Models of Nothing, PCT Research

[From Rick Marken (950509.2120)]

Bruce Abbott (950509.1205 EST)

With brief interruption, the "world model" can be fairly crude yet still
improve on the error that would accrue if the system simply stopped
controlling during the "blind" intervals.

You might want to try calculating how much the world model improves the
"error" (I presume you mean the objectively measured discrepancy
between cursor and target, not the error actually experienced by the
controller) if you add an arbitrary disturbance to the cursor. I think it
would be easy to find many disturbance patterns where the only way to
reduce this "error"is to stop "controlling" (ie. acting) during blind

Bruce Abbott (950509.1240 EST) --

Hans' system is not DESIGNED to control in the PCT sense of the
word...But it does illustrate an interesting principle, and I'm
suggesting we examine that principle with an open mind.

What sense of "control" other than the PCT sense is of any interest to PCT?
I can't think of any sense of "control" other than the PCT sense of control
that is not just "cause-effect". The principles we are interested in here are
the principles of control. Newton's laws illustrate an interesting principle,
too, but it is not a principle of control.

what I'm trying to get you to understand is that model-based control
is not an alternative theory of control; it's just a technique that might
provide some insight into how people manage to maintain control
following loss of signal, under some conditions, at least for brief

I already understand this. What you don't seem to understand is that
people _don't_ manage to maintain control following "loss of signal";
when people can no longer perceive a variable, the variable is no longer
protected from disturbance. What we don't know is what DOES happen when
a person can no longer perceive a controlled variable: do they switch to
a related perception? do they stop controlling: do they continue to
control an imagined version of the controlled perception? Model-based
control (as I said) is a theory in search of a phenomenon. I have nothing
against model based control; it's just that it's about NOTHING (yet). I like
TV sitcoms that are about nothing, but not in scientific theories;-)

You know, there's more to control than a closed loop.

Oh really? Disturbance resistance, perhaps?

For example, when you "run in imagination mode," where do
the imaginary inputs come from?

From the system getting the inputs; the outputs of the system are its

inputs. It's still a closed loop; and the perceptual signal is protected
from disturbances; but in this case the disturbances are neural signals
that add to the perceptual signal.

What do you DEFINE as PCT research? That is, to qualify as PCT
research, what elements must be present? I'd like to be able to
examine, say, a piece of published research and using your criteria,
classify it as PCT or non-PCT.

Excellent idea!! Here's my take on _real_ PCT research:

PCT research is research aimed at determining the perceptual variables
that organisms control and how they control them. For research to be
considered PCT research the following elements must be present:

1. Most important: PCT research is characterized by a vision of
organisms as _controllers_. PCT research sees the organism as a system
that is controlling it's own perceptual inputs. PCT researchers want to
find out what the organism is trying to perceive; they want to see the
world from the organism's perspective. In PCT research, the term
"controlling variables" always refers to what organisms do to variables
in the environment, NOT what variables in the environment do to

2. PCT research is always characterized by a determination of the
variable(s) being controlled by the organism. There is always evidence
that disturbances to some result of an organism's actions have almost
no effect on that result. All analysis and modelling is done after a
controlled variable has been identified and while it is being monitored.

3. PCT research is aimed at finding _lack_ of effect of environmental
variables on behavioral variables. Behavioral variables (like arm
position) are possibly controlled results of action; lack of effect of an
environmental variable on a behavioral variable suggests that the
behavioral variable is under control.

4. PCT research always looks at controlling on a one-organism-at-a-time basis.
Statistical tests are of no use in the study of living control systems; PCT
research is based on modelling, not statistics.

5. Published PCT research is always significant (in the normal sense
and the statistical sense). PCT research is significant if a controlled
variable has been identified and can be monitored. PCT research is not
published unless a controlled variable has been identified. PCT research
is not of the "manipulate and pray for statistical significance" variety.
PCT researchers keep developing their research techniques until controlled
variables can be clearly identified and readily monitored. PCT researchers
only publish their results when they KNOW what's being controlling and how.

Good luck on finding some good examples of PCT research out there in
the conventional literature.