RT, The _Relevance_ of Predictability

[Richard_Marken]

[From Rick Marken (960325.1300)]

Scott Graham (960325) --

Glasser discusses control, orders of perception, etc. in Stations of the
Mind.

I thought "Stations of the Mind" (now called "Control Theory" I think) was a
pretty good book, especially for someone who had only just barely started to
learn control theory. Still, I think that his covereage of levels (orders) of
perception was not strong; and it didn't strike me as being a big part of his
later work. I don't, for example, recall hearing him go into much detail
about how we control certain perceptions as the means of controlling other
perceptions; understanding this is central to understanding intrapersonal
conflict and how it can be resolved.

Bill Powers wrote the forward to "Stations of the Mind." Has Glasser
diverged from that much from his original thinking? How?

I don't know if Glasser diverged from his own original thinking, but I think
his discussions of control theory have either diverged from a correct
description of control theory or simply deteriorated in quality of
presentation.

I think Glasser's problem is that he assumed that he already knew control
theory when he ran into Bill Powers; therefore, he never took the trouble to
learn control theory. I have no particular complaints about Reality Therapy
per se; I think Glasser's approach to therapy is basically a decent one. My
only complaint about Glasser is that he doesn'tteach the facts or the
theory of control in any depth. In my opinion, the failure to teach the basic
theory turns the practice of Reality Therapy into ritual.

Bruce Abbott (960325.1040 EST) --

What is critical is that the participant have information about the current
state of e-coli's movement relative to target (i.e., change in error) and be
able to act on that information before the information becomes outdated.

This is true if the subject is controlling the perception of _movement_
relative to the target. But my revised version of the E.coli demo (now
available as a HyperCard stack called E. coli II) provides no information
about movement relative to the cursor; after a press (action) the dot moves
to a new location and stays there for a while. The press can be based only on
whether or not the dot is on the intended target.

Martin Taylor (960325 11:05)--

For the results of action to be unpredictable, that must apply to ALL of
our control action, not just occasional moments. You have to show that
control is possible even when the result of null action is unpredictable.

I believe I show this in E. coli II; the result of both action (press)
and null action (no press) is unpredictable because I have added a
disturbance that moves the dot to a new random position at some random time
during the null press period. So the subject cannot predict where the dot
will move after a _press_; nor can the subject predict where or _whether_ the
dot will move when there is _no press_.

Bruce Gregory (960325.1315 EST) --

I offer this challenge to those seeking a respite from the "Predictability
Wars."

I can understand that these "Predictability Wars" might seem rather academic.
But I think we are actually dealing (perhaps too indirectly) with issues that
are fundamental to how we view people (and learners) from a PCT perspective.
In particular, we are dealing with the role of perception in behavior.

From a conventional perceptive, perception is the "raw material"that

"starts" the process of behavior. Perception starts this process in the role
of _guide_ (Gibson's ecological optics), cause (behaviorism), or as
information that is _used_ as the basis for behavior (cognition/information
processing). From this perspective, the "predictability" of perception is a
very important issue because it makes perception a better guide, a more
effective cause or a more useful basis of behavior.

From a PCT perspective, perception is just perception; it is a neural signal

in a sensory neuron. It has no properties that make it "useful"; perception
is just what _is_. A dot on a screen is just a dot on a screen. PCT says that
people have learned or developed means for specifying references for the
values of certain perceptual variables -- and people act to make these
perceptual variables match these specifications -- a process called control.
Perceptions are not the start or the end of this process; rather, they are
one variable in a closed causal loop; the variable that is controlled.

When it is said that the "predictability" of a perception is important for
control, the implicit message is that this property of a perception (its
"predictability") guides, causes or is used by the control process; that
control can only occur if the perception has the property of being
"predictable". This view of perception reflects a very fundamental
misconception about how control works - - one that I think could have serious
practical consequences. For example, it could lead teachers to look for
characteristics of the perceptions kids control that make these perceptions
harder or easier to control; they might wonder what it is about the
perceptions controlled in a video game that make them so much easier to
controlthan the perceptions to be controlled in an English class. I think
teachers have already wasted enough time looking for such mythical entities;
it's time for teachers to learn about what kids are actually doing --
controlling.

Control systems coninuously act on the basis of any discrepency between
perception and reference to keep the perception in its reference state. The
predictability of perception is irrelevant to this process -- unless it's the
_perception_ of predictability, itself, that is being controlled.

Best

Rick

[_Martin_Taylor]

[Martin Taylor 960326 10:30]

Rick Marken (960325.1300)

For the results of action to be unpredictable, that must apply to ALL of
our control action, not just occasional moments. You have to show that
control is possible even when the result of null action is unpredictable.

I believe I show this in E. coli II; the result of both action (press)
and null action (no press) is unpredictable because I have added a
disturbance that moves the dot to a new random position at some random time
during the null press period.

That's an improvement, but it doesn't make the influence of null action
on dot behaviour unpredictable. It only makes it less predictable. When
you have any less than perfect predictability, you need to observe the
CEV if its state is to be what you want (i.e. that when you observe it,
the resulting perception matches its reference value). When you have zero
predictability, observing doesn't help, because nothing you do can have an
influence that is more likely to bring the perception nearer to rather than
farther from its reference. Control is required and is possible only between
the two extremes.

Look, you gripe about getting sucked into a quagmire. Me too. All I ever
tried to say was that the e-coli behaviour (in standard form) was largely
predictable, and entirely predictable during periods of null-action. You
turned it (with my complicity) into an analogue of "information theory
wars." My original intention was only to point out that e-coli could not
be taken as a demonstration that predictability was unnecessary.

···

---------------

When it is said that the "predictability" of a perception is important for
control, the implicit message is that this property of a perception (its
"predictability") guides, causes

If that's what worries you, I'm on your side. Predictability in no way
"causes" control.

or is used by the control process; that
control can only occur if the perception has the property of being
"predictable".

This, after the semicolon, is a fact. What implications people choose to
take from it is their business. Predictability matters two ways. If the
disturbance changes unpredictably, control is impossible, or if the
environmental feedback function changes unpredictably, control is impossible.

I think, as I have often said, that the problem is that you take
"predictability" to mean perfect predictability. A disturbance that changes
unpredictably is one whose magnitude and direction have no correlation
from moment to moment, not one that sometimes changes and cannot be observed
by the controller.

For example, it could lead teachers to look for
characteristics of the perceptions kids control that make these perceptions
harder or easier to control;

I'm sure there are hidden quagmires here. But, being no angel...

If a child (or an adult) cannot, by the nature of the environment, generate
actions that have a consistent influence on some perception, no kind of
reorganization will permit that perception to be controlled. Is it not
a good idea for a teacher to help people to assess whether they are likely
to be able to control the perceptions that are giving trouble, and perhaps
to assist them to reorganize so that the higher-level perceptions that
can be controlled are controlled?

they might wonder what it is about the
perceptions controlled in a video game that make them so much easier to
controlthan the perceptions to be controlled in an English class. I think
teachers have already wasted enough time looking for such mythical entities;
it's time for teachers to learn about what kids are actually doing --
controlling.

They would hardly be wondering about these "mythical entities" (controlled
perceptions) if they were not concerned about control. The problem surely
is that they don't know about, understand, or care about controlled
perceptions. As soon as they do, surely they would conceive their job as
assisting the student to control what the student wants to control?

That's quite a subtle proposition, when you realize that the student must
try to control in a social environment that inherently involves conflict.
It may be that to assist the student to control as an adult requires that
the student experience conflict with the teacher on some occasions as a
child. But what, when, where, over what perceptions? Those are the
hard questions.

Control systems coninuously act on the basis of any discrepency between
perception and reference to keep the perception in its reference state. The
predictability of perception is irrelevant to this process

Analytically, that's false. Practically, that's false. And there's the
quagmire.

When you (Rick) write equations, you usually write them without the time
variable. That gives rise to all sorts of misconceptions.

You write such things as p = o+d, o = G*(r-p) or whatever, whereas these
are not values but time functions. In particular the "p" in the two equations
refers to different moments in time. It is predictability over that time
gap that is crucial to control, and by omitting "t" in the equations, you
mislead yourself. You gloss over the problems of phase instability, of
rapidly changing disturbances, of changing environmental feedback functions.

To use an example you should like, what happens if the polarity of a
tracking handle is changed? You've done an experiment on that, as I recall.
I don't remember the number, but it takes something like 500 msec of
exaggerating the error before a higher-level control system reverses
the error-to-output polarity of the tracking system. That exaggeration
of the error was positive feedback, and it happened because the environmental
feedback function turned out to have a prediction error.

Now suppose that the polarity changed again, just about 500 msec after the
first change. What would happen? I suggest that the subject, having just
reversed the polarity of the tracking action, would not instantaneously
reverse it back, but would have another 500 msec or so of positive feedback,
and that with successive reversals of the handle polarity, this would
continue until the subject gave up (dropped the control gain near zero).

What would happen if the handle polarity were to change rapidly (i.e. every
10 msec on average) but erratically? I leave that question as an exercise
for the reader (who could try it if he/she wanted).

There are shades of grey between black and white. Prediction is partial.

Martin