[From Bill Powers (941214.1030 MST)]
Once again I'm piggy-backing off someone else's citation of a post that
hasn't got here yet ( 16 hours after its time-stamp), in this case Gary
Cziko citing Bruce Abbott:
Bruce Abbott (941213.1820 EST) said:
I'm afraid I'm just too used to thinking in terms of operants--
behaviors defined by their common (perceptual) consequences. While it
is true that every lever-press or string- pull is unique, the fact
?that we can perceive and describe them as lever- presses and string-p
ulls means that there is some common denominator by which we recognize
the essential "sameness" of these acts. I took it as implicit that
these perceptions are repeatable because the behavioral output will
vary to compensate for disturbances.
This way of defining behavior avoids the central question that PCT
answers: how is it that different actions are produced in such a way as
to have the same consequence? The concept of the operant merely
acknowledges that this problem exists. That would be all right if
matters were left there: if we merely said that regular consequences can
be produced by a variety of different behaviors. That is a true
statement supportable by observation.
But when explanations are offered, saying that the consequence selects
the behavior and thus explains why that behavior occurs, the question-
begging is shown up for what it is. The assertion is not just that one
of the behaviors that can produce a given consequence has occurred, but
that the consequence somehow selected a behavior in that class. Since on
different occasions we can find both a given behavior and its opposite
or even its absence resulting in the same consequence, this explanation
is clearly without substance, for we can say that the consequence
selects both a given behavior and its opposite (or no behavior at all,
if the consequence occurs spontaneously). This gives the consequence the
magical property of being able to create the particular behavior that is
required to produce it, whatever the amount, direction, and nature of
that behavior may be in a given circumstance.
The obvious way out of this dilemma is to propose that there are
discriminative stimuli that modify the effect of the consequence on the
behavior, so that the same consequence can produce the particular
behavior required in a given circumstance: the discriminative stimulus
represents the circumstance. This leads to a picture in which each
different behavior that produces the same consequence must be signalled
by a discriminative stimulus that indicates which specific behavior is
now required. So if a force of 1 newton is required to push a door open
when a green light is on, and a force of 2 newtons is required when a
red light is on, the appropriate amount of behavior will be selected
according to which light is now on, and which light was on when the door
opened on many previous occasions.
On the face of it this is a plausible model. If the organism actually
worked in the way supposed, we would see the observed behavior.
But suppose that in fact the organism is a control system which has a
reference signal specifying the perception of an open door. In this case
we would see (by running the model) that the amount of force applied
would vary automatically, becoming just enough to make the door be open.
The presence of red and green lights would be superfluous. However, we
would still observe that when the green light is on, the animal presses
with a force of 1 newton, and when the red light is on, with a force of
2 newtons. It would appear that the discriminative-stimulus and
selection-by-consequence explanation is working. In this case, however,
we would know that that explanation is wrong (assuming we have verified
in some other way that a control system is actually present).
The "essential sameness" of different acts that create the same
consequence lies _only_ in the fact that they produce the same
consequence. The acts themselves are unequivocally and measurably
different, often radically different. In an auction-room, raising a
card, scratching an ear, and uttering the words "two hundred seventy
five" all have the consequence of raising the bid. But in themselves
they have nothing in common that can explain their common effect. The
effect of a given act is determined by the physical state of that act,
and equally by the state of the rest of the environment that also
contributes to the final state of the consequence-variable.
In the absence of control theory, there is no simple way to explain the
operant, the behavior defined in terms of its consequences. The only
solution is to suppose that somehow, despite all the differences we can
see, all the different acts that lead to the same consequence must have
something in common, in themselves. But when we look at those behaviors,
all we can actually see are the differences. We can only imagine that in
some mysterious way, all these different acts are really the same
"behavior." And all we can really mean by that is that they produce the
same consequence.
PCT attacks this problem in a very different way. We start with as
realistic a view of the relationship between acts and consequences as
possible. We recognize that the connections between the act and the
consequence can vary, and that independent influences can also tend to
alter the state of the variable we use to characterize the consequence.
When we add up all the influences, we see that their net effect is the
observed consequence. If any of them were to change -- the influence of
the organism's acts or the environmental influences -- the consequence
would change. So if we observe that the consequence remains essentially
the same from trial to trial, and that the environment changes from
trial to trial, we must conclude that the actions must be changing in
exactly the way required to maintain a constant consequence.
We do not pretend that there is some magical property of the acts that
makes them somehow have the same consequence. We recognize that
different acts will create different consequences under constant
external conditions and we also recognize that different external
influences will create different consequences even if the acts are
unvarying. And we deduce that if the consequences are constant and the
external influences are varying, then the acts must be varying in a
precisely determined way to offset the variations in the external
influences.
This is the phenomenon that any theory of behavior has to explain.
The Law of Effect offers a particular explanation that relies on the
presence of discriminative stimuli that tell the organism when to use
one consequence-selected act and when to use a different one. In most
cases this requires assuming that if the consequence is observed to
repeat and the behaviors that produce it are different, the necessary
discriminative stimuli MUST HAVE BEEN PRESENT, and furthermore MUST HAVE
BEEN PERCEIVED. This assumption, however, only restates belief in the
theory, because the theory must be assumed to be true in order to
justify these assumptions.
In most cases of behavior, such as the cats in the puzzle box, we are
unable to say what the discriminative stimuli are; there is no objective
way to observe them or verify their sensory effects. We deduce their
presence from the fact that the cat finds an act that releases it from
the box, and from the assumption that the Law of Effect holds true. The
inability to verify the effect of the discriminative stimuli greatly
weakens the theory, because the critical assumptions on which it rests
are unverifiable. We are left with a plausible theory, but one that
depends on circular assumptions.
The Law of Effect model can, however, easily be disproven. If we
introduce an environmental effect on the consequence that is not
accompanied by any relevant discriminative stimulus (as in our tracking
experiments), different behaviors are seen to produce a constant
consequence, but the occurrance of those behaviors can't be associated
with any discriminative stimulus. According to the Law of Effect, it
should no longer be possible for the organism to maintain a constant
consequence by producing the necessary variable behavior. Since the
consequence is constant, it should be selecting the same behavior, but
it does not. And without a discriminative stimulus, there is no basis
for the same consequence to select different behaviors.
To disprove the Law of Effect in a specific situation is not necessarily
to disprove it for all situations. But the same PCT explanation that
accounts for the behavior in the absence of discriminative stimuli can
account for the behavior observed in most other situations where no
discriminative stimulus has actually been identified. So unless there is
specific evidence that a discriminative stimulus is present and
perceived, and that it is essential to creating the different behaviors
required to produce a constant consequence, the control model is to be
preferred.
Before the PCT model can be compared honestly with the LOE model, the
PCT model's properties must become as familiar as those of the LOE
model. I suggest again that we turn to some study of the PCT model and
see just how it works. I will work up some Pascal code in the next day
or two to get started, using a mouse to give the user the ability to
change variables "live" and get a feel for their effects.
···
--------------------------------------------------------------------
Best,
Bill P.