[From Bill Powers (941003.0800 MST)]
Bruce Abbott (941002.1630 EST)--
You guys are determined to see animalism [animism] here, and I'm
downright puzzled by it--in a way. I know your motivation: you are
trying to keep the focus on the causes of behavior that lie inside the
organism rather than "out there" in the environment. Understanding the
organism's internal organization is crucial for making proper sense out
of what relationships appear "out there" beyond the boundary separating
organism from environment. However, belief that the environment
"controls" behavior (in the EAB sense of the term) does not make one an
"animist." These folks no more believe in animism than you do.
You're right about our motivation.
In another field, a person named Klein has written that he is using the
term positive feedback to indicate when the controlled variable is
greater than the goal-state, and negative feedback to indicate when it
is less than the goal-state. In that same field, other people who use
the terms positive feedback mean "encouraging, supportive, approving,
helpful, admiring responses to a person's behavior", while negative
feedback means the opposite. In a Behavioral and Brain Sciences paper
now in press and distributed for commentary, Gray proposes a model of
the brain in which the primary function at each level is carried out by
a comparator: no input function, no output function.
The word control is often used in similar loose ways, with the added
problem that the word itself existed before control theory did.
Sometimes people use the word control just as a control-system engineer
would use it: the pilot controls the rate of descent of the airplane.
But without breaking stride, the same people can say that the shape of
an airfoil controls the amount of lift of an aircraft wing.
The pilot controls the rate of descent of the airplane by varying
factors that affect the lift of the wing: the plane's speed, the angle
of attack, the settings of flaps and spoilers, the angle of yaw (slip).
At the same time, the lift is affected by other factors such as
altitude, icing on the wings, smoothness of the airfoil surfaces, and
the shape of the airfoil. All of these factors, most of them variable,
contribute to the lift of the wing, but no control-system engineer,
while at work, would say that any of them controls the lift. The pilot
_varies_ the lift to _control_ the rate of descent of the airplane.
The rate of reinforcement contributes to behavior as many variables in a
control loop contribute to it through being affected by and affecting
other variables. An understanding of the role of reinforcement requires
understanding the organization of the control system and its
environment, including other factors beside the rate of behavior that
can influence the rate of reinforcement. All this is clear to a person
who understands the phenomenon of control and the kind of system
required to bring it about.
But a person who does not understand that kind of system will not see
what is and what is not control; in fact, behaviorists who have grown up
without realizing that negative feedback control systems exist, and
without understanding their properties, do not and cannot mean by this
word the same thing that the control engineer (or the PCTer) means. When
they think of control, they think of one variable acting as a cause with
respect to other variables. They know nothing of systems capable of
varying their outputs so as to maintain their inputs in preselected
states. Therefore when they use the word control they cannot be meaning
the actions of such systems.
When you say that reinforcement controls behavior, this is exactly like
saying that the temperature of a room controls the heat output of a
furnace, that the speed of a car determines how hard the driver will
press the accelerator pedal, that the flow of traffic controls how a
traffic cop will wave his arms, that the angle of bank of an airplane
controls how the pilot will move the control stick. In each case, it is
like saying that the variable actions which maintain the controlled
variable in a certain state are controlled by that state of the
controlled variable.
Controlled variables are affected by behavior and also by other
independent variables, disturbances. The state of a controlled variable
influences behavior, but no more than the setting of the internal
reference signal does; there is no way to predict in general what the
behavior will be for a given value of the controlled variable unless one
also knows the current states of the reference signal and external
disturbances. For a given state of the controlled variable, the behavior
can be in any state from maximum to minimum, depending on the reference
signal and the amount and direction of any disturbances that are acting
on the controlled variable.
The only observable causal relationship between behavior and
reinforcement runs from behavior, through the scheduling apparatus or
person, to the reinforcement. Reinforcement is observed to depend on
behavior. However, it does not necessarily depend exclusively on
behavior. There can be "noncontingent" increases and decreases in rate
of reinforcement that come from independent sources, not from an effect
of behavior. Yet, under conditions that make good control possible, we
find that the result of such disturbances is for the rate of
reinforcement to remain essentially constant, and the rate of behavior
to change. In fact, it is only the change in rate of behavior that makes
it possible for the rate of (obtained) reinforcement to remain nearly
constant while there are independent, noncontingent, disturbances acting
to alter the rate of reinforcement.
This situation makes it clear that the organism is varying its behavior
to keep the reinforcement rate under control at a particular level. To
say, under such circumstances, that the rate of reinforcement controls
the behavior is simply to misunderstand the situation.
All this is made very much harder to understand in relation to operant
conditioning by the way experiments are designed and carried out.
Because, in general, those doing the experiments do not understand
control, there is no constraint on those experiments to make sure that
the animal is faced with conditions that permit successful control to
continue. The animals are often required to work under conditions where
they have no control of their own body weight. They are put into
conditions where the loop gain is extremely low -- where variations in
actions have nearly no effect on the rate of reinforcement. In fact,
they are often tested under conditions which, if continued, would prove
fatal. Only the fact that the experimental run ends, and the animal is
returned to its living cage and fed back up to a survivable weight,
keeps the animals alive in many experiments.
These are not the conditions in which the animals evolved. Were the
environment suddenly to impose such severe restrictions, the species
Rattus Laboratorium would soon be extinct.
The only way to understand the underlying organization of behavior in a
rat is to study the rat under conditions where control can succeed well
enough to maintain the rat in a viable condition. When extreme
conditions are used, we are not seeing the animal's normal behavior. We
are seeing it in a state where error signals are extreme, where output
systems are driven hard against their limits, where perceptions of the
current state of affairs are masked by large random disturbances, and
where higher systems of control, such as weight control or foraging
control or mating control, are out of action completely or straining at
their utmost without being able to produce any effect.
All this comes about because behaviorists do not understand that there
is a very specific and important phenomenon behind the word "control" --
that this word means far more than "a repeatable relationship between
independent and dependent variables." I am going to hold out for using
the term control is its specific technical sense, because knowing when
and when not to use it will then be a sign as to whether someone
actually understands what PCT is all about.
ยทยทยท
-----------------------------------
And speaking of animism, what is this "selection OF consequences" I
keep hearing about? Who is doing the selecting? [No explanations,
please, it's a rhetorical question.] Sounds suspiciously like animism
to me...
It is perfectly proper to speak animistically about the behavioral
organization of a living system. Living systems contain reference
signals which specify the state to which a physical variable is to be
brought and in which it is to be maintained, as well as the control
organization required to make such reference signals effective. It is
not proper to speak of physical variables which do not contain the
required organization as if they were alive and had goals for the states
of other variables. Contrary to popular opinion, rain does not have the
purpose of ruining a picnic.
------------------------------------------------------------------------
More later -- got to watch the launch of Atlantis.
Best,
Bill P.