[From Rick Marken (950501.2120)]
Bruce Abbott (950501.1335 EST) --
The effect of CS intensity has been investigated.
But it was investigated under the assumption that the CS was the cause
of the CR. PCT shows that CS-CR relationships are, if anything, a side-
effect of controlling some weird perception (like CS+US+UR).
Actually, the parameters of classical conditioning have been
examined quantitatively.
But, again, this was done under the assumption that classical
conditioning is a lineal causal process: CS-->CR. So the parameters that
were measured are of no real value to a control theory analysis. A
control analysis requires, first, a determination of what variable is
under control. Once we know this we can start examining, via
modelling, the parameters of the control loop.
What has been missing is not so much the quantitative
manipulations as the proper way to model the system.
You keep ignoring the fact that control theory is about control. The
quantitative data on classical conditioning has nothing to do with
control. It was collected under the assumption that a CS causes a CR;
but there is no such causal path. There is just an organism controlling
a perception and acting, as necessary, to protect that perception from
the effects of ANY disturbance.
You continue to claim that PCT is the "proper" way to develop a model
of existing quantitative behavioral data. I suppose there is a sense in
which this is true; after all, these data are measures of aspects of the
behavior of a control system. But I think it is ultimately impossible to
demonstrate the value of PCT in terms of existing data. In fact, I think
it is impossible to use existing data to show that PCT is any better than a
cause-effect model. This is because existing data focus on apparent
cause-effect relationships rather than controlled variables.
Suppose, for example, that existing data showed a linear relationship
between disturbance and output in a tracking task. This relationship
could be modelled by a working cause-effect model or by a control
model (as Bourbon and Powers showed in their "Models and their
worlds" paper). It is highly unlikely that existing data would include
data that had been collected under conditions that would show that ONLY
the control model (controlling a particular perception) accounts for
all the observed relationships in the existing data.
It might help to think about this in terms of Bill's classical
conditioning model. The model just controls (CS+US+CR). But
imagine all the relationships you could observe by varying the relative
intensity and onset of these variables. All kinds of weird relationships
will be observed. What are the chances, based on observing all these
relationships, that you would guess that this data come from an
organism that is controlling (CS+US+CR)? I think they are pretty small.
With existing data you don't know what the organism is controlling
(as you do with Bill's program); all you see are all these weird
relationships between observable variables. I think the chances of
coming up with an accurate control model that happens, as a side effect,
to produce all these relationships, is exceptionally small. That is why
I argue that the only way to approach the study of living control
systems is to ignore all the existing data (for now) and start by doing
The Test to determine what these systems are controlling. Once you know
what living organisms control, all the existing data, to the extent that
it is a side effect of control, will fall into place
Best
Rick