pop quiz

CSGnet Archive CSG1995
1

[From Bill Powers (950630.1405 MDT)]

Martin Taylor (950630.14:00) --

     Fair enough... And yet, and yet... PCT researchers ALSO want to
     determine what kinds of control systems are inside the organism and
     how those internal structure are connected, etc., etc. The
     approximation o = -1/h(d) applies to ANY reasonably functioning
     control system, which means that if control is very good, it is not
     possible to distinguish among different control structures. Only
     structures that would not control well could be discarded.

I think that we are still at the stage of development of PCT where the
primary concern is to establish the reality of control for the
scientific community. This is a difficult and slow process, as evidenced
in our discussions of PCT and reinforcement theory.

There are many techniques for sorting out the control processes that
possibly go on at different levels of organization. These involve
looking at behavior with a microscope and picking out of it the details
that will tell us about the inner structure. But let's face it: we are
too few and have too little in the way of resources to tackle that job.
Building conjectures on conjectures is a way to set up a range of
possibilities for investigation, but we aren't in any position to take
more than a few halting steps of that kind. We come up immediately
against the need for careful well-instrumented research, and that is
beyond our reach for now. It will remain beyond our reach until a
substantial part of the world of science realizes what control theory
holds in store for all the behavioral sciences, and starts committing
physical and mental resources to exploring this new territory.

For now, Rick's approach is the right one, in my opinion. There are
fundamental features of control behavior that are highly
counterintuitive and that even seem to go against direct observation.
The relationship of controlled input to behavioral actions is one of
these features. If the output affects a controlled variable through a
function h, then we do not find that the controlled variable is simply
the function h of the output. Instead, we find that the controlled
variable acts like the independent variable, and the action becomes the
inverse-h function of a disturbance acting on the controlled variable.
Cause and effect appear to work backward.

Behavior does indeed appear to be a function of its direct consequences.
The most staggering task we have is to convince other scientists that
this appearance is an illusion, that behavior causes its consequences in
the normal way. Thousands upon thousands of scientists have been sucked
into a conceptual system in which the basic fact of control behavior has
been recognized -- and explained incorrectly. Until we can get across
the real explanation, the one that works without magic, we are going to
be stymied.

···

-----------------------------------------------------------------------
Best,

Bill P.

2

[Martin Taylor 950630 17:30]

Bill Powers (950630.1405 MDT)

I hope you don't mean that I am wrong in saying (950630.14:00):
     PCT researchers ALSO want to
     determine what kinds of control systems are inside the organism and
     how those internal structure are connected, etc., etc.

If I am wrong in this, I suspect that your quest is doomed to failure.
Most behavioural scientists, of whatever stripe, are interested in finding
out how organisms work. If you explicitly say that you are interested
only in the environmental part of the control loop, and that what goes
on inside the organism is "Any Old Control System That Works," then you
have asserted that what you do will be of no interest.

In my years on CSG-L, I have not previously observed you to be uninterested
in what goes on inside the organism. I think you once wrote a book with
some suggestions about what might be going on in there. But if one take's
Rick's correct comment at face value, one is led directly to the proposition
that to determine the internal structure of a perceptual control system is
impossible, and that anyone who tries to do so is a fool on a par with the
inventors of perpetual motion machines and angle trisectors.

There are many techniques for sorting out the control processes that
possibly go on at different levels of organization. These involve
looking at behavior with a microscope and picking out of it the details
that will tell us about the inner structure.

All of which depend on control NOT being perfect, and on the multiplicity
of signal paths between output and the CEV corresponding to the controlled
perception.

But let's face it: we are
too few and have too little in the way of resources to tackle that job.

And therefore assert that it is wrong to attempt it? As in Rick Marken
(950629.1845):

The causal relationship
between o and d is (approximately) the inverse of the feedback function,
h, relating o to q; it does not reflect a direct connection between o and
d and it does not reveal anything about the nature of the organism
that is controlling a perception of q (see Powers, Science, 1973).

That is very close to asserting that you could not, by observation
of behaviour, distinguish a sheep from a shepherd, since all you could
see would be the differences between their two environments.

The problem is in the juxtaposition of "(approximately)" and "does not
reveal anything." If the approximation is inexact, then the revelation is
not totally denied. And no living control system controls perfectly.

Martin