[From Bill Powers (2002.09.27.1609 MDT)]
Rick Marken (2002.09.26.0920)--
But I think we're getting a bit off point. This thread started when Bill
Williams
(UMKC 24 September 2002 12:33) asked the following:
> So, my question for you would be how would you justify from a control
theory
> perspective an argument that external circumstance is the cause of
behavior.
In the process of answering this question I realized that we have sometimes
characterized one of the "messages" of PCT as "Behavior is not caused by
external
circumstances". As I was constructing my reply to Bill W. I realized that
this
can be a confusing message because behavior certainly can be seen and
described ... as being caused by disturbances to controlled variables.
I think that's true of _deviations_ from normal behavior patterns, but if
you cast your mind back over a typical day, how many of your actions are
aimed only or primarily at counteracting the effects of disturbances of
static controlled variables? That's what has to be true if actions are to
be determined primarily by disturbances.
In my life, most of my actions are produced to create effects I want to
achieve, and serious disturbances are not common. Most of the variables I
control (at least as it seems to me right now) I am causing to change --
I'm not holding them in a constant condition. I pick up a glass and cup and
return them to the kitchen. I get out a map and look up Wolfgang''s
new location. I fix a peanut-butter sandwich and eat it. I pound away at
the keyboard making these letters appear. Of course there are small
independent disturbances all the time, but they aren't making me perform
all of these different actions. Controlling some variables like the peanut
butter sandwich is done to control other variables like hunger, but at the
level of making the sandwich the actions are produced primarily by varying
reference signals. My hunger can't tell me that the peanut butter is in the
pantry, or tell me how to spread peanut butter on bread.
There are always small errors that I correct with small adjustments of my
actions, but the main action is there simply to make something change the
way I want it to. If all disturbances from independent sources disappeared,
I wouldn't quit doing these things. I'm satisfying my own higher goals most
of the time, and the errors I'm correcting wouldn't be there if it weren't
for the other goals -- as well as whatever disturbances happen to be acting..
What control theory shows is that the "causal" path from environment to
behavior
runs through the environment, in particular through the controlled
variable, not
through the organism.
The causal path actually runs from the disturbance, to the controlled
variable, through the organism, and then to the output action, doesn't it?
It doesn't go backward through the environmental feedback function to the
output, bypassing the organism. It only SEEMS to do that because of the
relationship between the disturbance and the action.
I don't think we should lose sight of the literal physical effects. They go
(View this with Courier font, monotype)
r
v
d ---> qc ---> input function ----->comparator --->qo
^ |
> >
<-----feedback f <--------------------------
As you can see, the only way for a variation in d to get to qo is through
qc, the input function, and the comparator. Likewise for r: the only way
for r to affect qo is through the comparator, and the only way for it to
affect qc is through the comparator, qo, and the feedback function. It's
clear that qo is a function of both d and r (and itself), not of d alone or
r alone.
The approximations we talk about (qc = r, qd = -f(d)) are just that,
approximations made by assuming infinite gain. There is no actual effect of
d on qo going backward through the feedback function. The reference signal
does not literally determine the input quantity through the inverse of the
input function. The actual physical effects follow the arrows, always. We
mention the _apparent_ reverse relationships because they appear
paradoxical, and also are often mistaken for direct effects (as the
stimulus appears to cause the response directly).
So these "causal" relationships can only be _understood_ in
terms of _controlled variables_, a concept that doesn't even exist in
conventional
studies of behavior. So the clearest message of PCT is not that "There is no
causal relationship between behavior and environment" but, rather, "An
observed
causal relationship between behavior and environment can only be
understood once
the controlled variable has been identified".
That's better, in my opinion. It's just very difficult to get across the
way that closed loop screws up causal relationships.
It's ironic: one of Bandura's big beefs about PCT is that it fails to take
goal-setting into account (that is, behavior caused by "pro-active" changes
in the reference setting). So here I am arguing on Bandura's side.
Best,
Bill P.