What's so important?

[From Bill Powers (930405.1330 MDT)]

Greg Williams (930405) and Chuck Tucker (930405) --

RE: What's so important about PCT vs IT?

One could ask a similar question about any of the "PCT vs X"
subjects that have come up. X is an approach to the understanding
of behavior that has been around for a long time, which has
attracted the interest of a lot of people, and which has been
carried forward into one elaboration after another, building up a
tradition. Under the best of circumstances, people who believe in
X learn about control theory, and say "Wow, sure, PCT is the most
revolutionary thing since sliced bread, but you know that we Xers
have been saying the same things all along, and when you get
right down to the science of it, X is fundamental. We're glad to
see that PCT fully supports X."

I have to admit that this irks me. But being irked doesn't keep
me from wanting to work out the differences.

ยทยทยท

----------------------------------------------------------------
So, where are we in the PCT-IT discussion? I don't know where
Martin and Allen stand, exactly, but I know that I've come to a
clearer understanding of some premises of IT.

Just in the past few days I've realized that "information" in the
technical sense doesn't actually tell us anything about a
phenomenon that occurs through time. This changes the nature of
the question that has been going back and forth.

In the beginning of this long argument, the PCTers resisted any
suggestion that the disturbance told the control system what to
do. In the hands of the ITers, this became the question of
whether information in the disturbance could get into the control
system, where it could become the basis for action. The reason
for the PCT-side objections was that this sounds like input
causing output as in S-R theory, and it ignores the closed-loop
properties of the control system.

This argument eventually boiled down to the question of whether
the information in the disturbance was represented inside the
control system in such a way that the control system had enough
information to be able to reconstruct the disturbance. The PCTers
said no, the ITers said yes.

Just in the past few days I've realized that this argument is
spurious. The question isn't whether the information in the
perceptual signal is sufficient to reconstruct the disturbance.
It's whether "information" itself, as a measure of a phenomenon
taking place through time, is a sufficient basis for
reconstructing any phenomenon. After tracing back the definition
of information in IT to its origins, I have satisfied myself, if
no one else, that "information" is not the sort of measure from
which one could reconstruct the form of ANY time-varying
phenomenon. The "information" contained in a perceptual signal
could not even be used to reconstruct that same perceptual
signal. I tell you that this signal is transmitting 25 bits per
second. Draw the waveform.

I think that we have uncovered an invalid premise: it is the
assumption that "information" as technically defined has
something to do with how we obtain knowledge. By speaking of the
"information" in one variable that is "about" another variable,
we give the impression that we mean the process of learning
something about one variable from its effects on another
variable. But that is not what the term "information" means, in
IT. What it really means is log(R/r), and THAT IS ALL. If this
measure had been called "pergollity" from the start, nobody would
have confused it with knowledge-obtaining processes. It would
simply have remained a measure with some interesting properties,
but without all the connotations that are dragged in by using a
term like "information."

It's now clear that Martin's Mystery Function is simply a way of
manipulating signals that vary through time, and showing
quantitative relationships among them. This is the same thing we
do in PCT, except that we have a much simpler way of showing why
the output approximately equals and opposes the effects of the
disturbing variable outside the system. This sort of development
has nothing to do with IT or information; it is an argument
entirely within the analog world.

It is true that on the basis of knowing certain of the variables
in a control system, we can reconstruct the form of an external
disturbance. But this does not prove that there is sufficient
information getting into the system from the disturbance to
provide a basis for reconstructing the disturbance -- not unless
we take "information" in a strictly non-technical sense. We can
allow that the perceptual signal contains information in a
measurable amount, by the convention for computing information.
But there is no manipulation of the resulting number that will
result in reconstruction of either the perceptual signal or the
disturbance, or any other time-varying variable.

We can reconstruct the disturbance using the methods of PCT. We
cannot do so using the methods of IT. Therefore the process of
reconstruction does not depend on any measure of "information,"
but on properties of the varying quantities that are not captured
in the measure log(R/r).

If you want to understand how and why a control system works, the
only explanation so far is the one contained in the analog-world
equations.

(Chuck Tucker):I would also appreciate NOT being told that I am
too ignorant to understand.

How much is it worth to you?
----------------------------------------------------------------
Best,

Bill P.

From Ken Hacker [930405]

I am knee deep in Bill's book and am collecting more questions and
points about PCT which I think are important for me to make. I have also
collected various statements directed to me from folks on this net and
am thinking about them. I have not disappeared or given up on making
sense of PCT and how it might help explain human behavior. On the IV and
DV comments made by Rick, Bill, Gary, and others, I simply want to say that
I believe that I CAN test for signficant differences in linear relationships
between variables such as learning program A and learning program B without
and not intending to explain how subjects are controlling the programs, their
ouput, perceptions, or anything else. However, I acknowledge, and realized
part of it here, that the linear relationship between variables, whether
inferring causality or not, does not get to what is very important to any
student's learning and education, that is, self direction, pacing,
involvement, and motivation. I agree strongly that control theory may
a good additional (albeit to some, more useful and substituting) model for
explaining how students will learn through uses of Programs A and B.
Gary, the journal Communication Education is full of articles citing IV-DV
relationships and how the knowledge can improve education. But I DO subscribe
to the view that how students control their perceptions and output, how
they respond to teachers and programs, are critical to a fuller understanding.
Thus, I am completely open to ways to test which kind of program is more
workable, using a PCT paradigm. I will try it out in the fall if it makes
sense to me. In fact, I am thinking about testing this from two paradigms and
also describing to those in my discipline how a PCT model of
teaching shows different, and maybe more useful data, than the traditional
IV-DV approaches. I guess all of us in the social and behavioral learned
in our methods classes that linear relationships between variables are
best source of knowledge about how we can explain human behavior. As I have
read cybernetic literature over the past few years, I have suspected that
the ways in which individuals act upon their own perceptions in terms of
goals and control, are revealing of basic human behavioral principles that
traditional research simply does not explain. This does not mean that
the traditional research is useless. It means that it is inadequate.

Best, KEN

[From Rick Marken (930405.1400)]

Greg Williams (930405) --

I (and perhaps others on the net?) would appreciate some clarification
from those involved in the controversy regarding the importance of a
resolution of the controversy to yourselves and (potentially) to
other netters.

and CHUCK TUCKER (930405) echos --

I would like to echo Greg Williams request of today and ask for an
update on this exchange. I have read all of these comments and have
some difficulty believing that this is not the same old (and very
tiring to me) dispute about "boss reality."

In keeping with the spirit of the season I will try to answer these
questions as though they were formulated as: "why is this controvery
different from all other controversies?"

That is, "why care whether there is any information about the
disturbance in controlled perceptions?"

For me, this controvery gets to the very heart of the difference
between PCT and all other approaches to understanding human behavior.
Behind all the equations, models and words in this controvery is
an image of the nature of a living being. The living being I see
portrayed by the IT group is the old computer image of the
"information processing" school of psychology: perceptual
input, p, is transformed by a computer program into the output,o.
In diagram form, the information processing view is:

    p --> | info processing | --> o

The problem with this image (for me), other than the fact that it is
an incorrect image of a control system, is that it can divert
attention from what is most important (to me) about living
beings -- ie. their goals. The "information processing" image
leaves the door open to the seductive "IV-DV" approach to study-
ing behavior, which is based on measuring the effect of disturbance
variables on output variables. Disturbance - output relationships
are often very pronounced in control behavior -- especially when
disturbances are very abrupt, making the change in output easy to
notice. The IT image of living beings encourages (or, at least, does
not vigorously discourage) taking such relationships at face value.
For example, if we see that people move toward one type of speaker and
away from another (type of speaker being the disturbance, movement
direction being the output) then the IT approach suggests that we look
to characteristics of the speakers for an explanation of this phen-
omenon; what is it about the speakers (distrubance) that was used as
information by the listeners about whether they should move toward
or away?

The living being I see portrayed by PCT is quite different. In PCT,
I have an image of a system that is actively involved in making its
own perceptions match its goals for those perceptions. In IT, the
system does not seem to have much say about what happens to it; it
mechanically converts input into output, a process that has the
happy consequence of keeping perception equal to a reference value --
the offset in the conversion program. The IT and PCT people do agree
that input is kept at a reference level. But in PCT, keeping
perception at this reference level is what behavior is all about; and
the eye- catching relationships between disturbances and outputs is just
a side-effect of this process. With the PCT image clearly in one's
head, the movement of people toward and away from various speakers, for
example, would immediately be recognized as a side effect of a control
process; one in which people are controlling perceptions that are
influenced by (among other things) their distance from the speakers.
The PCT approach suggests that the goal of research is to characterize
the perceptual goals of the listeners rather than the informational
characteristics of the speakers.

So what is important to me in this apparently esoteric little debate
about "information about the disturbance in perception" is the PCT
notion of how to conceptualize and study the behavior of living control
systems. It may be that I am making a mountain out of a molehill; it
certainly is true, for example, that the IT people understand that
behavior (output) is the process by which organisms control their
own perceptual experience (p = r). But there is another side to this
insight that, I think, is being missed, namely, that distrubance -
output relationships (which are often the most noticable aspects
of control behavior) depend mainly on charateristics of the environ-
ment, not of the control system itself (o=-g(d)). Such relationships
may provide hints about controlled variables -- but these hints
must be turned into a specific hypotheses about the variable being
controlled which is then tested using "The Test for the Controlled
Variable".

IV-DV studies may provide hints about the existence of controlled
variables -- but they cannot be a basis for understanding the
behavior of control systems. For my part, I view this whole debate
about whether or not there is "information about the disturbance in
controlled perception" as an attempt to formally demonstrate that the
last sentence is true.

Best

Rick

[Martin Taylor 930406 18:30]
(Rick Marken 930405.1400)

A quick last word before leaving. I may say something between April
15 and 20, but otherwise not till June.

Rick says:

For me, this controvery gets to the very heart of the difference
between PCT and all other approaches to understanding human behavior.
Behind all the equations, models and words in this controvery is
an image of the nature of a living being. The living being I see
portrayed by the IT group is the old computer image of the
"information processing" school of psychology: perceptual
input, p, is transformed by a computer program into the output,o.
In diagram form, the information processing view is:

   p --> | info processing | --> o

If this is why Rick has been arguing, all along, the whole discussion
has been founded on a misapprehension.

All of what I have been writing has been based on the idea that the
above description is just as false as Rick says it is. The whole reason
that I say that information in the perceptual signal is used in the output
is PRECISELY that this is a wrong model. Without the feedback, how could
it work? It may be the old S-R model, but it is simply an unworkable
view of how a living system could survive in the world.

I BASE the argument on the notion that the output CONTROLS the perception.
That's why it works. That's why living systems work. That's why information
from the disturbance is eventually found almost completely in the output.
If there is no control, it doesn't (and can't) get there. If there is
no control, a noisy and uncertain version of the state of the disturbance
may be found in the perceptual signal, but what use is that, except (as
Bill pointed out early in the discussion) to higher level systems, or
(as I pointed out long ago) to a possible alerting system that might
reconfigure the hierarchy so as to permit this perceptual signal to be
brought under control.

The problem with this image (for me), other than the fact that it is
an incorrect image of a control system, is that it can divert
attention from what is most important (to me) about living
beings -- ie. their goals.

I recognize that the discussion has been omitting reference to the goals.
It has been almost inevitable, under the circumstances. But future
(if permitted) information-theory discussions will not omit them. I agree
with Rick that they are the most important distinction between living and
non-living things. But living things would not live, if they had to rely
on magic to achieve their goals. I regret the diversion of attention,
but don't apologize for it. We can't always remain focused on the primary
factor, when there are many to consider.

The living being I see portrayed by PCT is quite different. In PCT,
I have an image of a system that is actively involved in making its
own perceptions match its goals for those perceptions.

Me, too.

In IT, the system does not seem to have much say about what happens to
it; it mechanically converts input into output, a process that has the
happy consequence of keeping perception equal to a reference value --
the offset in the conversion program.

This is the image of IT that riles me. I don't identify with it. I don't
like it, and I won't accept it. If my writing has been so bad as to permit
this image to persist, I can only apologize for a lack of skill. But I
have tried not to conform to this image, and have emphatically said so
in many of the postings, with words like "this works only through the
feedback."

The IT and PCT people do agree
that input is kept at a reference level. But in PCT, keeping
perception at this reference level is what behavior is all about; and
the eye- catching relationships between disturbances and outputs is just
a side-effect of this process.

I think that word "side-effect" isn't right. It is the mechanism. If
the opposition between output and disturbance were not pretty close,
the whole system would not work. Goals would not be attained, perceptions
would not stay near reference levels. The whole system is designed to
permit the opposition to be made as exact as possible. That's why there
is perception to be controlled. If we didn't control perception, we might
vaguely be able to anticipate and sort of oppose disturbances for a little
while, but we wouldn't last long trying to do it. Try just cutting off one
of your senses for a while.

The PCT approach suggests that the goal of research is to characterize
the perceptual goals of the listeners rather than the informational
characteristics of the speakers.

I don't disagree with the goal. I assert that investigation of the
informational characteristics of the things to be controlled is an
essential part of achieving it.

it certainly is true, for example, that the IT people understand that
behavior (output) is the process by which organisms control their
own perceptual experience (p = r). But there is another side to this
insight that, I think, is being missed, namely, that distrubance -
output relationships (which are often the most noticable aspects
of control behavior) depend mainly on charateristics of the environ-
ment, not of the control system itself (o=-g(d)).

Actually, you should attack me from the other side on this. I argue
that this g(d) is not a function at all. Bill P. agrees with you that it
is. But we all agree that it is not the equality function. There's no
time to argue this point now, but what you seem to see as a problem is
merely a point I had hoped to be able to bring up once we got past
what I thought was a quite simple matter, but which swamped the net
over the last weeks.

IV-DV studies may provide hints about the existence of controlled
variables -- but they cannot be a basis for understanding the
behavior of control systems. For my part, I view this whole debate
about whether or not there is "information about the disturbance in
controlled perception" as an attempt to formally demonstrate that the
last sentence is true.

I think these two sentences bear no relation to each other. If I change
the gain or transport lag in a control system model, and measure the
correlation between the model's data and the human's, I am doing an
IV-DV study. You do it all the time. What has that to do with whether
information about the disturbance goes through the perceptual signal?

Enjoy Spring, though I guess in California you don't notice it happening.

Martin