[From Rick Marken (931230.1430)]
From Osmo Eerola (931230 08.20 GMT)--
The dog stimulus (perception) in MY EXAMPLE MUST exists before
anything else will happen. Okay?
If you say it, it must be so. But I cannot see how this fits into a control
theory model of "dog avoidance behavior".
That perception (or a futher manipulated function of it) forms a new
reference value for another (lower level) control loop of my actions
(part of my holistic survival control system) causing a kind of step
response. This response will happen as a result of seeing the dog! Okay?
Well, no. This is just a bunch of verbal handwaving. How can the
perception of a dog become a reference signal? A reference signal
is the value of a variable. What variable is "dog" a value of? What is
it's value; 1, 10, 1000? The reference value in a control system specifies
the intended value of a perceptual variable. What is the perceptual variable
that is specified? If you tried to draw a diagram of the control loop that
you describe verbally you would see that a LOT more needs to be
At first there was the dog then my action. This is the time sequency.
You cannot deny that.
I cannot deny that you keep saying it. In order to understand the causal
relationships in the control system that might be implied by your
description, however, you must first design the control system.
The sudden deviation of the controlled variable from it's reference level.
What did cause the sudden deviation from the reference value?
Mainly. BUT NOT EXCLUSIVELY, the disturbance to the controlled variable,
p, which I take to be a perception of "distance from angry dogs". The value
of this variable depends (continuously) on 1) environmental disturbances, d,
which is the location of angry dogs in the environment and 2) the outputs of
the control system itself, o, such as walking and turning. So the value of the
controlled variable depends, at any instant, on both the location of any
angry dogs in the environment (d) and what the system itself is doing (o);
p = f(d + o). I assume that this is a "pull only" control system so that
the reference signal,r, specifies the _mimimum_ value of p; if p is
greater than r then there is no output (the dog is farther than the reference
distance); if p is less than r then there is error. When there is no angry dog
in the environment then d is very large and p will also be large, regardless
of the value of o (what the system is doing in the environment). If a
dog suddenly jumps into the enviroment, d takes on some value which
depends on the location of the dog relative to the system; the perceived
distance to the dog also depends on the current outputs of the system. So
a sudden deviation of p from the reference value depends on BOTH d (the
location of the angry dog) and o (which way the system is moving relative
to that location). So d (whose value suddenly changes when an angry dog
is introduced into the environment) is NOT the cause of the actions that
bring p back to the reference. It is p that causes these actions (actually
it is the difference between p and the current setting of r that is the
cause of those actions. Note also that the systems own actions (o) are part
of the cause of those very actions.
So, o = g(r-p) and p = o+d. Thus, o (the "running away" actions), are caused
by r-(o+d), NOT by d (the sudden presence of the angry dog).
Perhaps a simpler way to look at it is like this; the dog is not the
"cause" of "running away" because the perception of the dog is NOT
independent of what one is doing. What we perceive at any instant is
always a function of what we happen to be doing at that instant. If our
actions had not taken us to that location, we would not have perceived
the angry dog. p = f(d+o) and o = g(r-p). Cause and effect are not that
easy to identify in a closed loop, since any cause is, at the same time
an effect and any effect is, at the same time, a cause -- continuously.
1. What is behavior?
In my own (bio)physical terms as I see it:
behaviour is muscle work (W=Fs), excluding in most cases at least
the unvoluntarily controlled muscles (e.g. heart).
Any higher level description of behaviour can be reduced
back to muscle work.
In fact, higher level descriptions of behavior CANNOT be reduced
back to muscle work. Take a behavior like "driving to work". The
muscle forces that produced this result yesterday will NOT produce
that result today or the next day. That's because behaviors are the
COMBINED result of muscle actions and environmental disturbances.
Environmental disturbances (like traffic, road conditions, even the
capabilities of the muscles themselves) are unpredictably variable. In
order to produce a consistent "higher level" behavioral result, muscle
work must VARY appropriately to compensate for changing disturbances.
Most people refer to the consistent results of variable muscle work as
"behavior". In PCT we recognize that these behaviors are _controlled_
results of mucle work. PCT explains how this phenomenon (the
production of controlled results) occurs; it shows that what is controlled
is not the result itself but a perceptual representation of the result.
So in PCT, we define behavior are "controlled perceptual variables".
2. What is control?
Again, as I see it:
Getting the desired function (muscle work) to be done in a proper
(intended) way by comparing the measured output
(or its representation) to the reference.
Control means ACTIVELY keeping a variable in a predetermined state.
It is an ACTIVE process because control only exists if there is a system
that is generating outputs that oppose the effects of disturbances to the
controlled variable. The opposition must be nearly equal, opposite and
simultaneous with the effects of disturbance in order for the system
generating the outputs to count as a control system (and for the results
of its actions to count as control).
3. How does one determine what a living system is doing
at any particular time (ie. how does one determine it's
In wide sense adapting to its environment in an optimal way
by functioning in the intended way.
The correct answer is very clear and simple once you know the answer
to the two questions above. You determine what a living system is
doing by determining what variable(s) it is controlling. The main goal
of PCT is to determine what living systems are doing -- and how they
do it. This can be done in a completely objective, non-arbitrary way
using "the test for the controlled variable". This approach to understanding
living systems has NEVER been used (systematically) in the behavioral
sciences; this is what is most revolutionary about PCT. PCT points the be-
havioral sciences in a completely new direction -- one that most behavioral
scientists (and control engineers) don't even know exists.
4. Is there a difference between intentional and accidental
behavior?If so, what is the difference (in CT terms)? How
does one distinguish between intentional and accidental
behavior (if there is a difference)?
The correct answer is: YES. Intentional behavior refers to controlled
variables; unintentional behavior refers to uncontrolled variables.
It is typically not possible to tell the difference between controlled
and uncontrolled variables simply by looking; one tells by doing
"the test for controlled variable". This test is demonstrated in my
"mind reading" program (which I thought would put an end to
anyone's belief in conventional behavioral science as soon as they
experienced it: but no, the power of faith transcends the power of
reasoned observation -- alas, experience with the program has not
produced a single convert to PCT). In the "mind reading" program
there are five numbers on the screen; all are influenced by the
movement of the mouse controller. The subject picks any one
of the five numbers and traces out a two dimensional pattern on
a computer screen. Of course, all five numbers trace out patterns
due to the mouse movements, but only one pattern (made with one
of the numbers) is being produced intentionally. An observer cannot
tell, by looking at the movements of the mouse or the movements of
the numbers on the screen, which number is is being moved
intentionally. But the computer can tell -- it can read the subject's
mind and determine his or her intention. The computer indicates the
intentionally moved number by bolding it on the screen. If the
subject now switches (mentally) to a new number and starts moving
it intentionally, the computer picks up the switch. The computer can
tell, at any time, which of the five numbers is being moved
intentionally. Movement of the remaining four numbers is, thus, an
accidental side effect of actions used to control the intentionally
No one looking at the movement of the numbers could tell which one
was being moved intentionally. The computer can tell, however,
because it is continuously doing the "test for the controlled variable"
on all five numbers. The test involves applying disturbances
continuously to the position of all five numbers and watching
for LACK OF EFFECT.
The "mind reading" program illustrates two important principles of
PCT: 1) you cannot tell what a person is doing by just looking --
"behavior", contrary to the most basic assumption of the behavioral
sciences, is NOT an objective phenomenon and 2) the basic goal
of the PCT approach to understanding behavior is to determine
the difference between intentional and unintentional results of
Got a good humor for a happy new year?
I'll be in a MUCH better humor next year if people would LEARN PCT
before EXPLAINING it.