there are no
disturbances. The control system can do everything that the system
organized the other way can do. But the system that learns to repeat the
same action cannot do what a control system does.

Best,

Bill P.

[Bruce_Gregory8]

[From Bruce Gregory (2000.1123.0632)]

Bill Powers (2000.11.22.1832 MST)

In the behaviorist's laboratory, of course, things are arranged so
repeating the same behavior _will_ cause the same consequence to repeat.
Thus it may appear that what the organism learns is to repeat the same
behavior that was associated with the consequence. Only by introducing
disturbances can we distinguish between learning to repeat a behavior and
learning to _vary_ behavior when that is required to keep producing the
same consequence. The latter, of course, is what a control system does.
Note that if an organism learns in the latter way, it will naturally repeat
the same behavior to produce the same consequence when there are no
disturbances. The control system can do everything that the system
organized the other way can do. But the system that learns to repeat the
same action cannot do what a control system does.

We might reason that the thermostat is simply an S-R system. Temperature
less than t -- if the furnace is on, leave it on; if the furnace is off,
turn it on. Temperature greater than t -- if the furnace is on, turn it
off; if the furnace is off, leave it off. So the thermostat mediates its
output by applying a simple rule to the input (which includes the state of
the furnace). We can therefore map the output of the furnace to the input
temperature and control the behavior of the furnace by manipulating the
reinforcer -- the temperature. As long as an S-R model allows the link
between environment and response to include more than one variable
(including "internal" variables), why can't such a model replicate the
behavior of a control system?

Happy Thanksgiving.

BG

[Bill_Powers1]

[From Bill Powers (2000.11.23.0534 MST)]

Bruce Gregory (2000.1123.0632)--

We might reason that the thermostat is simply an S-R system. Temperature
less than t -- if the furnace is on, leave it on; if the furnace is off,
turn it on. Temperature greater than t -- if the furnace is on, turn it
off; if the furnace is off, leave it off. So the thermostat mediates its
output by applying a simple rule to the input (which includes the state of
the furnace). We can therefore map the output of the furnace to the input
temperature and control the behavior of the furnace by manipulating the
reinforcer -- the temperature. As long as an S-R model allows the link
between environment and response to include more than one variable
(including "internal" variables), why can't such a model replicate the
behavior of a control system?

By reducing the description of the system to these terms, you have
effectively designed a control system with a reference level of t. A
control system actually is an S-R system with a reference level added
(which defines the level of input that the system prefers) and of course an
input that depends on its output and disturbances. From this design, you
can deduce (approximately) how the system will behave under conditions of
disturbance and no disturbance.

When there is no disturbance (or a constant one), a given on-off ratio of
the furnace will result in a certain equilibrium temperature. So every time
the same furnace behavior occurs, the same room temperature will result. If
we were thinking of the temperature as a reinforcer, we might conclude that
occurrance of a temperature of t degrees reinforces an on-off behavior
ratio of p percent. If the room started at some other temperature, we would
see the actual on-off ratio change until it produced a temperature of t, at
which point the thermostat would continue to produce the same on-off ratio, p.

If, however, there were a varying outside temperature, we would not find
this simple relationship. The temperature would still be brought to t and
be maintained there, but the on-off ratio of the furnace would vary,
apparently at random. Sometimes it would be on 90 percent of the time, and
sometimes 30 percent, depending on the outside temperature. The consequence
(the temperature) would no longer be the same every time the behavior (the
on-off ratio) repeated. In fact, if the thermostat did learn what on-off
ratio would produce a temperature of t, and reproduced that behavior, it
would be unable to produce the same room temperature when the outside
temperature varied.

My point is that you can't tell, when no variable disturbances are acting,
whether the system has learned to produce a specific behavior or to vary
its behavior to produce a specific consequence. The two types of system
would show the same behavior. Only when we disturb the consequence can we
see a difference between the two types of system. When disturbances are
present, the system that learned a specific behavior will no longer
reproduce the same consequence; only the control system will be able to
vary its behavior to keep the consequence the same.

Nice example, and you made me think a bit to come up with an answer. I
think it makes the point clearer.

Best,

Bill P.

[Bruce_Gregory8]

[From Bruce Gregory (2000.1123.1026)]

Bill Powers (2000.11.23.0534 MST)

If, however, there were a varying outside temperature, we would not find
this simple relationship. The temperature would still be brought to t and
be maintained there, but the on-off ratio of the furnace would vary,
apparently at random. Sometimes it would be on 90 percent of the time, and
sometimes 30 percent, depending on the outside temperature. The consequence
(the temperature) would no longer be the same every time the behavior (the
on-off ratio) repeated. In fact, if the thermostat did learn what on-off
ratio would produce a temperature of t, and reproduced that behavior, it
would be unable to produce the same room temperature when the outside
temperature varied.

The issue seems to revolve around what we call the behavior. You have
provided a quantitative formulation, the on-off ratio, but a less
quantitative description, behavior consists of turning the furnace on and
off, allows the behaviorist to "save" his model. (I use the masculine,
since behaviorism is something I suspect few women would want "credit" for.)

BG

[Bruce_Gregory8]

[From Bruce Gregory (2000.1123.1245)]

Rick Marken (2000.11.22.1330)

The behaviorist prediction is that the organism will have to learn
anew the behavior (square rather than sine wave action variations)
that reproduce the consequence because the newly appropriate behavior
(square wave action variations) was not previously reinforced. The
PCT prediction, on the other hand, is that the organism will continue
to reproduce the consequence, using a never before performed (and,
thus, never before reinforced) set of action variations, without
difficulty.

What do you think?

Here's what I think. The behaviorist would say that the organism learned to
make whatever motion of the arm was necessary to track the target. This was
"the behavior" that was reinforced. So changing the disturbance did _not_
require the organism to learn a "new" behavior.

BG

[Richard_Marken2]

[From Rick Marken (2000.11.23.1020)]

Bruce Gregory (2000.1123.1245)

Here's what I think. The behaviorist would say that the organism
learned to make whatever motion of the arm was necessary to track
the target. This was "the behavior" that was reinforced. So
changing the disturbance did _not_ require the organism to learn
a "new" behavior.

I agree. But, as I said, I'm not developing the experiment for
the sake of committed conventional psychologists. I'm developing
it for the sake of those who are willing to learn (a small
audience, I admit, but a precious one). I would like to get
comments on the experiment from that perspective. That is,
will the results of the experiment clearly distinguish the
reinforcement from the control theory view of learning?

Best

Rick

···

---

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: marken@mindreadings.com
mindreadings.com

[Bruce_Gregory8]

[From Bruce Gregory (2000.1123.1436)]

[From Rick Marken (2000.11.23.1020)]

I agree. But, as I said, I'm not developing the experiment for
the sake of committed conventional psychologists. I'm developing
it for the sake of those who are willing to learn (a small
audience, I admit, but a precious one). I would like to get
comments on the experiment from that perspective. That is,
will the results of the experiment clearly distinguish the
reinforcement from the control theory view of learning?

It would clearly distinguish between one model of reinforcement and the
control theory model.

BG