[From Bill Powers (2000.11.16.0219 MST)]
Bruce Abbott (2000.11.15.1545 EST)--
So, let's apply this to the case that started all this: the rat pressing a
lever to earn a food pellet, when the apparatus requires that the lever be
pressed with a force that falls within a narrow range or "window" of values,
initially unknown to the rat. You say that the control system "selects by
setting a reference signal." I'm a little troubled by this -- it sounds
like the control system is a little person in the head that knows what it
wants and knows how to get it.
Yes, that is what Rick, and I, and everyone else who understands PCT is
saying -- all except for the "little person" part. There is a system inside
the rat that knows what it wants (it has a positive nonzero reference level
for food intake). In the case where a fully-formed control system already
exists, the system knows how to get it (send reference signals to
lower-level control systems that perform the acts that, as a side-effect,
produce the food). In a novel situation (novel for the individual rat)
there is no existing system that will produce the food, but there is
another kind of system in the rat that can institute processes that will
construct a control organization that will produce food. Either way, is it
the rat that determines what is wanted, and the rat that reorganizes itself
and finds the means of getting it. The environment merely provides the
setting in which the rat must find the means of getting food if it doesn't
already know how to get it.
Can you be more specific? How does the
control system in question "know" what variable to control? How does it
"select" that variable for control? How does it "know" what reference value
to select for that variable? How does it "know" what actions to take to
control the variable?
These are, of course, very basic questions, with which I tried to deal in
considerable detail in B:CP (and even before that). In fact, my answers to
these questions constitute what we now call perceptual control theory,
together with the theory of reorganization. I probably did not cover every
possible situation, nor did I come up with every possible answer. But the
answers I have proposed have seemed fairly general to many people, enough
to form a starting point for the replacement of previous concepts about
what behavior is, how it works, and how behavioral organization is acquired.
Most germane to your questions is the concept of a reorganizing system.
Let's consider, once again, how it is proposed to produce new organizations
that end up providing food in a novel situation where the rat has no
previously-learned method for causing food to appear.
First, we must find out or try to guess what effect of ingesting things the
rat is trying to control. A number of variables can be postulated, from a
general notion of "nutrition" to specifics like blood glucose
concentration, electrolyte balance, level of toxins in the blood, and so
forth. The specifics are, of course what is actually controlled; general
concepts are only for the convenience of a theoretician.
First, how does the rat know that it should control these variables? That's
a rather anthropomorphic way of asking the question, for it implies that
there is some central consciousness in the rat that does what we call
"knowing." We know nothing about a rat's consciousness, if there is any,
but we do know that rats inherit certain structures and that all rats that
stay alive through their own efforts, without exception, eat when certain
variables like blood glucose get out of range. So it makes sense to propose
that there are certain very basic reference levels built into every rat as
part of its inheritance, among them being a reference level for a certain
concentration of blood glucose.
So we have a basic perception, potentially controllable (remember that in
PCT, perception does not imply anything about consciousness: a thermostat
perceives temperature in the sense used in PCT). We have an inferred
reference level for the perception.
Deviations of the perception from the reference level are detected by an
inferred comparator, which produces an error signal as usual. But now we
have a problem, because (to stick to one example) the error in blood
glucose concentration can be cured only by performing unknowable acts in an
environment of unpredictable structure created at the whim of an experimenter.
In the worst case, there is no way in which previous experience could help
in deducing what the experimenter has decided that the rat must do in order
to produce food. We have to assume that the rat has certain capabilities
such as those of moving itself around, but what it has to find out is how
and where to move that will correct the basic glucose concentration error.
This implies that reorganization is going on at a specific level in a
partly-organized system. Since we're assuming the worst case, the only
solution I can see is for the rat to produce varied actions in variable
ways and variable places until either food appears to be eaten or it
starves to death. And of course when it has found one of the actions that
produces food, it must stop the random variations and continue doing what
it was doing when the food appeared.
And what is it that tells the rat it has done the right thing? The blood
glucose concentration error diminishes. This is how we get to the E. coli
method of reorganization, which I had defined 25 years before I knew
anything about E. coli (see the 1960 papers with Clark and MacFarland).
The diminution of blood glucose error, I have since found by using
simulations, must slow the rate of reorganization, not terminate it
abruptly, if a systematic approach to a final solution is to occur. So the
naive rat is not expected to stop searching the cage and start pressing the
lever the first time a lever-press results in a food pellet's being
delivered. Rather, the rat starts dwelling longer and longer in the
vicinity of the lever, as if it's not sure what it was doing while it was
there that produced the food (another anthropomorphism as well as a
metaphor). Only after many trials does it eliminate the irrelevant actions.
So this model, after quite a few cycles of modification, now predicts more
or less correctly how a rat will behave in a Skinner box, at least in
general. And it answers one of your questions: how does the rat find out
what to do in order to get what it wants?
Notice that the rat gets no guidance from the environment to lead it to the
behavior that is needed. Of course if the needed behavior is something
bizarre like walking in a figure-8 pattern, there would be little chance of
finding it through reorganization in a single jump. However, if the
experimenter started with simple requirements such as turning left which
the animal would be likely to produce during a normal search process, the
amount of reorganization required to find it would be small. Then the
requirement could be changed so another small reorganization might come up
with the next approximation in a reasonable time, and so on through the
process known as "shaping." The need for a single large reorganization of
low probability is replaced by a series of small reorganizations of high
probability. In no case does the experimenter determine how the animal will
reorganize: that is strictly up to the animal. Shaping merely sets the
stage so that fewer reorganizations need to be done in order to
re-establish control of food intake during a given phase of shaping.
The relationship between the experimenter and the animal is one that can
easily be illustrated with the rubber bands. The experimenter varies what
does not matter to him, the organism's food intake, as a means of
controlling what does matter to him: the organism's behavior. And the
organism varies what does not matters to it, its behavior, as a means of
controlling what does matter to it, its food intake. Each party to the
relationship is controlling what matters to it, its own perception, by
varying what does not matter to it, its own behavior. This can be a
completely conflict-free interaction.
This is all standard PCT, with only minor changes from what I started
saying in print 40 years ago. I must have been explaining it very poorly,
for these questions to be arising again at this late date.
Best,
Bill P.