Talkin' PCT

[From Rick Marken (970411.2230 PDT)]

Martin Taylor (970411 15:15) --

My suggestion of conflict comes only from your answer "No" to the
question, which seems inconsistent with the methods you use to
suggest to Hans Blom, Bruce Abbott, Jeff Vancouver and others
(including me) exactly how our understanding is flawed. Basically
that method is "If you can't say or see it my way, don't say it
at all."

I don't think so, Martin. I don't want you guys to stop saying what
you're saying. Your postings provide me with what little amusement
I am able to squeeze out of this vale of tears;-) Of course, I would
be much happier if you guys would just drop all your baggage and
start helping us do PCT science. But, apparently, a lot of PCT
science conflicts with what you already believed before you stumbled
upon PCT: information theory, MCT, the cause-effect model of research,
whatever. I'm obviously not going to change your minds about these
beliefs -- and you guys are clearly not going to change my mind about
the merits of these beliefs. I've already been though those beliefs
myself and rejected them a long time ago (for two good reasons -- data
and modeling) when I started to understand living systems from a PCT
perspective.

I am afraid I am a hopeless case when it comes to PCT. If you want
to convert me to beliefs that I have already given up it's going to
be a tough road to hoe. I certainly don't want you to stop trying to
convince me of the merits of your beliefs if that's what you want
to do; but (as I tell the Jehovah's Witnesses who come by
occasionally) you're probably not going to have much luck.

I admit that I get tired having to deal with people who are more
interested in teaching me what they believe (and how those beliefs
are supported by or support PCT) than in learning about and DOING PCT.
But what can I do? I don't think anyone who stumbles across PCT has any
idea what they're getting into (I sure didn't). I think just about
everyone comes to PCT with a great deal of baggage; I came with the
baggage of experimental psychology; others come with the baggage of some
favorite "wise person" (Dewey, Popper, etc), others come with the
baggage of some favorite theory, like natural selection, etc etc.

When I started reading B:CP and getting interested in PCT I had no idea
that PCT would challenge the foundations of experimental psychology;
that I would have to drop my baggage if I wanted to ride in comfort on
the PCT express. It was not easy for me to drop that baggage; I imagine
that most people resent the idea (as I did) of giving up what they
consider their very precious baggage. In fact, I know only a very few
people who have been able to drop their baggage and jump on the PCT
express. Most people stick with their baggage at the station and just
catch the PCT train when it comes through town.

When CSGNet was formed I had hoped that it could be used to develop PCT
science. For some reason I imagined that the people who joined
the list would be like me; they would be people who had already dropped
the baggage and were ready to start working with others who had dropped
their baggage too. Out in the "real world" I spent all my time trying to
get my PCT work past non PCT-wise reviewers. Here in the PCT cyberworld
I thought there would be no resistance to PCT; no condescending
instruction about how behavior and control "really" work; no cries of
"straw man" every time we tried to compare PCT to conventional models of
behavior; no appeals to the authority of "experts in the field". Boy,
was I ever wrong.

I certainly don't have anything against people who don't want to drop
their baggage. I understand their problem and I'm very sympathetic.
Anyway, I can't do anything about it. But that doesn't mean that I'm not
going to occasionally jibe and ridicule baggage that I find ridiculous.
I'm only human;-)

I'm not at all sure that a little knowledge isn't a dangerous
thing.

Oh, come on, Martin. People will find any justification they want
for doing nasty things to each other. Heck, nowadays (in the US) people
use Chistianity rather than social Darwinism to justify the fact that
people are homeless and starving (they have bad morals rather than bad
genes;-)). But I still think people should have
free access to both "The Origin of Species" and "The Bible":wink:

If I believed whatever it is that you apparently believe about
the work of Hans, Bruce, Jeff, and me, I would try to cast my
views in such a way that the errors you see could be perceived
by the person concerned (or by myself as an erroneous critic).

We spent over a year, using simulations and all other means we could
think of, to show you that information theory contributes nothing
to our understanding of control. We had a similar discussion about
reinforcement theory. The same is happening right now in the discussion
of MCT. The "errors" in information theory, reinforcement theory and MCT
are (from my perspective) so large, the holes so gaping, that I cannot
understand how anyone, confronted with all the evidence we posted, could
continue to see any merit in these idesa. But you guys do continue to
find merit in these ideas. I think we are at a point where we just have
to admit that people believe what they want to believe and that's pretty
much it. Evidence -- data and models-- convinces only if people are
willing to be convinced.

Your messages to Hans mostly take the form of "MCT doesn't work
so why keep going on about it" or less polite equivalent statements.
I'd take that to be rather a persistent attempt to suppress
whatever it is that Hans is trying to get across.

I don't have any interest in supressing Hans' statements. But it's
getting pretty tiresome. Hans has been on this net for at least 4 years
and he should know by now why MCT makes no sense as a model of
purposeful behavior. The evidence against MCT are not subtle; it is
glaring. The fact that Hans just ignores this evidence (mainly the fact
that an MCT controller cannot possibly control in a world of
unpredictable disturbances ,-- something that all living organisms do
all the time) shows that he has absolutely no interest in either
abandoning MCT or learning PCT. This is sure no surprise; Hans has a LOT
invested in MCT. Since it's clear that neither evidence nor
reason will change Hans' mind, I sometimes succumb to my frustration and
say things that are impolite. But I'm certainly not interested in
supressing Hans.

Please note that I'm taking you at your word when you say:

>Why worry about causing disturbances to perceptions I'm
. controlling?

See Bill's (970311.1939 MST) most recent post. Rest assured that
your disturbances have had virtually no effect on my controlled
variables. I didn't even detect them;-)

Best

Rick

[Martin Taylor 970414 11:55]

Rick Marken (970411.2230 PDT)

I would
be much happier if you guys would just drop all your baggage and
start helping us do PCT science. But, apparently, a lot of PCT
science conflicts with what you already believed before you stumbled
upon PCT: information theory, ... I've already been though those beliefs
myself and rejected them a long time ago (for two good reasons -- data
and modeling) when I started to understand living systems from a PCT
perspective.

I've said it before, and no doubt will say it again: To assert that
PCT conflicts with information theory is equivalent to saying PCT conflicts
with Fourier Analysis, or that PCT conflicts with Pythagoras's theorem.
If you believe it so, then you are asserting that PCT is non-scientific.
To investigate PCT under these conditions is about as realistic as to
study planetary motions in the belief that they conflict with the
differential calculus or the theory of Abelian groups.

When CSGNet was formed I had hoped that it could be used to develop PCT
science.

I remember being severely chastized by you for holding to this same hope.
I don't remember the words exactly, but they were something like: "What
do you think you are trying to do? Improve PCT?" as if PCT had been delivered
on stone tablets from a mountain top. I remember making a plea that CSGnet
should be used to develop PCT science, a plea that was, how shall I put it,
laughed out of court. And that was before the information-theory discussion
even started.

···

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

> I'm not at all sure that a little knowledge isn't a dangerous
> thing.

Oh, come on, Martin. People will find any justification they want
for doing nasty things to each other. Heck, nowadays (in the US) people
use Chistianity rather than social Darwinism to justify the fact that
people are homeless and starving (they have bad morals rather than bad
genes;-)). But I still think people should have
free access to both "The Origin of Species" and "The Bible":wink:

So do I. But I'm just not as sure about things as you seem always to be.
And to think that people should have access to the sources of knowledge
is by no means incompatible with believing that getting it half-right can
be more dangerous than knowing nothing at all about it. Look at the results
we get by having politicians who follow conventional economics and "free
market" pseudo-Darwinism.

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

We spent over a year, using simulations and all other means we could
think of, to show you that information theory contributes nothing
to our understanding of control.

"Our" in the last line being Marken and possibly Powers, I think. With this
interpretation, I have to agree with your statement.

All the data that disagreed with your position was finessed by arguments
worthy of a lawyer. That's why you were unable to "show" me. And I was unable
to show you because I was unable to--not allowed to because of specious
arguments that cost much time to refute--develop the basic understanding
you needed in order to follow what was going on. Your rhetoric consisted
largely of failing to understand that there is a difference between
zero and a non-zero number less than unity. "There is NO information about
the disturbance in the perceptual signal." You took the correct assertion
that there is imperfect information about the disturbance passed by the
perceptual function, and turned it into an assertion about information
about the disturbance "in" the perceptual signal, and from that to an
assertion that imperfect information is the same as zero information, and
thence into a mishmash of concepts that I find hard to follow. It's not a
labyrinth I want to tread again.

Oh well. I'm afraid I just stepped into another mess of tar at the entrance
to the labyrinth.

Martin

[From Bruce Gregory (970414.1315 EST)]

Martin Taylor 970414 11:55
Rick Marken (970411.2230 PDT)

Oh well. I'm afraid I just stepped into another mess of tar at the entrance
to the labyrinth.

Is this a private tar pit, or can anyone step in? I draw a
distinction between those who understand PCT but look at the
world in ways that I do not always find illuminating
(although they clearly do) and those who seem bent on improving
PCT without necessarily bothering to understand it first. (Of
course the latter are convinced that they _do_ understand PCT,
but that seems beyond help; you can only help those who
recognize they could use help :wink:

Bruce

[From Bruce Abbott (970414.1410 EST)]

Martin Taylor 970414 11:55, replying to Rick Marken (970411.2230 PDT) --

Well said, Martin. I couldn't agree more.

Regards,

Bruce

[From Rick Marken (970414.1320)]

Martin Taylor (970414 11:55) --

To assert that PCT conflicts with information theory is equivalent
to saying PCT conflicts with Fourier Analysis

The whole information theory discussion started (years ago) when you
claimed (based on information theory) that a control system had to have
information about the disturbance(s) to a controlled variable in order
to control. This ridiculous assertion conflicts (in the most fundamental
way) with the facts of control.

Since the information theory discussion started, you have changed your
story about what information theory says about control in various ways.
I think we are currently at the point where you agree that the
control system itself doesn't have (or need) any information about
disturbing variables in order to control (looking ahead I see that I am
wrong -- the delusion persists); your current story (the one I agree
with) is that a person observing the control system has information
about the net disturbance to a controlled variable in the form of the
output of the control system; knowing this output, the feedback function
and the function relating disturbance to controlled variable, the
observer can derive the net disturbance. We (Bill and I) agree with this
but we still don't see what this contributes to our understanding of
control beyond what we already know from simple algrebra.

Me:

When CSGNet was formed I had hoped that it could be used to
develop PCT science.

Martin:

I remember being severely chastized by you for holding to this
same hope. I don't remember the words exactly, but they were
something like: "What do you think you are trying to do? Improve
PCT?" as if PCT had been delivered on stone tablets from a
mountain top.

No. As if PCT had been delivered on paper (and on computer) to be
improved by TESTING IT AGAINST OBSERVATIONS.

Your idea of doing science is my idea of doing nothing at all. You can't
improve a model by just thinking about it. You have to derive
predictions from that model and see if the behavior that is being
modeled (the behavior of a living control system) is the same as the
behavior of the model. This is the only way to develop PCT science.

I remember making a plea that CSGnet should be used to develop PCT
science, a plea that was, how shall I put it, laughed out of court.

Why plead? Why not just DO IT! If there was any laughing it was probably
me laughing at the idea that it is possible to develop
PCT science using information theory or attractor theory or whatever.
You develop a science by 1) observing, then 2) modeling, then
3) testing 4) testing 5) testing 6) testing and 7) testing.

Bill has already done steps 1 through 3. I'm working on step 4. The only
way to develop PCT science is by doing steps 4 through 7.

Bruce Abbott (970414.1410 EST) --

Martin Taylor 970414 11:55, replying to Rick Marken (970411.2230 PDT)

Well said, Martin. I couldn't agree more.

What do you agree with, Bruce? Martin's notion of how to do science? Or
his remarkable analytic discovery that

there is imperfect information about the disturbance passed by
the perceptual function

Best

Rick

[Martin Taylor 970415 10:10]

Bruce Gregory (970414.1315 EST)]

Martin Taylor 970414 11:55

Oh well. I'm afraid I just stepped into another mess of tar at the entrance
to the labyrinth.

Is this a private tar pit, or can anyone step in?

Anyone can step in any tar-pit they like. We don't have traffic cops here
on CSGnet (do we?). But I think that to step into tar-pits (or rather,
to pick up tar-babies) is rather ill-advised.

I draw a
distinction between those who understand PCT but look at the
world in ways that I do not always find illuminating
(although they clearly do) and those who seem bent on improving
PCT without necessarily bothering to understand it first. (Of
course the latter are convinced that they _do_ understand PCT,
but that seems beyond help; you can only help those who
recognize they could use help :wink:

A good distinction--but who is to be the judge among people who believe
they understand PCT, but understand it differently? Doesn't each person
have to judge individually their own degree of understanding, and using
that judgment, judge their own opinion of the degree of understanding
displayed by others?

Martin

[Martin Taylor 970415 11:10]

Rick Marken (970414.1320)]

Martin Taylor (970414 11:55) --

> To assert that PCT conflicts with information theory is equivalent
> to saying PCT conflicts with Fourier Analysis

The whole information theory discussion started (years ago) when you
claimed (based on information theory) that a control system had to have
information about the disturbance(s) to a controlled variable in order
to control. This ridiculous assertion conflicts (in the most fundamental
way) with the facts of control.

Since the information theory discussion started, you have changed your
story about what information theory says about control in various ways.

The last time you said this, I went to the trouble of going back to the
archive to check whether you were right. I don't trust my memory to the
extent that you trust yours. Having found that you were not correct
(except in minor details that reflect an increase in my precision of
understanding--and I would hope we all change for this kind of reason),
I posted a long message consisting of the relevant parts of the first
series of interchanges. You chose to ignore this _factual_ history, and
there's nothing much I can do about that. Believe what you will. It's
no skin off my nose.

I think we are currently at the point where you agree that the
control system itself doesn't have (or need) any information about
disturbing variables in order to control

I don't know what you mean about being "at the point", since it's a
red herring you (or perhaps Bill--I don't remember) inserted. Long before
we ever got into any _technical_ discussion of information theory, it was
(and remains) trivially obvious that a scalar signal cannot distinguish
anything more than variations along one dimension. It is conceptually
ridiculous to suppose that the scalar perceptual signal could distinguish
among _causes_ of variation, and alwasy has been. That's never been an
issue, no matter how often you assert that I (and Allan Randall) believe
such nonsense. It's a purely political trick of rhetoric, and I like it
now no more than I did the first time you brought it up. That first time,
I acknowledge that there was a confusion of usage that could have justified
your misapprehension. This excuse no longer persists, and should not have
persisted past the first couple of days of discussion years ago,

(looking ahead I see that I am
wrong -- the delusion persists); your current story (the one I agree
with) is that a person observing the control system has information
about the net disturbance to a controlled variable in the form of the
output of the control system; knowing this output, the feedback function
and the function relating disturbance to controlled variable, the
observer can derive the net disturbance. We (Bill and I) agree with this
but we still don't see what this contributes to our understanding of
control beyond what we already know from simple algrebra.

Well, at least that's a huge step forward. And, do you also agree that the
output is derived from the operation of the output function on the
perceptual and reference variables? If you do, then there might yet be
hope for a renewed discussion of the value of information theory (as
opposed to its applicability, about which there is less question than
there is about the applicability of "simple algebra"). ("Simple algebra,"
in case it isn't obvious, applies to control systems only in the long-term
limit of an unvarying disturbance waveform and an unvarying reference
level. To handle transient variation, you need other tools).

However, your principle is correct, if you substitute "what we already
know from Laplace Transform or similar analyses in the case of linear
control systems." I said from the outset that one will never learn more
about the detailed behaviour of a specified control system from informational
analysis than one will by doing a proper dynamical analysis of that
control system. And one won't. One may, however, be able to say things
about the typical behaviour of typical systems of different classes, and
say some things about what might and what might not work.

Bruce Abbott (970414.1410 EST) --

>Martin Taylor 970414 11:55, replying to Rick Marken (970411.2230 PDT)

>Well said, Martin. I couldn't agree more.

What do you agree with, Bruce? Martin's notion of how to do science? Or
his remarkable analytic discovery that

> there is imperfect information about the disturbance passed by
> the perceptual function

That's a vast improvement on "There is NO, repeat NO, information about
the disturbance in the perceptual signal. None, Nada." There may be hope
for you yet;-)

Martin

[From Bill Powers (97045.0848 MST)]

Rick Marken (970414.1320)--
Martin Taylor (970414 11:55) --

To assert that PCT conflicts with information theory is equivalent
to saying PCT conflicts with Fourier Analysis

The whole information theory discussion started (years ago) when you
claimed (based on information theory) that a control system had to have
information about the disturbance(s) to a controlled variable in order
to control.

A very large part of this argument stems from failing to distinguish between
(a) the _effect_ of a disturbing variable on the controlled variable, and
(b) the _state_ of the disturbing variable and the _form of the function_
through which it exerts its effect. Any information about the disturbance
available to the control system can only exist in the form of the _effect_
of the disturbing variable. There is no information available to the control
system about either the _state_ of the disturbing variable or the _form of
the function_ through which it has its effects. A disturbance d of 10 units
acting through the function d^2 has exactly the same effect as a disturbance
of 50 units acting through the function 2*d.

This is why I proposed distinguishing between "the disturbance" (the state
of the disturbing variable) and a "perturbation" (the amount of change in
the controlled variable that would result from this amount of disturbance if
the system's output remained the same). The perturbation can be deduced from
knowing the state of the controlled variable, the form of the environmental
feedback function, and the state of the output variable. But having deduced
the perturbation, we would also have to know the form of the function
connecting the disturbance to the controlled variable in order to deduce the
state of the disturbing variable.

Since the standard control system diagram shows no sensory connection from
the output or the forms of the environmental feedback function and
disturbing function to the perceptual signal, it is clear that this basic
unit of organization cannot _itself_ know anything about the output, the
external functions, or the disturbing variable. To allow the system itself
to make use of this information, you would have to modify the diagram,
showing the information pathways by which knowledge of the various parts got
inside the system, and you would have to supply computing functions to carry
out all the deductions that are assumed. This would greatly change the
structure of the diagram, and make it enormously more complex.

However, if we interpret Martin's statement (as amended over the years) in a
different way, we do not have to assume he means that the control system
itself makes any use of the knowledge of these unsensed variables and
functions. If there is an analyst who knows about the output, the
environmental feedback function, the disturbing function, the disturbing
variable, and the controlled variable, then the _analyst_ can (in principle)
do an Informational analysis of the system, showing how a hypothetical
quantity called Information could be traced through the various processes.
This is, of course, aside from the fact that nobody has yet done a
successful Informational analysis of a closed-loop system, all attempts to
do actual numerical calculations having failed so far.

In a quite similar way, an analyst could compute the _average values_ of all
the signals and variables, over some time like a year, and show that certain
relationships exist among them. This would not mean that the control
system's operation can be explained in terms of these average values, but
they could certainly be calculated. The average values would behave as they
do _because of the way the control system works_, but the opposite would not
be the case: the control system would not work _because of the way the
average values behave_.

This is the view to which I have come concerning Information theory. If we
were to succeed in calculating the flows of Information in various parts of
a control system, using whatever knowledge the analyst has available, we
would come up with numbers assignable to various parts of the system. These
numbers would behave over time as they do because of the way the control
system operates. If the organization of the control system were different,
we would find different numbers, behaving differently. But in no way could
we then assert that the control system behaves as it does _because of the
way these numbers behave_.

I suppose there is some deep philosophical principle at issue here. Do our
abstract generalizations have any effect on the way natural systems operate?
Or are they just dependent on the way natural systems operate? Physicists
seem to fall into a mode of speaking in which it seems that such things as
quantum principles somehow "govern" nature, as if the principles came first
and nature then obeyed them. I think the PCT view would suggest the opposite
relationship: nature behaves as it does, and we construct perceptual systems
through which we make sense of its behavior.

Best,

Bill P.

[Martin Taylor 970416 09:00]

From Bill Powers (97045.0848 MST)]

Rick Marken (970414.1320)--
Martin Taylor (970414 11:55) --

To assert that PCT conflicts with information theory is equivalent
to saying PCT conflicts with Fourier Analysis

The whole information theory discussion started (years ago) when you
claimed (based on information theory) that a control system had to have
information about the disturbance(s) to a controlled variable in order
to control.

Bill, your message is a very positive contribution to the interaction
on information theory. Naturally, I have a few comments, but I'm much
encouraged by the message as a whole.

I proposed distinguishing between "the disturbance" (the state
of the disturbing variable) and a "perturbation" (the amount of change in
the controlled variable that would result from this amount of disturbance if
the system's output remained the same).

This is a new introduction of the _word_ perturbation. In previous
discussion, we have used "disturbing influence" for this. The word
"perturbation" suggests to me the _observed_ effect of the disturbing
variable (as opposed by the control system's output). I don't care
what words we use when we are talking among ourselves, all of us knowing
what concept we mean. But as we are aware, words can mislead--as my original
use of "disturbance" to mean this same concept misled you to think I
meant what you mean by "disturbing variable." Nowadays, I do tend to use
"disturbance" again as a short form for "disturbing influence" --what
you now call "perturbation"--when I think I will not be misunderstood,
but I shouldn't, because it does get misunderstood, even By Rick as
recently as yesterday.

However, if we interpret Martin's statement (as amended over the years) in a
different way, we do not have to assume he means that the control system
itself makes any use of the knowledge of these unsensed variables and
functions. If there is an analyst who knows about the output, the
environmental feedback function, the disturbing function, the disturbing
variable, and the controlled variable, then the _analyst_ can (in principle)
do an Informational analysis of the system, showing how a hypothetical
quantity called Information could be traced through the various processes.
This is, of course, aside from the fact that nobody has yet done a
successful Informational analysis of a closed-loop system, all attempts to
do actual numerical calculations having failed so far.

Right. I'll go along with that. Including the last sentence. But I
dispute "(as amended over the years)". Your understanding of what I say
may have been amended over the years, but my intent has not.

If you really believe what you say here, I might be encouraged to go
back and try to make the last sentence false.

In a quite similar way, an analyst could compute the _average values_ of all
the signals and variables, over some time like a year, and show that certain
relationships exist among them. This would not mean that the control
system's operation can be explained in terms of these average values, but
they could certainly be calculated. The average values would behave as they
do _because of the way the control system works_, but the opposite would not
be the case: the control system would not work _because of the way the
average values behave_.

No, no more than a linear control system works the way it does because
its Laplace Transform relations are the way they are. The control system
works as it does because Nature is the way it is. The Laplace Transform
analysis is a way of describing it, and helping a person familiar with
LT analysis to understand what the control system is doing. I think of
the information theory analysis in the same way. It helps a person
accustomed to thinking in that way to understand what is going on, and
sometimes to assess easily what may happen in a new configuration.

This is the view to which I have come concerning Information theory. If we
were to succeed in calculating the flows of Information in various parts of
a control system, using whatever knowledge the analyst has available, we
would come up with numbers assignable to various parts of the system. These
numbers would behave over time as they do because of the way the control
system operates. If the organization of the control system were different,
we would find different numbers, behaving differently. But in no way could
we then assert that the control system behaves as it does _because of the
way these numbers behave_.

Is this in some way different from the previous paragraph? I think I agree
with it in the same way as the previous one, but the apparent repetition
makes me think that I may be missing some crucial distinction.

We can't assert that the control system will behave as it does because the
numbers behave the way they do. But if an analysis (Laplace or other) with
new numbers behaves in a particular way, we can be pretty sure that if a
control system is built according to those new numbers, it will behave the
way the numbers do. Description and prediction are opposite sides of the
same coin. Neither implies causality.

···

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

I suppose there is some deep philosophical principle at issue here. Do our
abstract generalizations have any effect on the way natural systems operate?

No. They help us relate the way one natural system operates to the way
other natural systems operate, perhaps in vastly different circumstances.

Or are they just dependent on the way natural systems operate? Physicists
seem to fall into a mode of speaking in which it seems that such things as
quantum principles somehow "govern" nature, as if the principles came first
and nature then obeyed them. I think the PCT view would suggest the opposite
relationship: nature behaves as it does, and we construct perceptual systems
through which we make sense of its behavior.

That's my view, too. But I have had unresolved arguments with philosophers
who believe the contrary--that we are allowed to perceive the workings of
a great machine that obeys these governing principles that we may now be
discovering. I'm unable to argue my (and your) position against this
notion, and I'm inclined to think that there cannot be any direct way
of distinguishing the two positions, given that the rules we find by
our constructed perceptions may be unrelated to the "rules" by which
Nature may be governed--if indeed there are any such rules.

Overriding this unresolved debate is the question "Does it matter?"
I think not, for our purposes here.

Martin

[From Bill Powers (970416.1043 MST)]

Martin Taylor 970416 09:00--

Bill, your message is a very positive contribution to the interaction
on information theory. Naturally, I have a few comments, but I'm much
encouraged by the message as a whole.

Good, maybe we can lay this long and confused dispute to rest.

I proposed distinguishing between "the disturbance" (the state
of the disturbing variable) and a "perturbation" (the amount of change in
the controlled variable that would result from this amount of disturbance
if the system's output remained the same).

This is a new introduction of the _word_ perturbation. In previous
discussion, we have used "disturbing influence" for this. The word
"perturbation" suggests to me the _observed_ effect of the disturbing
variable (as opposed by the control system's output). I don't care
what words we use when we are talking among ourselves, all of us knowing
what concept we mean. But as we are aware, words can mislead--as my
original use of "disturbance" to mean this same concept misled you to
think I meant what you mean by "disturbing variable."

Unfortunately, "influence" 1s no less ambiguous than "disturbance." We can
say "His friend WAS a bad influence on him", or we can say "His friend HAD a
bad influence on him." Influence can mean either the cause or the effect.

However, I agree with you that "perturbation" is more likely to imply the
actual effect, rather than the effect of the disturbing variable with the
output held constant. Unfortunately, the observed (apparent) effect depends
as much on the control system's action as on the disturbing influence,
especially when you go to the continuous case. We never see the actual
effect of the disturbance independently of the effect of the action (and
neither does the control system). Even with a step-change in the disturbing
variable, there could be a step-change in the action at the same time, from
a previous change in either the disturbing variable or the reference signal.

Perhaps we simply have to give up on trying to find a single word that
commonly means exactly what we want it to. The nearest I can come is
"contribution": there is a contribution to the state of the controlled
variable from the disturbing variable acting through the disturbance
function, and from the output variable acting through the environmental
feedback function; the state of the controlled variable is the sum of these
contributions (in a linear universe).

As to the remaining philosophical points, I think we can agree on them and
drop the subject (unless one of us finds the other violating the agreement,
or unless we find that we're not agreeing to the same thing!).

···

-----------------------------
We seem to have among us a mathematician of considerable analytical talents,
Richard Kennaway. I hope you have obtained and read the paper he mentioned.
When I get permission from Richard, I will post the PostScript version that
Wolfgang Zocher whipped up for me last night, on my FTP page (or, come to
think of it, I'll just send it to Richard who can post it on his).

I would like to invite Richard into this discussion, to see if he, perhaps,
could find a way to analyze the Information content of various signals in a
control system, the simplest one with an integrating output and a
proportional environmental feedback function. A fresh pair of eyes might
find the easy way to do this, uncontaminated by preconceptions on either
side. I don't have anything to contribute here, but I realize that there are
people on CSGnet who might like to see the resolution of the "information in
perception" debate carried to a definitive stage.

Best,

Bill P.

Nowadays, I do tend to use

"disturbance" again as a short form for "disturbing influence" --what
you now call "perturbation"--when I think I will not be misunderstood,
but I shouldn't, because it does get misunderstood, even By Rick as
recently as yesterday.

However, if we interpret Martin's statement (as amended over the years) in a
different way, we do not have to assume he means that the control system
itself makes any use of the knowledge of these unsensed variables and
functions. If there is an analyst who knows about the output, the
environmental feedback function, the disturbing function, the disturbing
variable, and the controlled variable, then the _analyst_ can (in principle)
do an Informational analysis of the system, showing how a hypothetical
quantity called Information could be traced through the various processes.
This is, of course, aside from the fact that nobody has yet done a
successful Informational analysis of a closed-loop system, all attempts to
do actual numerical calculations having failed so far.

Right. I'll go along with that. Including the last sentence. But I
dispute "(as amended over the years)". Your understanding of what I say
may have been amended over the years, but my intent has not.

If you really believe what you say here, I might be encouraged to go
back and try to make the last sentence false.

In a quite similar way, an analyst could compute the _average values_ of all
the signals and variables, over some time like a year, and show that certain
relationships exist among them. This would not mean that the control
system's operation can be explained in terms of these average values, but
they could certainly be calculated. The average values would behave as they
do _because of the way the control system works_, but the opposite would not
be the case: the control system would not work _because of the way the
average values behave_.

No, no more than a linear control system works the way it does because
its Laplace Transform relations are the way they are. The control system
works as it does because Nature is the way it is. The Laplace Transform
analysis is a way of describing it, and helping a person familiar with
LT analysis to understand what the control system is doing. I think of
the information theory analysis in the same way. It helps a person
accustomed to thinking in that way to understand what is going on, and
sometimes to assess easily what may happen in a new configuration.

This is the view to which I have come concerning Information theory. If we
were to succeed in calculating the flows of Information in various parts of
a control system, using whatever knowledge the analyst has available, we
would come up with numbers assignable to various parts of the system. These
numbers would behave over time as they do because of the way the control
system operates. If the organization of the control system were different,
we would find different numbers, behaving differently. But in no way could
we then assert that the control system behaves as it does _because of the
way these numbers behave_.

Is this in some way different from the previous paragraph? I think I agree
with it in the same way as the previous one, but the apparent repetition
makes me think that I may be missing some crucial distinction.

We can't assert that the control system will behave as it does because the
numbers behave the way they do. But if an analysis (Laplace or other) with
new numbers behaves in a particular way, we can be pretty sure that if a
control system is built according to those new numbers, it will behave the
way the numbers do. Description and prediction are opposite sides of the
same coin. Neither implies causality.

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

I suppose there is some deep philosophical principle at issue here. Do our
abstract generalizations have any effect on the way natural systems operate?

No. They help us relate the way one natural system operates to the way
other natural systems operate, perhaps in vastly different circumstances.

Or are they just dependent on the way natural systems operate? Physicists
seem to fall into a mode of speaking in which it seems that such things as
quantum principles somehow "govern" nature, as if the principles came first
and nature then obeyed them. I think the PCT view would suggest the opposite
relationship: nature behaves as it does, and we construct perceptual systems
through which we make sense of its behavior.

That's my view, too. But I have had unresolved arguments with philosophers
who believe the contrary--that we are allowed to perceive the workings of
a great machine that obeys these governing principles that we may now be
discovering. I'm unable to argue my (and your) position against this
notion, and I'm inclined to think that there cannot be any direct way
of distinguishing the two positions, given that the rules we find by
our constructed perceptions may be unrelated to the "rules" by which
Nature may be governed--if indeed there are any such rules.

Overriding this unresolved debate is the question "Does it matter?"
I think not, for our purposes here.

Martin

[From Bruce Gregory (970416.1430 EST)]

Bill Powers (97045.0848 MST)

This is the view to which I have come concerning Information theory. If we
were to succeed in calculating the flows of Information in various parts of
a control system, using whatever knowledge the analyst has available, we
would come up with numbers assignable to various parts of the system. These
numbers would behave over time as they do because of the way the control
system operates. If the organization of the control system were different,
we would find different numbers, behaving differently. But in no way could
we then assert that the control system behaves as it does _because of the
way these numbers behave_.

Need to keep straight the system and the model. The system
behaves however it does. The model works according to fixed
rules. It would be _very_ peculiar to say that the system
behaved some way because of the rules used to construct the
model. All we can ever say is how well the model agrees with the
system. The system behaves _as if_.....

I suppose there is some deep philosophical principle at issue here.

I doubt it!

Do our
abstract generalizations have any effect on the way natural systems operate?
Or are they just dependent on the way natural systems operate? Physicists
seem to fall into a mode of speaking in which it seems that such things as
quantum principles somehow "govern" nature, as if the principles came first
and nature then obeyed them.

Yes, but this is just an example of reifying abstractions --
something we human beings are prone to do despite our best
efforts. Physicists, like mathematicians are particularly
susceptible to this disorder :wink:

I think the PCT view would suggest the opposite
relationship: nature behaves as it does, and we construct perceptual systems
through which we make sense of its behavior.

Indeed!

Bruce

[From Bruce Gregory (970416.1545 EST)]

Martin Taylor 970416 09:00

>From Bill Powers (97045.0848 MST)]

>Or are they just dependent on the way natural systems operate? Physicists
>seem to fall into a mode of speaking in which it seems that such things as
>quantum principles somehow "govern" nature, as if the principles came first
>and nature then obeyed them. I think the PCT view would suggest the opposite
>relationship: nature behaves as it does, and we construct perceptual systems
>through which we make sense of its behavior.

That's my view, too. But I have had unresolved arguments with philosophers
who believe the contrary--that we are allowed to perceive the workings of
a great machine that obeys these governing principles that we may now be
discovering. I'm unable to argue my (and your) position against this
notion, and I'm inclined to think that there cannot be any direct way
of distinguishing the two positions, given that the rules we find by
our constructed perceptions may be unrelated to the "rules" by which
Nature may be governed--if indeed there are any such rules.

I ask philosphers the following question, "Granted that we
_might_ have access to the rules God used to construct the
universe, how can we tell if _this particular_ rule is one of
God's or just one whose limitations we have yet to discover?" No
one has yet provided even an implausible answer. Giving
support to your observation:

Overriding this unresolved debate is the question "Does it matter?"
I think not, for our purposes here.

Bruce