[From Bill Powers (950716.1120 MDT)]
Bruce Abbott (950716.1145 EST) addressing Bill Leach --
Bill L.:
[I think that you will find that PCT does NOT assert that "these" CAN
ALWAYS be given mechanistic, perfectly deterministic ...
interpretations]
Bruce:
So in PCT behavior sometimes just comes out of nowhere? (;->
Please explain what you meant here. Surely you can't mean that the
organism's actions are not to be explained within PCT in purely
mechanistic terms.
Yes, behavior can come from nowhere, and yes, actions are explained in
purely mechanistic terms. This is not a contradiction: it shows that
"mechanistic" does not mean "caused from outside the system."
Given a few capacitors and resistors and one operational amplifier, I
can solder together a physical system which will start behaving in a
highly organized way as soon as a power source is connected to it. No
external stimulus is required -- just an undifferentiated supply of
energy with some minimum capacity. The circuit is called a oscillator.
While the sine-wave (or other) oscillations may be modified by external
signals, the external signals are not the cause of the oscillations. The
physical organization of the system causes them. This is a deterministic
physical system whose behavior comes from nowhere, in terms or prior
causative events. Such systems are commonplace in electronics and
computing.
A control system is the same sort of system. Control does not happen
because there is some stimulus that causes control to occur. Control
occurs spontaneously when a physical system with the right physical
properties is connected to an environment with the right physical
properties. This is exactly analogous to the oscillator: behavioral
actions will be produced that control the input variable simply because
of the parameters and connectivity of the system. Control is an
INTRINSIC PROPERTY of any system having the right organization and
connected in the right way to its environment.
What Bill Leach is talking about is reorganization, but I think this
concept of a spontaneously controlling system is even more basic.
When you bring reorganization into the picture, we get even more
autonomy. Reorganization alters the physical parameters of a hierarchy
of control systems. This reorganization is done to maintain certain
basic variables inside the organism near built-in reference levels. The
specific organizations produced by reorganization are unpredictable; all
that is predictable is that reorganization will not stop until all
critical internal variables are under control.
As reorganization takes place, the control actions of the organism
change, so that new aspects of the environment are brought under
control. Always, the criterion for changing organization is the benefit
to the organism in terms of intrinsic controlled variables. All that the
environment does is to determine what consequences follow from a given
behavior; the environment does not determine that any particular
consequence will be of interest to the organism, what amount of that
consequence will be maintained by behavior, or that that consequence
rather than another will be controlled.
So in terms of events in the current external world, both behavior and
learning are entirely spontaneous.
···
------------------------------------------
The randomness you describe refers to the fact that one's history
of experience (and the sum total of its "imprint" on the nervous
system) is largely a matter of circumstance. A major job of the
nervous system is to sort out those relationships which are
repeatable from those which are merely accidental.
On the contrary, the randomness of reorganization is independent of
history, and that is what makes it such a powerful method. It works, I
have proposed, like the E. coli method works: random changes in output
which are applied at intervals that depend on the results relative to
the results that the system wants.
------------------------------------------
... EAB is just terribly concerned about the scientific status of
that word "wants." What does it mean to "want" something?
As you know, it means that there is a difference between the perception
being controlled and the reference level for that perception. This is
the physical explanation for "wanting." The word is used rather loosely;
you can say "I want some ice-cream," meaning to refer to the state of
the controlled variable (I don't have any ice cream), the nature of the
goal (the actual amount wanted), or the error signal (the degree of lack
of ice cream in relation to the amount wanted).
It isn't surprising that EAB has a problem with "want" or other terms
that are specific to control phenomena. EAB contains nothing that
corresponds to a reference signal or a comparator. It's assumed that
when people use terms like wanting, desiring, intending, and so forth,
they are talking about vague mentalistic concepts. In PCT we don't have
that problem.
If I ask why you wrote the above sentence, would a satisfactory
answer be that you did it because you "wanted to"?
Yes. That would be a correct answer at the level of function directly
addressed by the question. That answer says that I wrote the sentence
not accidentally in the course of doing something else, but because I
intended to write exactly that sentence. There was, in other words, a
set of reference signals that was satisfied by the act of writing the
sentence: the reference signals probably specified the various elements
of the sentence that were to be created in present-time perception.
You can also ask WHY I wanted to write that sentence. Now the question
addresses a higher-level goal; generating the specific reference signals
that led to writing that sentence is now to be seen as the output of a
higher system controlling something else by means of writing that
sentence. Perhaps writing the sentence was being used to answer a
question, in which case the correct answer would be "because I wanted to
answer the question."
Hardly! This is no explanation at all, because now you have to
ask, "but why did you want to?"
Exactly. By pursuing that line of reasoning, you end up with a
hierarchical control model. When you actually do this, you find that the
spectre of infinite regress which has kept others from following the
same line of investigation never appears. There is a finite number of
levels. When you get to the top level, there are no phenomena left to
explain. If you do find another one, you simply add one more level.
Most people can answer "why" about three or four times, after which the
answer is "I don't know."
The term "food" and the term "reinforcer" are not even close to
being isomorphic. This is like claiming that "gasoline" and
"power" are the same, and so we can dispense with the concept of
power.
I'm surprised at this admission. A "reinforcer" is supposed to be
something observable. But if you admit that "food" and "reinforcer" are
two different things, then you will have a hard time showing me any
reinforcers in an operant-conditioning experiment. You can easily show
me the food, or water, or whatever, but showing me the reinforcer is, to
say the least, difficult.
-------------------------------------
RE: deprivation
Bill L:
Doesn't anyone in EAB wonder why you can not train rats without first
creating a deprivation condition?
Bruce:
Bill, I've explained reinforcement theory's view of this already,
so I'm more than a bit surprised that you're asking this question.
Deprivation is viewed as an "establishing operation" that
influences the effectiveness of food as a reinforcer.
Yes, but this doesn't explain why you have to deprive a rat of food in
order to make food effective as a reinforcer.
It is assumed that the reason deprivation has such an effect on the
reinforcing value of food is that deprivation creates a need which
the food tends to correct.
But why does food correct ONLY a food deprivation? This is not self-
evident, unless you make certain tacit assumptions, such as the
assumption that a "need" is the difference between the obtained amount
of something and the amount that is desired. If "need" is just an
abstract concept, then depriving a rat of water would create a need-
state, and there is no a priori reason why food should not eliminate it
if food ever eliminates a need state. What is still missing is a
statement of why the need is so specific to the deprivation, and why
only an input that reduces THAT need will be effective.
PCT, of course, explicitly addresses this question.
But from the strictly Skinnerian point of view, one can use the
empirical fact that level of deprivation affects rate of responding
for food pellets without waiting for the underlying biological
reasons for that linkage to be elucidated. This is not to say that
these reasons are irrelevant, only that the scientific study of
behavior need wait for them to be discovered.
If Skinner had been a modeler he would have discovered the problem with
this purely empirical approach. Our investigations of the Motheral curve
are starting to show one of them. If you look at the region near the
origin, where increasing reinforcement goes with increasing behavior,
you find that behavior increases with _decreasing_ deprivation, not
increasing deprivation. Over on the right side, decreasing deprivation
goes with decreasing behavior -- but _more_ reinforcement.
I'm all for the empirical approach, but you don't get explanations from
pure empiricism. Skinner never seemed to put together all the facts at
his disposal into a single picture of behavior. He just looked at little
part-relationships. His tunnel vision kept him from seeing the
contradictions he was discovering. He concluded triumphantly that
intermittent reinforcement is more effective in producing behavior than
continuous reinforcement -- and at the same time believed, presumably
because of other data, that more reinforcement should produce more
behavior. His use of verbal classifications (intermittent vs.
continuous) kept him from seeing the quantitative contradictions (less
reinforcement and more behavior on FR-10 than on FR-1).
Bill L:
It is not the nutrient and it is not even the environmental
circumstances but rather the organism's internal reference level for
nutrients that control behaviour. This is so obvious that it hurts.
Bruce:
Actually, there's a good argument to be made that nutrient levels
remain rather steady under ordinary (nondeprivation) conditions and
that under these zero-error conditions variables other than
nutrient level are far more important.
These are not zero-error conditions. At zero error there is no behavior.
Obtained nurtient levels are maintained constant only by the behavior
that is going on, and that behavior is driven only by the small
remaining error.
If other factors become more important (for what?), then they can't be
causing any change in the obtained nutrient levels if there is no effect
on nutient-obtaining behavior. If you introduce other factors like
taste, you can create a conflict, as by loading the water supply with
quinine. You bring a second control system into play which uses the same
behavior but has a goal for a different aspect of the same physical
input. If the behavior lessens because of the bad taste, then the animal
gets less water and the nutrient level is affected (counting water as a
nutrient, but you could use food if that bothers you). So I don't
understand what you're talking about.
Environmental circumstances determine what actions will lead to
what changes in the organism's controlled variables; experience
with those circumstances determines what among several possible
actions will actually take place. If I am currently experiencing
an error in some controlled perception, what will I do? Your
statement suggests that the answer, given only this information, is
obvious. Obviously it is not.
Not to an EABer. To a PCTer the answer is that the animal will either
produce more behavior to bring the controlled perception near its
reference level, or start reorganizing until it finds a behavior that
will have the right result. If there are several behaviors that could
bring the controlled perception to its reference level, we can't predict
which of them will be used, but we can predict that most of the time one
of them will be used. If only one behavior can produce the desired
result, than that behavior will be used, if the animal can find it and
if it can produce the required amount of output.
How you reach these conclusions is beyond my understanding. Where
is your evidence that "when the experimental data does [do] not
match what is expected then the results are dismissed as failing to
account for the effects of "previous reinforcers" or "satiation"?
I don't know of anyone in EAB who "dismisses" contradictory results
in the way you describe.
Then why, in talking of using operant conditioning in the classroom, did
you explain that sometimes it doesn't work because of the previous
history of reinforcement or because some more effective reinforcer must
have been present? It seems to me that this is doing exactly what Bill
described.
In a research report, it is considered to be the researcher's
obligation to note conflicts between a study's outcome and previous
findings or the theory being evaluated within the study. An author
may then speculate on factors or interpretations which could
"account" for the discrepancy, not to "dismiss" the contradiction,
but to suggest directions for future research.
That sounds wonderful, but how many of these authors have speculated
that perhaps reinforcement does NOT maintain behavior, or that the
probability of behavior does NOT increase with reinforcement, or that
reinforcement may NOT be an actual phenomenon? The "directions for
future research" are uniformly cast in terms of possibilities that leave
the basic theoretical structure intact.
Furthermore, if research is conducted to evaluate the effect of
such factors as satiation or previous history of reinforcement and
this research supports the explanation, this is hardly what I would
term "dismissing" the results.
Well, I doubt that anyone can write a complete "history of
reinforcement" even for a laboratory animal reared from birth under the
experimenter's control. Reinforcement is going on 24 hours per day and
no experimenter could possibly notice every possible reinforcer in
action. "History of reinforcement" remains a fiction.
Well, I didn't mean to pre-empt Bill's reply altogether, so I'll quit.
-----------------------------------------------------------------------
Best to all,
Bill P.