The Snark is a Boojum

[From Rick Marken (970901.1040)]

Bruce Abbott (970901.1215 EST)

The explanations I am offering are control explanations for
empirical phenomena which have been given certain labels in EAB.
I am not offering PCT as an explanation of reinforcement
_theory_, but of certain phenomena that reinforcement theory
is supposed to explain. I am sorry that you are having such a
hard time telling the difference.

You are offering control theory explanations for empirical
phenomena that DON'T EXIST. There is no such thing as the
phenomenon of reinforcement. It doesn't happen. That's
what Bill has been showing you over and over again. This
is not a theoretical point. It's a fact.

Reinforcement is _not_ a phenomenon. Reinforcers don't do what
you say they do; they don't strengthen (increase the frequency
of) behavior. Trying to explain this "fact" with PCT is as
ridiculous as trying to explain the movement of the sun around
the earth using Copernican theory. The phenomenon (sun moving
around the earth) doesn't happen; it's an _appearance_ that is
seen only under special circumstances, like when you look at
the sun from one of its rotating satellites. The same is true
of reinforcement; it is an _appearance_ that is seen only
under spacial circumstances, like when you look for a change
in the _particular_ behavior that a hungry animal, placed in
a cage, can suddenly use to feed itself.

I know this is tough to hear, Bruce, but, trust me. It's for
your own good. I don't want to see you softly and suddenly
vanish away, for the snark of reinforcement is a boojum you see.

Best

Rick

···

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

[From Bruce Abbott (970902.0920 EST)]

Rick Marken (970901.1040) --

Bruce Abbott (970901.1215 EST)

The explanations I am offering are control explanations for
empirical phenomena which have been given certain labels in EAB.
I am not offering PCT as an explanation of reinforcement
_theory_, but of certain phenomena that reinforcement theory
is supposed to explain. I am sorry that you are having such a
hard time telling the difference.

You are offering control theory explanations for empirical
phenomena that DON'T EXIST. There is no such thing as the
phenomenon of reinforcement. It doesn't happen. That's
what Bill has been showing you over and over again. This
is not a theoretical point. It's a fact.

Rick, you are allowing yourself to be confused by terminology. What I am
calling the phenomenon of reinforcement is a real phenomenon: response rates
_are_ higher during the contingency phase of the experiment I described, and
reduce to baseline levels when the contingency is withdrawn. The fact that
this observation is called "reinforcement" is unfortunate, because that name
also suggests an explanation. But I am here restricting that term to the
observations to be explained; the theoretical concept called reinforcement
is something else entirely.

Reinforcement is _not_ a phenomenon. Reinforcers don't do what
you say they do; they don't strengthen (increase the frequency
of) behavior.

I have not said that reinforcers strengthen behavior. That is theory.

Trying to explain this "fact" with PCT is as

ridiculous as trying to explain the movement of the sun around
the earth using Copernican theory. The phenomenon (sun moving
around the earth) doesn't happen; it's an _appearance_ that is
seen only under special circumstances, like when you look at
the sun from one of its rotating satellites. The same is true
of reinforcement; it is an _appearance_ that is seen only
under spacial circumstances, like when you look for a change
in the _particular_ behavior that a hungry animal, placed in
a cage, can suddenly use to feed itself.

No, you have the analogy wrong. It is like using Copernican theory (rather
than Ptolemy's) to explain the apparent retrograde motion of the planets. I
am arguing that Copernical theory explains this retrograde motion (as it is
observed in the heavens against the fixed stars) and you are arguing that
retrograde motion does not exist, meaning an actual physical reversal of the
planet's motion. That is, you are confusing my reference to the observed
phenomenon (appearance of retrograde motion) with an explanation of that
observation (the planets are actually reversing course).

Regards,

Bruce

[From Bruce Gregory (970902.1650 EDT)]

Rick Marken (970901.1040)]

You are offering control theory explanations for empirical
phenomena that DON'T EXIST. There is no such thing as the
phenomenon of reinforcement. It doesn't happen. That's
what Bill has been showing you over and over again. This
is not a theoretical point. It's a fact.

I hate to agree with the "evil" Bruce, but I see a
terminological problem here. It might be well to avoid talking
about whether things exist or not. Our models tell us what
exists. Reinforcement is part of the Skinnerian model and
not part of a PCT model. Arguments over the existence of
the features of models tend to become theological, I
fear. It seems to me that the _only_ question we can settle
empirically is which model is more useful in making sense of the
world. No contest here as far as I am concerned.

The "good" Bruce

[From Rick Marken (970903.0810)]

Bruce Gregory (970902.1650 EDT)

I hate to agree with the "evil" Bruce

It happens.

Our models tell us what exists. Reinforcement is part
of the Skinnerian model and not part of a PCT model.

To some extent, this is true. But our experience also tells us
what exists. And our experience tells us that the phenmenon of
reinforcement (an increase in behavior associated with an
increase in reinforcer) only occurs in one special circumstance:
when an animal in an operant experiment first discovers the
means of feeding itself. The animal goes from never using that
means of feeding itself (and never getting a reinforcer) to
frequently using that means of feeding itself (and getting many
reinforcers). From then on, however, the relationship between
behavior and reinforcer is negative; an increase in reinforcer
is associated with a _decrease_ in behavior.

We don't need a model to see that there is no such thing as
reinforcement; and we don't need a model to see that there
is such a thing as control. Once the animal has discovered
the means of influening food input it can be shown that the
animal is controlling a perceptual aspect of that input. We
can demonstrate control by showing that disturbances to the
controlled variable (rate of food input,say), such as changes
in the number of presses required to produce food or addition
of noncontingent food pellects, are resisted; the animal
changes its influence on food input so as to protect rate of
food delivery from disturbance.

I think behaviorists have imagined the nonexistant fact of
reinforcment (and failed to notice the actual fact of control)
because they have made their observations though the lens of
a (usually unstated) model that places the derminants of
behavior outside of the organism. If external events determine
what organisms do then you certainly don't want those events
doing nothing but surpressing behavior. You'll end up with an
organism that does nothing. No, what you want is external events
that strengthen or "select" certain behaviors. Yes, that would
explain why organisms do some things (like press bars) and not
others. So you imagine that consequences _reinforce_ (strengthen)
behavior and you find justification for this imagined relationship
between consequences and behavior in the fact that animals will
go from none of a particular behavior to a lot of it if this
behavior produces certain consequences ("reinforcers"). This
one observation has justified the model driven assumption that
consequences select behavior. Unfortunately, all the other facts
show that this model is wrong.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

[From Bruce Gregory (970903.1300 EDT)]

Rick Marken (970903.0810)]

Bruce Gregory (970902.1650 EDT)

> Our models tell us what exists. Reinforcement is part
> of the Skinnerian model and not part of a PCT model.

To some extent, this is true. But our experience also tells us
what exists. And our experience tells us that the phenmenon of
reinforcement (an increase in behavior associated with an
increase in reinforcer) only occurs in one special circumstance:
when an animal in an operant experiment first discovers the
means of feeding itself. The animal goes from never using that
means of feeding itself (and never getting a reinforcer) to
frequently using that means of feeding itself (and getting many
reinforcers). From then on, however, the relationship between
behavior and reinforcer is negative; an increase in reinforcer
is associated with a _decrease_ in behavior.

With your permission, I will worry this bone a bit more. I would
describe what you say in the following way. From our
observations of animals in operant experiments we observe that
when the animal discovers the means of feeding itself it uses
the method frequently. One possible interpretation of this
phenomenon is that the access to the pellets "causes" the animal
to press the level (a model). How well does this model work? Not
very well, as you point out. This means the model must be
modified to account for the new data.

We don't need a model to see that there is no such thing as
reinforcement; and we don't need a model to see that there
is such a thing as control.

If this is true, why doesn't everyone recognise the existence
and importance of control? I think that one begins to observe
the ubiquity of control after one reads B:CP or MR. It is the
utility and applicability of the model that makes the
difference. It is the model that transforms the experience. I
see the same world I always saw, only now I see control wherever
I look.

Once the animal has discovered
the means of influening food input it can be shown that the
animal is controlling a perceptual aspect of that input. We
can demonstrate control by showing that disturbances to the
controlled variable (rate of food input,say), such as changes
in the number of presses required to produce food or addition
of noncontingent food pellects, are resisted; the animal
changes its influence on food input so as to protect rate of
food delivery from disturbance.

Yes, but as an economist whose name escapes me said, "Without a
theory, the facts are silent."

I think behaviorists have imagined the nonexistant fact of
reinforcment (and failed to notice the actual fact of control)
because they have made their observations though the lens of
a (usually unstated) model that places the derminants of
behavior outside of the organism. If external events determine
what organisms do then you certainly don't want those events
doing nothing but surpressing behavior. You'll end up with an
organism that does nothing. No, what you want is external events
that strengthen or "select" certain behaviors. Yes, that would
explain why organisms do some things (like press bars) and not
others. So you imagine that consequences _reinforce_ (strengthen)
behavior and you find justification for this imagined relationship
between consequences and behavior in the fact that animals will
go from none of a particular behavior to a lot of it if this
behavior produces certain consequences ("reinforcers"). This
one observation has justified the model driven assumption that
consequences select behavior. Unfortunately, all the other facts
show that this model is wrong.

Rather than wrong, I would say woefully limited. It requires so
much juggling and adjustment to fit the data that the model
looses much of its charm ;-. (Someone once said, "With seven
free parameters, I can fit the Manhattan skyline..." By the way,
that's the problem with the "standard" theory of leptons and
quarks -- it has, count 'em, _seventeen_ free paramters!)

I appreciate the advantages of saying that an inadequate or
limited theory is wrong, but I know that in physics we
successfully use theories that are "wrong" by confining our uses
to domains where the theories work. (Newtonian physics is still
incredibly useful even though it breaks down when relative
velocities approach the speed of light and where gravitation
fields are extremely strong.)

Bruce

[From Rick Marken (941029.1245)]

Bill Leach (941028.19:52 EST) --

Gotcha! I see the points. Such remarks are likely a "sign" that the
researcher is vaguely aware of "something wrong" but not quite what.
It is almost amazing how few words need to be added or changed
though to make it come out right.

Yes. It is amazing. For the sake of lurkers, I would like to try to make
this point a bit more concrete. The quote we are discussing suggested
that the goal of behavioral research is to determine the relationship
between stimuli and responses. Such research seems close to The Test
because we are applying a disturbance (the stimulus) and watching for
a response. For example, the stimulus might be your hand moving
toward the subject's face; the response (if one occurs) might be
movement of the subject's head. If you do this study, you will find that
there is a relationship between movement of your hand and movement of
the subject's head. This is the kind of psychological fact that is
very common in the behavioral sciences -- though it is usually expressed
statistically: "on the average, the subject's head moves this much when
a hand approaches the face at that rate.

This kind of research is, indeed, close to The Test (especially when it is
done one subject at a time) but (as I noted) it leaves out one crucial step:
the hypothesis about the variable being controlled. The hypothesis
about the controlled variable really states _why_ you expect to see a
relationship between the stimulus (disturbance) and the response. In
this case, there are several possible hypotheses: the subject might be
controlling the distance between his or her face and objects in the
environment. Or the subject might be controlling the amount of pain
he or she experiences (remember, the controlled variable is really a
perception). (These particulr hypotheses are not mutually exclusive;
the subject might be controlling vayriables both and there may be a
hierarchical relationship between the control systems).

The hypothesis about the controlled variable is the central focus of The
Test; disturbances (stimuli) are applied to see if they actually have an
effect on the controlled variable. And it is usually best to monitor the
relationship between disturbance and controlled variable rather than
that between disturbance and response (as in conventional research). For
example, suppose the subject is controlling for pain and suppose also
that he or she is standing against a wall so that there is no room for the
head to move back. In this case, if you take a swing at the subject's face,
instead of the head going back, the hands go up to protect the face from
the blow. The subject feels less pain than if he or she had done nothing
so the disturbance (the swing) has little effect on the controlled
variable. An S-R psychologist, who looks only at the relationship
between disturbance and response, would count this result as a
statistical deviation; the subject usually responds by moving the head back
but this time there was no head movement. The fact that variable means
(responses) must be used to produce a consistent result (such as keeping the
pain variable near zero) is the reason why most behavioral science research
now relies almost exclusively on statistics; the addiction to statisitics is
a result of ignoring controlled variables.

Conventional psychologists, looking at S-R relationships, have noticed
that responses to stimuli tend to produce results that could be
interpreted as "what the subject wants". For example, the reflexive
movement that occurs when you touch a hot stove produces the happy
result of keeping your flesh from burning. Indeed, it has been noticed
that "reflexive" responses seem to have the _purpose_ of producing
particular results. But without control theory, the achievement of these
purposes was seen as a lucky result of the way S-R connections are hooked up.
As we know, however, there is no way to produce control (a purposful result)
with an S-R system (Bourbon and Powers show on reason why in their "Models
and their worlds" paper in Closed Loop).

The failure of conventional psycholgists to consider the possibility of
controlled perceptual variables in their S-R research may seem like a
small omission: but it makes all the difference. Once you realize that
responses to stimuli are just a visible side effect keeping controlled
perceptual variables in reference states, then you see that conventional
psychologists have been on a Snark hunt ever since the methods of con-
ventional behavioral research were developed in the early 1900s. Eventually,
these researchers will softly and suddenly vanish away -- for the Snark is
a Boojum, you see.

Your feet dry again yet?

Yes, thanks. They're back out here havin' fun in the warm California sun.

Best

Rick