Simultaneous Action and S-R Thinking Around the Control Loop

[From Bruce Abbott (2014.01.08.0920 EST)]

BA: I've changed the subject title to reflect the content.

Rick Marken (2014.01.07.1730) --

Adam Matic (2014.01.08. 0010 CET)

AM: Great to see the videos. Just a quick note. The PCT-Part1.m4v
video file is quite large. I converted it down to 50 MB to see how it
will look and
sound:
https://dl.dropboxusercontent.com/u/70399093/PCT-Part1_x264_002.mp4

RM: These work great Adam. And how timely! His talk is about exactly what
we've been talking about on the net lately: 1) the theory explains a
_phenomenon_ so the talk starts with a description of the behavioral
phenomenon to be explained: the production of consistent results by variable
means in the face of disturbances, ie. the phenomenon of control 2) this
phenomenon occurs in a closed loop where all events in the loop are
happening _at the same time_; looking at the loop in sequential terms is a
mistake that leads to predictions that are inconsistent with the observation
(of control).

RM: I highly recommend this series of videos. But this first video gives a
nice, clear introduction to the phenomenon the PCT explains -- control
-- and it's sure nice to see Bill.

BA: Nobody disputes the fact that, in a control loop, all events in the loop
are happening at the same time. Nobody disputes the fact that analyzing
events around the loop as if they were happening one at a time, in sequence,
will give the wrong result when analyzing how the system will behave.
Nevertheless, changes in the values of the variables around the loop do
propagate around the loop in real time, like a wave in a doughnut-shaped
tank moving clockwise around the tank. If the changes all took place
simultaneously, then by definition there would be no such thing as loop
delay. Yet we know that loop delay is always a factor that must be taken
into account if the control system is to be stable -- too much delay and
changes in the system's output get more than 90 degrees out of phase with
the changes in the disturbance that the output changes are supposed to
oppose. The consequence is the conversion of negative feedback into runaway
positive feedback.

BA: But don't take my word for it; here's what Bill Powers actually has to
say. The following excerpts are from the Appendix to B:CP under the heading
of "Stability":

BP: The common beginner's error in analyzing how a control system works is
to trace the effects of an abrupt disturbance step by step around a closed
loop. . . . [Bill gives an example of computations taking place step by step
around the loop] . . . Even though we set up the system for negative
feedback, it is producing not control but wilder and wilder oscillations. .
. . Clearly we have left something out, since a real system does no such
thing. In fact, if any control system were designed so that events took
place sequentially as in our analysis, exactly that kind of extreme
oscillation would inevitably occur, even if the time between events were
microseconds. The problem is not the time scale, but the fact that time
itself is left out of a sequential-state analysis.

BP: There are time delays in every real control system. The output cannot
actually cancel the effect of a disturbance on a controlled quantity at the
instant the disturbance occurs. But in every successful control system,
there is a physical limitation built into the system which in effect slows
the action of the system dynamically. This limitation prevents some variable
in the system--usually its output--from jumping abruptly from one value to
another. If a sudden error occurs implying that the output should suddenly
change from 1 unit to 10 units, the output does not suddenly become 10
units. Rather it BEGINS TO CHANGE toward a new value.

BP: The speed of change is the critical factor. For a system with a given
sensitivity and a given inherent time delay, the speed must be such that
oscillations do not build up.

BP: [Bill introduces the leaky-integrator in the output function without
calling it such, and with it the slowing factor, which allows only a
fraction of the computed change in output to affect the CV on each iteration
of the loop.]

BP: The more sensitive the control system, the smaller fraction of the
calculated correction must be permitted on each round of calculation and the
more slowly must the system change its output if stability is to be
maintained. This is how time can be taken into account in a sequential-state
analysis of a control system. When time is properly taken into account, the
sequential analysis gives the same steady-state result as the
continuous-variable (algebraic) analysis.

BA: So sayeth Bill Powers, so sayeth we all.

Bruce

[Martin Taylor 2014.01.08.10.08]

[From Bruce Abbott (2014.01.08.0920 EST)]

BA: I've changed the subject title to reflect the content.

Good idea.

Long-time participants in CSGnet will remember a period in which Bill P was having a hard time trying to convince Rick that every link in a control loop was a simple S-R structure whose output was completely determined by its inputs (including, of course, the history of the inputs). As I remember, Rick deferred to Bill, but I get the impression that he was one "convinced against his will, who retains his opinion still", his opinion being that the completed closure of the loop somehow invalidates the S-R nature of the individual links. I judge this by several of his recent messages in which he rants on about perception being influenced jointly by output and by disturbance, and about variables simultaneously changing all around the loop, both of which I would have thought not to be in PCT 101, having been dealt with in kindergarten.

PCT 101 probably starts with the fact that although the effects propagate around the loop the S->R direction, which defines "forwards", the analysis of the loop has to proceed in the other direction, deducing S from knowing R for each link in the loop. Both ways of looking at the loop have to start somewhere, as Fred says in a message heavily criticized by Rick for that assertion. To say that the effects in a loop propagate around the loop is not "S-R thinking". It is physics.

Martin

[From Rick Marken (2014.01.09.1430)]

Bruce Abbott (2014.01.08.0920 EST)--

RM: I highly recommend this series of videos. But this first video gives a
nice, clear introduction to the phenomenon the PCT explains -- control
-- and it's sure nice to see Bill.

BA: Nobody disputes the fact that, in a control loop, all events in the loop
are happening at the same time.

RM: Yes, one dispute was about about whether control was a phenomenon
that is explained by control theory. I think control is a phenomenon
and that intentional (purposeful) behavior is control. So control
theory is a theory of purposeful behavior. The other dispute was about
the best way to describe the controlling done by a control system:
"protecting a controlled variable from disturbance" or "compensating
for the effects of disturbance". I think both are fine but I often
prefer the first because the second can imply an S-R process.

BA: Nobody disputes the fact that analyzing
events around the loop as if they were happening one at a time, in sequence,
will give the wrong result when analyzing how the system will behave.
Nevertheless, changes in the values of the variables around the loop do
propagate around the loop in real time, like a wave in a doughnut-shaped
tank moving clockwise around the tank.

RM: The problem with thinking about the control process this way is
that it ignores what is most important about control: the controlled
variable. And a wave propagating around the loop is very much like a
sequential view. The "bump" in the wave is just the state of the cv at
a particular instant in time (it's not the effect of the disturbance
because what is perceived is the joint result of disturbance and
output). The "propagation" is a continuous sequence of changes in the
location of that bump until it gets to the output that is the
"response" to that bump.

While this way of thinking about control may help you remember that
temporal relationships between variables in a control loop are
important for the stability, I think it takes your eye off what I
think is the main PCT "ball": determining the perceptual variables
around which behavior is organized. This is what distinguishes the PCT
approach to understanding behavior from all others, including other
applications of control theory to behavior. Of course, when you use
control models to determine the perceptual variables around which
behavior is organized you have to take temporal relationships between
the variables into account in order to produce a model that behaves
properly (controls the hypothesized controlled variable); and the
addition of a transport lag can often improve the fit of model to
data.

But I believe that if you just focus on the dynamics of control --
which is the aspect of control that non-PCT control theorists have
been focusing on for decades and seems to be the main focus of the
model Martin Taylor posted -- you miss what is most important about
PCT: recognition of the fact that behavior is organized around the
control of perceptual variable and that understanding behavior is a
matter of figuring out what those variable _are_.

Control engineers are all into dynamics because they have to _build_
stable control systems; these engineers are not interested in testing
for controlled variables because they have built these systems to
control particular variables, so they already know what the controlled
variables are. PCT shows us that psychologists are in a very different
position than a control engineer. The systems we study have already
been built and they are quite stable (although there are diseases,
like Parkinson's, that produce instability). What we don't know is
what variables these systems control. I think it's easier to remember
that this is the main goal of PCT research -- figuring out what
variables are controlled -- when you focus on the fact that control
involves protecting these unknown variables -- controlled variables --
from disturbance rather than on the fact that control involves a
somewhat delayed compensatory response to the effects of disturbance.

Best

Rick

···

--
Richard S. Marken PhD
www.mindreadings.com

The only thing that will redeem mankind is cooperation.
                                                   -- Bertrand Russell

Control engineers are all into
dynamics because they have to build

stable control systems; these engineers are not interested in
testing

for controlled variables because they have built these systems to

control particular variables, so they already know what the
controlled

variables are. PCT shows us that psychologists are in a very
different

position than a control engineer. The systems we study have already

been built and they are quite stable (although there are diseases,

like Parkinson’s, that produce instability). What we don’t know is

what variables these systems control. I think it’s easier to
remember

that this is the main goal of PCT research – figuring out
what

variables are controlled – when you focus on the fact that control

involves protecting these unknown variables – controlled variables

from disturbance rather than on the fact that control involves a

somewhat delayed compensatory response to the effects of
disturbance.
[From Dag Forssell (2014.01.09.1610 PST)]

Rick, I think this is a very good point, succinctly stated. Many
thanks.

Dag

Rick Marken (2014.01.09.1430)]

[From Bruce Abbott (2014.01.09.1940 EST)]

Rick Marken (2014.01.09.1430) --

Bruce Abbott (2014.01.08.0920 EST)

BA: Nobody disputes the fact that analyzing events around the loop as
if they were happening one at a time, in sequence, will give the wrong
result when analyzing how the system will behave.
Nevertheless, changes in the values of the variables around the loop
do propagate around the loop in real time, like a wave in a
doughnut-shaped tank moving clockwise around the tank.

RM: The problem with thinking about the control process this way is that it
ignores what is most important about control: the controlled variable. And a
wave propagating around the loop is very much like a sequential view. The
"bump" in the wave is just the state of the cv at a particular instant in
time (it's not the effect of the disturbance because what is perceived is
the joint result of disturbance and output). The "propagation" is a
continuous sequence of changes in the location of that bump until it gets to
the output that is the "response" to that bump.

BA: The bump in the wave moves around the control loop, from CV to input to
perception to error to output to CV. It represents changes in the states of
ALL these variables, not just the CV. Changes are taking place
simultaneously all around the loop, but they are not the same changes that
are taking place simultaneously. The effect visible at the CV is followed
after a brief delay by a change in the perceptual signal, etc. While the
perceptual signal is changing, the CV is undergoing further changes that
have not yet affected the perceptual signal, including further changes
induced by the disturbance and feedback. Do you agree?

RM: While this way of thinking about control may help you remember that
temporal relationships between variables in a control loop are important for
the stability, I think it takes your eye off what I think is the main PCT
"ball": determining the perceptual variables around which behavior is
organized. This is what distinguishes the PCT approach to understanding
behavior from all others, including other applications of control theory to
behavior. Of course, when you use control models to determine the perceptual
variables around which behavior is organized you have to take temporal
relationships between the variables into account in order to produce a model
that behaves properly (controls the hypothesized controlled variable); and
the addition of a transport lag can often improve the fit of model to data.

RM: But I believe that if you just focus on the dynamics of control -- which
is the aspect of control that non-PCT control theorists have been focusing
on for decades and seems to be the main focus of the model Martin Taylor
posted -- you miss what is most important about PCT: recognition of the fact
that behavior is organized around the control of perceptual variables and
that understanding behavior is a matter of figuring out what those variables
_are_.

RM: Control engineers are all into dynamics because they have to _build_
stable control systems; these engineers are not interested in testing for
controlled variables because they have built these systems to control
particular variables, so they already know what the controlled variables
are. PCT shows us that psychologists are in a very different position than a
control engineer. The systems we study have already been built and they are
quite stable (although there are diseases, like Parkinson's, that produce
instability). What we don't know is what variables these systems control. I
think it's easier to remember that this is the main goal of PCT research --
figuring out what variables are controlled -- when you focus on the fact
that control involves protecting these unknown variables -- controlled
variables -- from disturbance rather than on the fact that control involves
a somewhat delayed compensatory response to the effects of disturbance.

BA: Dynamics are important, even when analyzing living control systems.
Anyone who studies control should be familiar with how various factors, such
as loop gain and transport delay, among others, affect how a control system
behaves -- whether it overshoots the reference or approaches it smoothly,
whether it approaches the reference quickly or seems to take forever,
whether it stabilizes quickly or oscillates for a time before settling down.
You mentioned the characteristic tremor of Parkinson's Disease; another
possible example is bipolar disorder, with its unstable swings between mania
and depression. It's the dynamics that distinguish these disorders from
normality.

A good model of the first-order systems that operate the muscles and joints
must exhibit the same pattern of behavior as is actually observed under
various conditions of load, friction, or damping, even when this control
exhibits less than perfect control (e.g., overshoots, brief periods of
oscillation). A control-system model that fails to duplicate the dynamics of
the real thing is likely to be an incorrect model.

I don't share your concern that talking about dynamics will obscure what you
see as a more important focus on CSGnet -- teaching what control is and how
to determine what the controlled variables are. Those who do not feel the
need to learn about dynamics are free to ignore the discussions about them.

Bruce

[Martin Taylor 2014.01.09.17.50]

[From Rick Marken (2014.01.09.1430)]

Bruce Abbott (2014.01.08.0920 EST)--

RM: I highly recommend this series of videos. But this first video gives a
nice, clear introduction to the phenomenon the PCT explains -- control
-- and it's sure nice to see Bill.

BA: Nobody disputes the fact that, in a control loop, all events in the loop
are happening at the same time.

RM: Yes, one dispute was about about whether control was a phenomenon
that is explained by control theory.

Not for the first time, I'm wondering whether you read the same CSGnet I do. I have no memory of such a dispute, and I can't really imagine what it could be about if there was one. Control theory is by definition the theory that explains control, so how could there be a dispute about that? Or, what could be explained by control theory other than control?

I think control is a phenomenon

In which meaning of the word? Here are the choices: 1. something directly observed, 2. something out of the ordinary, and 3 (obsolete). what seems to you to be the correct view. I'm guessing you mean 3.

and that intentional (purposeful) behavior is control. So control
theory is a theory of purposeful behavior.

Anybody on CSGnet not believe that?

  The other dispute was about
the best way to describe the controlling done by a control system:
"protecting a controlled variable from disturbance" or "compensating
for the effects of disturbance". I think both are fine but I often
prefer the first because the second can imply an S-R process.

To me, the wording "protecting a variable from disturbance" seems to imply the existence of a shield or shell that doesn't let the disturbance get to the environmental variable defined by the perceptual function. If you have such a shield, you have no need of control. It's personal preference, obviously, but if you are talking to someone outside of the small group of people who have a PCT background, I suspect they would be more likely to understand what you mean if you use "compensating".

BA: Nobody disputes the fact that analyzing
events around the loop as if they were happening one at a time, in sequence,
will give the wrong result when analyzing how the system will behave.
Nevertheless, changes in the values of the variables around the loop do
propagate around the loop in real time, like a wave in a doughnut-shaped
tank moving clockwise around the tank.

RM: The problem with thinking about the control process this way is
that it ignores what is most important about control: the controlled
variable.

Why would you think it is ignored? Unless you post just in order to show how wrong everybody is except for yourself. What is it that the control system is controlling? What is all the other stuff _for_?

  And a wave propagating around the loop is very much like a
sequential view. The "bump" in the wave is just the state of the cv at
a particular instant in time (it's not the effect of the disturbance
because what is perceived is the joint result of disturbance and
output).

Wrong. That initial bump in the cv IS the effect of the disturbance bump alone, because the effect of the disturbance bump has not yet affected the value of the output. That's why the bump exists in the perceptual waveform at all. Later, when the influence of the disturbance "bump" has begun to appear at the output and then at the environmental variable (which happens slowly because of the integrating output function), it begins to disappear from the perceptual signal.

  The "propagation" is a continuous sequence of changes in the
location of that bump until it gets to the output that is the
"response" to that bump.

While this way of thinking about control may help you remember that
temporal relationships between variables in a control loop are
important for the stability, I think it takes your eye off what I
think is the main PCT "ball": determining the perceptual variables
around which behavior is organized.

We disagree about what is the main PCT "ball". I think it's an important secondary "ball" that helps in the explanation of, say, social interactions but is not critical in doing so. What (to me and perhaps to nobody else) is important about PCT is not what is controlled, because what is controlled varies from moment to moment, but what the theoretical assumption that perceptions are controlled implies for the interactions among entities (organisms, social groups, modules within organisms ...). That is a very rich field of research in PCT.

But I believe that if you just focus on the dynamics of control --
which is the aspect of control that non-PCT control theorists have
been focusing on for decades and seems to be the main focus of the
model Martin Taylor posted

Given the short history of the thread, that's an odd thing to say, isn't it? Bruce had suggested that intuitively he thought he was controlling certain variables in a tracking task, rather than the variable usually taken to be the controlled perception. I showed a model that had proved to fit 1300+ tracks better than any other of the several models I had tested. That best model happened to control the same variables that Bruce had intuited, so I mentioned it to support his conjecture about what were his controlled variables. Which, I guess, justifies your saying that the model ignores the controlled perception and just focuses on the dynamics.

Yeah, right.

It gets boringly repetitive to suggest that it might be more productive if you were to comment on what is written rather than on what you imagine or wish that others had written, and stop posting your repetitions of the same old mantra while pretending that they are comments. Almost as boring as reading your mantra repetitions.

You have a good background and do great demos and experiments, so it would be wonderful if you were to comment carefully on what others write. Your comments might actually advance the science, but I have the impression that you are controlling for that not to happen. It's too bad, because PCT could be, and should be, the foundation of a rich science of social, biological, and psychological systems. To keep repeating that it should ONLY be a search for "the" controlled variable strikes me as very defeatist. It's protecting your private turf by fencing it off to prevent research forays into the big bad world of phenomena (things directly observed).

Martin

[From Rick Marken (2014.01.10.1200)]

Bruce Abbott (2014.01.09.1940 EST)-

BA: While the
perceptual signal is changing, the CV is undergoing further changes that
have not yet affected the perceptual signal, including further changes
induced by the disturbance and feedback. Do you agree?

RM: Yes.

BA: Dynamics are important, even when analyzing living control systems.

RM: Yes, of course.

BA: I don't share your concern that talking about dynamics will obscure what you
see as a more important focus on CSGnet -- teaching what control is and how
to determine what the controlled variables are. Those who do not feel the
need to learn about dynamics are free to ignore the discussions about them.

RM: You have my permission to continue the discussion of dynamics;-)
Actually, we will be getting into some dynamics stuff when we do the
course on LCS III. So dynamics will not be ignored.

Best

Rick

···

--
Richard S. Marken PhD
www.mindreadings.com

The only thing that will redeem mankind is cooperation.
                                                   -- Bertrand Russell

[From Rick Marken (2014.01.10.1250)]

Martin Taylor (2014.01.09.17.50)--

MT: Control theory is by definition the theory that
explains control, so how could there be a dispute about that?

RM: Got me?

MT: Or, what could be explained by control theory other than control?

RM: Indeed. And what could explain control other than control theory?
That's why Bill's observation that "behavior is control" is so
important. Because behavior is control, only control theory can
explain it. All other theories of behavior -- S-R, cognitive
(including information theory, I'm afraid), reinforcement, etc) ,
which I call "causal" (as opposed to "control") theories -- are a
priori wrong because they do not explain control. They persist (as
Bill also pointed out) because they appear to work, due to the
behavioral illusion (the appearance that circumstances --
disturbances--cause behavior). So I believe there is nothing more
important for the eventual triumph of the PCT revolution than getting
psychologists (and all other human beings) to understand that behavior
is control (and, of course, to understand what control is).

RM: I think control is a phenomenon

MT: In which meaning of the word? Here are the choices: 1. something directly
observed, 2. something out of the ordinary, and 3 (obsolete). what seems to
you to be the correct view. I'm guessing you mean 3.

RM: No, I intend meaning 1: something directly observed. You can
directly observe a variable being maintained in a preselected state,
protected from disturbance. It's what you observe in the basic
tracking task when the person is carrying out the task successfully.

RM: And a wave propagating around the loop is very much like a
sequential view. The "bump" in the wave is just the state of the cv at
a particular instant in time (it's not the effect of the disturbance
because what is perceived is the joint result of disturbance and
output).

MT: Wrong. That initial bump in the cv IS the effect of the disturbance bump
alone, because the effect of the disturbance bump has not yet affected the
value of the output.

RM: I'm afraid your "wrong" is wrong. The output is always having an
effect on the CV. Time lags in the loop mean that the state of the
output at time t is related to the state of the CV at time t - tau
(tau being the duration of the transport lag in the loop). But the CV
at time t - tau is equal to the sum of the effects of disturbance and
output at time t-tau.

CV(t-tau) = d(t-tau) +o(t-tau)

The state of the CV at an instant is _always_ a simultaneous, combined
result of disturbance and output.

MT: That's why the bump exists in the perceptual waveform at all.

RM" The "bump" is a bump in the current value of the CV. And if we
look at the "bump" that is the cause of output at time t we get the
above: CV(t-tau) = d(t-tau) +o(t-tau). So the "bump" could be due to a
large value of d(t-tau) and a small value of o(t-tau) or a small value
of d(t-tau) and a large value o(t-tau). All that the control system
knows is the value of the "bump" -- the CV -- and the output (after a
delay of tau seconds) changes appropriately.

So it's really the CV that determines the (future) output; the
disturbance alone doesn't determine the (future) output. But because
the delayed output to CV happens in a closed loop the output will end
up varying in opposition to the disturbance, and in so doing it
protects the controlled variable from the effects of the disturbance.
The system does this by adding it's own effects (of its output) to
those of the disturbance(s) to the controlled variable. So the control
system is compensating for the disturbance but it is not doing this by
reacting to the effects of the disturbance (alone) to the controlled
variable; it's reacting to changes in the controlled variable itself,
changes that depend (simultaneously) on both disturbance and output.
That's why I prefer (often) to talk about control systems as acting to
_protect_ the controlled variable from disturbance, rather than acting
to compensate for the effects of disturbance. But, again, both ways of
saying it are correct since they are describing the same thing:
control.

MT: We disagree about what is the main PCT "ball".

RM: Yes, indeed we do.

MT: You have a good background and do great demos and experiments, so it would
be wonderful if you were to comment carefully on what others write. Your
comments might actually advance the science, but I have the impression that
you are controlling for that not to happen. It's too bad, because PCT could
be, and should be, the foundation of a rich science of social, biological,
and psychological systems. To keep repeating that it should ONLY be a search
for "the" controlled variable strikes me as very defeatist. It's protecting
your private turf by fencing it off to prevent research forays into the big
bad world of phenomena (things directly observed).

RM: I disagree, of course. But it will take a whole paper to explain
why. But for now I'll just say that I think the search for controlled
variables is central to PCT. It is because you would only do this
search if you understood that the behavior of living organisms _is_
control and that, therefore, understanding their behavior is a matter
of understanding what variables they control. Controlled variables are
to the science of living control systems what atoms are to the science
of chemistry: the basic unit of analysis of control.

Best

Rick

···

--
Richard S. Marken PhD
www.mindreadings.com

The only thing that will redeem mankind is cooperation.
                                                   -- Bertrand Russell

[Martin Taylor 2013.01.10.16.42]

[From Rick Marken (2014.01.10.1250)]

Martin Taylor (2014.01.09.17.50)--
MT: Control theory is by definition the theory that
explains control, so how could there be a dispute about that?

RM: Got me?

MT: Or, what could be explained by control theory other than control?

RM: Indeed. And what could explain control other than control theory?
That's why Bill's observation that "behavior is control" is so
important. Because behavior is control, only control theory can
explain it.

Ah, yes. We can partially agree on that. But "behaviour" is a different class of concept from "control", isn't it? "Behaviour" is something an outside observer can see happening. "Control" is a process largely hidden from the observer.

  All other theories of behavior -- S-R, cognitive
(including information theory, I'm afraid), reinforcement, etc) ,
which I call "causal" (as opposed to "control") theories -- are a
priori wrong because they do not explain control. They persist (as
Bill also pointed out) because they appear to work, due to the
behavioral illusion (the appearance that circumstances --
disturbances--cause behavior). So I believe there is nothing more
important for the eventual triumph of the PCT revolution than getting
psychologists (and all other human beings) to understand that behavior
is control (and, of course, to understand what control is).

We can totally agree on that. But why you call "information theory" a theory of behaviour is beyond me. It is as much a theory of behaviour as is Fourier or Laplace transform analysis. And what it has to do with "cognitive" I cannot imagine.

RM: I think control is a phenomenon

MT: In which meaning of the word? Here are the choices: 1. something directly
observed, 2. something out of the ordinary, and 3 (obsolete). what seems to
you to be the correct view. I'm guessing you mean 3.

RM: No, I intend meaning 1: something directly observed. You can
directly observe a variable being maintained in a preselected state,
protected from disturbance. It's what you observe in the basic
tracking task when the person is carrying out the task successfully.

But we can't agree on that. You forget the second mantra of PCT: "You can't see what someone is doing by looking at what they are doing". Yes, there are occasions when you can see that there should be an effect of a disturbance on something you can observe, and you can see that an organism is able to sense the effects of that disturbance, and you can see the action that counters those effects. In the lab, doing the appropriate experiments, you can indeed see all of that. The TCV is based on that ability, which the tester sets up "on purpose". But most of the time in the real world, you can't.

RM: And a wave propagating around the loop is very much like a
sequential view. The "bump" in the wave is just the state of the cv at
a particular instant in time (it's not the effect of the disturbance
because what is perceived is the joint result of disturbance and
output).

MT: Wrong. That initial bump in the cv IS the effect of the disturbance bump
alone, because the effect of the disturbance bump has not yet affected the
value of the output.

RM: I'm afraid your "wrong" is wrong.

Two wrongs make a right. We are talking about the bump, the deviation of the controlled variable from what it was before the disturbance bump happened, not about the total influence of disturbance and output on the controlled variable. Initially, the bump is solely due to the disturbance, because the effect of the bump has not yet reached the output, let alone the environmental variable. When it does, the bump begins to disappear. Then and only then, is the size and shape of the bump affected jointly by the disturbance and the output.

That ought to be kindergarten PCT, but I guess it isn't.

The state of the CV at an instant is _always_ a simultaneous, combined
result of disturbance and output.

Of course it is. I wish you wouldn't keep assuming I'm not yet ready for Control 101, and must be kept in kindergarten.

So it's really the CV that determines the (future) output; the
disturbance alone doesn't determine the (future) output.

Of course.

I'm not sure who you are writing for, but I don't think it is anyone who has been reading CSGnet for more than a few weeks.

MT: We disagree about what is the main PCT "ball".

RM: Yes, indeed we do.

MT: You have a good background and do great demos and experiments, so it would
be wonderful if you were to comment carefully on what others write. Your
comments might actually advance the science, but I have the impression that
you are controlling for that not to happen. It's too bad, because PCT could
be, and should be, the foundation of a rich science of social, biological,
and psychological systems. To keep repeating that it should ONLY be a search
for "the" controlled variable strikes me as very defeatist. It's protecting
your private turf by fencing it off to prevent research forays into the big
bad world of phenomena (things directly observed).

RM: I disagree, of course. But it will take a whole paper to explain
why.

What do you disagree about? That it would be wonderful if you were to comment carefully on what others write?

What would this paper for? To explain why PCT should not be the foundation of a science of social, biological, and psychological systems? To explain why PCT should ONLY be used to search for what variables are controlled by _this_ person in _that_ situation at _this_ moment?

  But for now I'll just say that I think the search for controlled
variables is central to PCT. It is because you would only do this
search if you understood that the behavior of living organisms _is_
control and that, therefore, understanding their behavior is a matter
of understanding what variables they control.

That's rather a reversed piece of logic isn't it? Because understanding that behaviour is control of perception allows you to understand that controlled variables exist, therefore the only research you should do is look for that/those variable(s) in specific individual situations? Because understanding that atoms exist, the only research scientists should do is seek what particular atoms are in a piece of stuff? How they fit together should not be studied? Certainly, without understanding control, one will not search for controlled variables; without understanding control, one will not search for the effects of gain and transport lag in Parkinsonian tremor, either. Should we say that because we understand control, therefore we should only study gain and transport lag in Parkinsonianism? It makes exactly as much sense.

I certainly agree that if you want to figure out what _this_ organism will do when disturbed in _that_ way, you need to know what variables it is controlling. But you also need to know how it has reorganized and what effordances it has that provide opportunities for its actions to influence the controlled variables. Discovering what variables it controls is necessary, but far from sufficient, even in if you are interested only in that very particular situation at that particular moment. A minute later, outside the laboratory, the organism will probably be controlling a quite different set of variables. So why bother trying to figure out just which of the thousands of variables an organism might be controlling over the course of a day or a month it is controlling at this particular moment, unless it is that organism at that moment at the centre of your concern? Wouldn't it be much better to work on the implications of the fact that it IS controlling while perhaps others are also controlling in the same environment?

  Controlled variables are
to the science of living control systems what atoms are to the science
of chemistry: the basic unit of analysis of control.

I don't think that's a workable metaphor, but use it if you want. I think the basic unit of the analysis of control is the control loop, but to each his own. The reason I think as I do is that without the completed loop, there is no control, no controlled variable. To the loop, the controlled variable has much the same importance as does the proton number to the chemistry of an atom. It is important, but it is not everything. The atom, with all its electron shells and potential for interaction with other atoms, is what makes chemistry. The control loop, with all its potential for interaction with other control loops, is what makes life.

Martin

[From Rick Marken (2014.01.10.1812)]

Martin Taylor (2013.01.10.16.42)--

RM: Because behavior is control, only control theory can
explain it.

MT: Ah, yes. We can partially agree on that. But "behaviour" is a different
class of concept from "control", isn't it? "Behaviour" is something an
outside observer can see happening. "Control" is a process largely hidden
from the observer.

RM: Not at all. Control is completely visible. Even William James (to
whom Bill referred in the that first lecture) knew how to demonstrate
it without knowing anything about control (he called it "purpose") or
control theory. You introduce disturbances to a variable that seems to
be under control, like the path of filings to a magnet or of Romeo to
Juliet (James' examples). The disturbance has little effect on the
variable that is under control but is fully effective on the variable
that is not. So the filings are stoped by an interfering card; Romeo
is not stopped by an interfering wall.

RM: No, I intend meaning 1: something directly observed. You can
directly observe a variable being maintained in a preselected state,
protected from disturbance.

MT: But we can't agree on that. You forget the second mantra of PCT: "You can't
see what someone is doing by looking at what they are doing".

RM: You can't see what someone is doing by _just looking_ at what they
are doing. But you can see what they are doing by manipulating an
independent variable (disturbance) and looking for _lack_ of expected
effect on a possible controlled variable. It's true, however, that you
can't tell _precisely_ what a person is doing by carrying out this
testing procedure; you have to do a lot of testing to determine
exactly what variable -- what aspect of one's own perceptions --
represents the perceptual variable that is actually under control by
the controller. But just this rather informal implementation of this
testing procedure can tell you whether you are dealing with control or
not. For example, you can tell whether you are being followed --
whether the distance between you and the car behind you is under
control -- by making random left and right turns. If the car stays
being you then it's a good bet that the distance between you and the
other car -- or a variable that is very closely related to that
distance -- is under control. Control theory is not needed to see that
control (intention or purpose) is going on here.

MT: Two wrongs make a right. We are talking about the bump, the deviation of the
controlled variable from what it was before the disturbance bump happened,
not about the total influence of disturbance and output on the controlled
variable. Initially, the bump is solely due to the disturbance, because the
effect of the bump has not yet reached the output,

RM: Your sequential S-R assumptions are showing. Think of it in terms
of a tracking task. The output (mouse movement) is occurring
continuously as is the disturbance. At any instant in time, t, the
position of the cursor (the CV) is a joint function of disturbance and
output: CV(t) = d(t) + o(t). Due to the transport lag in the system,
the output at time t, o(t), depends on the state of the CV at time
t-tau. So these is always an output happening when the disturbance is
happening even though it is an output whose cause occurred tau seconds
earlier. Any "bump" in a CV results in an output that happens tau
seconds later, joining with the new disturbance that also happens tau
seconds later.

In a feedback loop there is never a time when only the disturbance is
affecting the CV; the output is always affecting the CV as well. Maybe
what is confusing you is that sometimes a person appears to be
producing no output -- such as not moving the mouse when doing a
tracking task -- when a sudden change in the value of a disturbance
occurs. But not moving the mouse is not producing no output; it's just
that the output is at 0. Output is a _variable_ so just standing still
(for example) is an output that is affecting you sensory inputs; those
inputs are different than they would be if you were moving.

MT: That ought to be kindergarten PCT, but I guess it isn't.

RM: No I think it's just looking at control as an S-R process:
disturbance causes bump causes output causes reduction in bump causes
reduction in output. It doesn't work that way, as Bill has shown and
as I have shown in several paper and demos (see in particular
Stimulus-Response vs. Control)

RM: The state of the CV at an instant is _always_ a simultaneous, combined
result of disturbance and output.

MT: Of course it is. I wish you wouldn't keep assuming I'm not yet ready for
Control 101, and must be kept in kindergarten.

RM: But you keep saying that it's the disturbance that causes the bump
in the CV that results in the output that corrects the disturbance. If
you know that the state of he CV is a simultaneous, combined result of
_both_ disturbance and output, then I can't see how you could continue
to think of the process of control this way.

RM: So it's really the CV that determines the (future) output; the
disturbance alone doesn't determine the (future) output.

MT: Of course. I'm not sure who you are writing for, but I don't think it is anyone who has
been reading CSGnet for more than a few weeks.

RM: I'm writing it for you, Martin, though it seems many others on the
net see things your way so I'm writing it for them too. If you know
that a CV = o + d then o cannot possibly be caused by d; it's caused
by d + o, which is a very different thing because it means that part
of what causes the output of a control system is the output of the
control system. So the cause of the output of a control system is not
an independent variable (which d is); it is the controlled variable.
The result of this closed loop causal process is that outputs that
compensate nearly perfectly for the disturbance; so it _looks like_
the disturbance is the cause of these outputs. But it's not; that's an
_illusion_: the behavioral illusion.

RM: But for now I'll just say that I think the search for controlled

variables is central to PCT. It is because you would only do this
search if you understood that the behavior of living organisms _is_
control and that, therefore, understanding their behavior is a matter
of understanding what variables they control.

MT: That's rather a reversed piece of logic isn't it? Because understanding that
behaviour is control of perception allows you to understand that controlled
variables exist, therefore the only research you should do is look for
that/those variable(s) in specific individual situations?

RM: I said that the search for controlled variables is _central_ to
PCT research; I didn't say it's the only thing PCT researchers should
study. Possible hierarchical (or heterarchical or whatever)
relationships between control systems is another important area of for
research; we'll deal with an interesting aspect of hierarchical
control in chapter 5 of LCS III. And of course dynamics of control are
very interesting, particularly in disease (like Parkinsons) and
conflict (like binge-purge eating disorders).

But all research on living control system must start with knowledge of
what variables are under control when we study a particular behavior
(like limb position control; is it muscle length, rate of change in
length, force, some combination of these). And even if we don't do
formal testing for controlled variables we have to be aware in our
modeling that we are making assumptions about the perceptual variables
that are under control and that our assumptions might be wrong.

Certainly parameters like gain, slowing and transport lag can affect
the dynamic behavior of a control system; but before we start studying
how these parameters affect the dynamics we have to know what the
system is controlling; a system that is controlling optical velocity
is not the same as one that is controlling optical acceleration, for
example, and the effect of control parameters on the dynamics of
control are going to be quite different depending on what variable is
controlled.

MT: So why bother trying to figure out just which of
the thousands of variables an organism might be controlling..Wouldn't it
be much better to work on the implications of the fact that it IS
controlling while perhaps others are also controlling in the same
environment?

RM: The implications of the fact that a behavior involves control
(purpose) is that it is organized around the control of some
perceptual variable(s). Catching a fly ball is a behavior that clearly
involves control. If you want to understand that particular example of
control you have to figure out _which_ perceptions are being
controlled and _how_ they are being controlled. That's how I see PCT
research progressing; figureing out the controlled variables that are
the basis of different examples of behavior (control) and figure out
how control of those variables results in the behavior (control) that
we observe. that what I think Bruce Abbott is doing, for example, with
some work being done on trying to understand how people control the
position of their limbs.

Best

Rick

···

--
Richard S. Marken PhD
www.mindreadings.com

The only thing that will redeem mankind is cooperation.
                                                   -- Bertrand Russell

[Martin Taylor 2014.01.11.10.43]

[From Rick Marken (2014.01.10.1812)]

A long message in three parts. Part 1 chastizes me by expanding on what I said about being able to determine the controlled variable if you can see or manipulate a disturbance to the putative environmental correlate of the controlled perception, can ascertain that the organism cna detect the effect of the disturbance, and can ascertain that the organism's output can influence the environmental variable.

Part 2 demonstrates that Rick can't tell the difference between the value of a variable and a change in the value of that variable.

I don't propose to answer in detail, but I would just like Rick to note that if x = y+z, and y changes while z stays the same, x changes by (strange coincidence) the value of the change in y, no matter what the value of z.

Part 3 relaxes his position that the search for the controlled variable is the only permissible PCT research, and agrees at some length with what I had said about when it is appropriate and necessary. The writing, however, suggests that he thinks he is correcting me.

I don't think Rick and I are so very far apart on what actually happens in control, except on a few matters in which our different intellectual backgrounds become important. I come from engineering (both academically and through my heritage), and I believe he comes from psychology.

I started as an engineer and had considered doing graduate work in control systems. Since I got into psychology, I have always addressed it as primarily an engineering problem ("what influences what", "what are the practical limits", and "where are the feedback mechanisms are typical questions"). When I first became acquainted with research psychology during my Master's work in Operational Research, one of my most respected advisors (and boss when I got a job) was trying to get the Canadian Psychological Association, of which he was President, to issue a recommendation that anyone intending to go into research psychology should avoid undergraduate psychology, as most of it would have to be unlearned before the student could do effective research. When you get into PCT, there's even more unlearning to do. Luckily, I didn't have to unlearn as much as most psychologists would need to unlearn, and I don't think I have succumbed to the missionary complex of the converted, mainly because I found PCT after having had an appropriate background to understand its engineering implications, and after having independently developed a theory of communication that later turned out to be a special case of PCT.

Martin

···

Martin Taylor (2013.01.10.16.42)--

RM: Because behavior is control, only control theory can
explain it.

MT: Ah, yes. We can partially agree on that. But "behaviour" is a different
class of concept from "control", isn't it? "Behaviour" is something an
outside observer can see happening. "Control" is a process largely hidden
from the observer.

RM: Not at all. Control is completely visible. Even William James (to
whom Bill referred in the that first lecture) knew how to demonstrate
it without knowing anything about control (he called it "purpose") or
control theory. You introduce disturbances to a variable that seems to
be under control, like the path of filings to a magnet or of Romeo to
Juliet (James' examples). The disturbance has little effect on the
variable that is under control but is fully effective on the variable
that is not. So the filings are stoped by an interfering card; Romeo
is not stopped by an interfering wall.

RM: No, I intend meaning 1: something directly observed. You can
directly observe a variable being maintained in a preselected state,
protected from disturbance.

MT: But we can't agree on that. You forget the second mantra of PCT: "You can't
see what someone is doing by looking at what they are doing".

RM: You can't see what someone is doing by _just looking_ at what they
are doing. But you can see what they are doing by manipulating an
independent variable (disturbance) and looking for _lack_ of expected
effect on a possible controlled variable. It's true, however, that you
can't tell _precisely_ what a person is doing by carrying out this
testing procedure; you have to do a lot of testing to determine
exactly what variable -- what aspect of one's own perceptions --
represents the perceptual variable that is actually under control by
the controller. But just this rather informal implementation of this
testing procedure can tell you whether you are dealing with control or
not. For example, you can tell whether you are being followed --
whether the distance between you and the car behind you is under
control -- by making random left and right turns. If the car stays
being you then it's a good bet that the distance between you and the
other car -- or a variable that is very closely related to that
distance -- is under control. Control theory is not needed to see that
control (intention or purpose) is going on here.

MT: Two wrongs make a right. We are talking about the bump, the deviation of the
controlled variable from what it was before the disturbance bump happened,
not about the total influence of disturbance and output on the controlled
variable. Initially, the bump is solely due to the disturbance, because the
effect of the bump has not yet reached the output,

RM: Your sequential S-R assumptions are showing. Think of it in terms
of a tracking task. The output (mouse movement) is occurring
continuously as is the disturbance. At any instant in time, t, the
position of the cursor (the CV) is a joint function of disturbance and
output: CV(t) = d(t) + o(t). Due to the transport lag in the system,
the output at time t, o(t), depends on the state of the CV at time
t-tau. So these is always an output happening when the disturbance is
happening even though it is an output whose cause occurred tau seconds
earlier. Any "bump" in a CV results in an output that happens tau
seconds later, joining with the new disturbance that also happens tau
seconds later.

  In a feedback loop there is never a time when only the disturbance is
affecting the CV; the output is always affecting the CV as well. Maybe
what is confusing you is that sometimes a person appears to be
producing no output -- such as not moving the mouse when doing a
tracking task -- when a sudden change in the value of a disturbance
occurs. But not moving the mouse is not producing no output; it's just
that the output is at 0. Output is a _variable_ so just standing still
(for example) is an output that is affecting you sensory inputs; those
inputs are different than they would be if you were moving.

MT: That ought to be kindergarten PCT, but I guess it isn't.

RM: No I think it's just looking at control as an S-R process:
disturbance causes bump causes output causes reduction in bump causes
reduction in output. It doesn't work that way, as Bill has shown and
as I have shown in several paper and demos (see in particular
Stimulus-Response vs. Control)

RM: The state of the CV at an instant is _always_ a simultaneous, combined
result of disturbance and output.

MT: Of course it is. I wish you wouldn't keep assuming I'm not yet ready for
Control 101, and must be kept in kindergarten.

RM: But you keep saying that it's the disturbance that causes the bump
in the CV that results in the output that corrects the disturbance. If
you know that the state of he CV is a simultaneous, combined result of
_both_ disturbance and output, then I can't see how you could continue
to think of the process of control this way.

RM: So it's really the CV that determines the (future) output; the
disturbance alone doesn't determine the (future) output.

MT: Of course. I'm not sure who you are writing for, but I don't think it is anyone who has
been reading CSGnet for more than a few weeks.

RM: I'm writing it for you, Martin, though it seems many others on the
net see things your way so I'm writing it for them too. If you know
that a CV = o + d then o cannot possibly be caused by d; it's caused
by d + o, which is a very different thing because it means that part
of what causes the output of a control system is the output of the
control system. So the cause of the output of a control system is not
an independent variable (which d is); it is the controlled variable.
The result of this closed loop causal process is that outputs that
compensate nearly perfectly for the disturbance; so it _looks like_
the disturbance is the cause of these outputs. But it's not; that's an
_illusion_: the behavioral illusion.

RM: But for now I'll just say that I think the search for controlled

variables is central to PCT. It is because you would only do this
search if you understood that the behavior of living organisms _is_
control and that, therefore, understanding their behavior is a matter
of understanding what variables they control.

MT: That's rather a reversed piece of logic isn't it? Because understanding that
behaviour is control of perception allows you to understand that controlled
variables exist, therefore the only research you should do is look for
that/those variable(s) in specific individual situations?

RM: I said that the search for controlled variables is _central_ to
PCT research; I didn't say it's the only thing PCT researchers should
study. Possible hierarchical (or heterarchical or whatever)
relationships between control systems is another important area of for
research; we'll deal with an interesting aspect of hierarchical
control in chapter 5 of LCS III. And of course dynamics of control are
very interesting, particularly in disease (like Parkinsons) and
conflict (like binge-purge eating disorders).

But all research on living control system must start with knowledge of
what variables are under control when we study a particular behavior
(like limb position control; is it muscle length, rate of change in
length, force, some combination of these). And even if we don't do
formal testing for controlled variables we have to be aware in our
modeling that we are making assumptions about the perceptual variables
that are under control and that our assumptions might be wrong.

Certainly parameters like gain, slowing and transport lag can affect
the dynamic behavior of a control system; but before we start studying
how these parameters affect the dynamics we have to know what the
system is controlling; a system that is controlling optical velocity
is not the same as one that is controlling optical acceleration, for
example, and the effect of control parameters on the dynamics of
control are going to be quite different depending on what variable is
controlled.

MT: So why bother trying to figure out just which of
the thousands of variables an organism might be controlling..Wouldn't it
be much better to work on the implications of the fact that it IS
controlling while perhaps others are also controlling in the same
environment?

RM: The implications of the fact that a behavior involves control
(purpose) is that it is organized around the control of some
perceptual variable(s). Catching a fly ball is a behavior that clearly
involves control. If you want to understand that particular example of
control you have to figure out _which_ perceptions are being
controlled and _how_ they are being controlled. That's how I see PCT
research progressing; figureing out the controlled variables that are
the basis of different examples of behavior (control) and figure out
how control of those variables results in the behavior (control) that
we observe. that what I think Bruce Abbott is doing, for example, with
some work being done on trying to understand how people control the
position of their limbs.

Best

Rick

[From Rick Marken (2014.01.11.1030)]

Martin Taylor (2014.01.11.10.43) --

MT: Part 2 demonstrates that Rick can't tell the difference between the value of
a variable and a change in the value of that variable.

I don't propose to answer in detail, but I would just like Rick to note that
if x = y+z, and y changes while z stays the same, x changes by (strange
coincidence) the value of the change in y, no matter what the value of z.

RM: But what Rick knows that you don't know is that if x = y + z then
x will change if y changes and z remains constant, _or_ if z changes
and y remains constant _or_ if both x and y change by different
amounts. For example, if x changes from 5 to 7, then it could be
because y increased by 2 while z remained constant or because z
increased by 2 while y remained constant or because y increased by 4
while z decreased by 2 or because y increased by 1 while z increased
by 1.

Substitute q.o for x, d for y and o for z (q.o = d + o) and you can
see that a change in the value of q.o (the controlled variable) could
have resulted from an infinite number of different changes in the
disturbance, d, and output, o. A system that is controlling q.o has no
way of "knowing" what is causing changes in q.o; and it doesn't have
to. All it has to know is the state of q.o and produce outputs, o, in
proportion to the difference between the intended (reference) and
current state of that variable (perception).

The effect of the delay between the perception of q.o and the
resulting output, o, depends on the rate of change in the disturbance
as well as the slowing characteristics of the control system itself.
That's why the performance of a control system depends on the
frequency spectrum of the disturbances to which the controlled
variable is subject.

Best

Rick

···

--
Richard S. Marken PhD
www.mindreadings.com

The only thing that will redeem mankind is cooperation.
                                                   -- Bertrand Russell

[Martin Taylor 2014.01.11.15.01]

[From Rick Marken (2014.01.11.1030)]

Martin Taylor (2014.01.11.10.43) --

The effect of the delay between the perception of q.o and the resulting output, o, depends on the rate of change in the disturbance as well as the slowing characteristics of the control system itself. That's why the performance of a control system depends on the frequency spectrum of the disturbances to which the controlled variable is subject.

Well, at least you got that right. I hope you remember it when I use that fact in later discussions.

The unquoted part of your message also seems technically correct, but since it is quite irrelevant to the preceding messages in the thread, I didn't check the math carefully.

Martin

[From Rick Marken (2014.01.11.1340)]

Martin Taylor (2014.01.11.15.01) –

MT: The unquoted part of your message also seems technically correct, but since
it is quite irrelevant to the preceding messages in the thread, I didn’t

check the math carefully.

RM: Here’s the unquoted part of my message:

RM: But what Rick knows that you don’t know is that if x = y + z then

x will change if y changes and z remains constant, or if z changes
and y remains constant or if both x and y change by different
amounts. For example, if x changes from 5 to 7, then it could be
because y increased by 2 while z remained constant or because z

increased by 2 while y remained constant or because y increased by 4
while z decreased by 2 or because y increased by 1 while z increased
by 1.

Substitute q.o for x, d for y and o for z (q.o = d + o) and you can

see that a change in the value of q.o (the controlled variable) could
have resulted from an infinite number of different changes in the
disturbance, d, and output, o. A system that is controlling q.o has no
way of “knowing” what is causing changes in q.o; and it doesn’t have

to. All it has to know is the state of q.o and produce outputs, o, in
proportion to the difference between the intended (reference) and
current state of that variable (perception).

Could you please tell me how this is “quite irrelevant to the preceding messages in the thread”? I thought it was relevant to what you said, which was:

MT: Part 2 demonstrates that Rick can’t tell the difference between the value of a variable and a change in the value of that variable.

I don’t propose to answer in detail, but I would just like Rick to note that if x = y+z, and y changes while z stays the same, x changes by (strange coincidence) the value of the change in y, no matter what the value of z.

Best

Rick

···


Richard S. Marken PhD
www.mindreadings.com

The only thing that will redeem mankind is cooperation.
– Bertrand Russell

It’s irrelevant because you are talking about the value of a
variable, whereas the topic was the progression of the changes in
the loop variables after a change in the disturbance. The VALUE of
the perceptual signal, for example, is determined by the output and
the disturbance jointly, but after a change in the disturbance, the
CHANGE in the value of the perceptual signal is unaffected by
consequent changes in the output until the lapse of at least one
loop transport lag. The CHANGE in the value of the perceptual signal
is influenced initially only by the CHANGE in the disturbance (what
we were calling the “bump”.
Martin

···

On 2014/01/11 4:37 PM, Richard Marken
wrote:

[From Rick Marken (2014.01.11.1340)]

      > Martin Taylor (2014.01.11.15.01) --



      >MT:  The unquoted part of your message also seems

technically correct, but since

      > it is quite irrelevant to the preceding messages in the

thread, I didn’t

      > check the math carefully.



      RM: Here's the unquoted part of my message:
        RM:

But what Rick knows that you don’t know is that if x = y + z
then

        x will change if y changes and z remains constant, _or_ if z

changes

        and y remains constant _or_ if both x and y change by

different

        amounts. For example, if x changes from 5 to 7, then it

could be

        because y increased by 2 while z remained constant or

because z

        increased by 2 while y remained constant or because y

increased by 4

        while z decreased by 2 or because y increased by 1 while z

increased

        by 1.



        Substitute q.o for x, d for y and o for z (q.o = d + o) and

you can

        see that a change in the value of q.o (the controlled

variable) could

        have resulted from an infinite number of different changes

in the

        disturbance, d, and output, o. A system that is controlling

q.o has no

        way of "knowing" what is causing changes in q.o; and it

doesn’t have

        to. All it has to know is the state of q.o and produce

outputs, o, in

        proportion to the difference between the intended

(reference) and

        current state of that variable (perception).
    Could you please tell me how this is "quite irrelevant to the

preceding messages in the thread"? I thought it was relevant to
what you said, which was:

      MT: Part 2 demonstrates that Rick can't tell the difference

between the value of a variable and a change in the value of
that variable.

      I don't propose to answer in detail, but I would just like

Rick to note that if x = y+z, and y changes while z stays the
same, x changes by (strange coincidence) the value of the
change in y, no matter what the value of z.

[Martin Taylor 2014.01.12.09.51]

Here's an example in graphic form. A simple control loop with a pure

integrator output function and otherwise straight-through unity gain
functions has a transport lag of 1 second in each of its connections
(if you prefer, you could assign the lag to the relevant processing
rather than to the connector). The total transport loop lag is 4
seconds. At the start of the graphic, the system has been undisturbed for a
while, so all the variables have steady values of zero (reference
assumed to be zero). One second into the graphic, the disturbance
has a “bump” of 6 seconds duration. The graphic traces the effect of
the bump around the loop. It shows what the prematurely posted text
says, but maybe it will be easier to understand, and to understand
why the “unquoted part” of Rick’s message was irrelevant to the
prior discussion.
Martin

image00148.jpg

···

Sorry, but my message was posted prematurely. Here, I hope, is the
complete message.

  It's irrelevant because you are talking about the value of a

variable, whereas the topic was the progression of the changes in
the loop variables after a change in the disturbance. The VALUE of
the perceptual signal, for example, is determined by the output
and the disturbance jointly, but after a change in the
disturbance, the CHANGE in the value of the perceptual signal is
unaffected by consequent changes in the output until the lapse of
at least one loop transport lag. The CHANGE in the value of the
perceptual signal is influenced initially only by the CHANGE in
the disturbance (what we were calling the “bump”.

    On 2014/01/11 4:37 PM, Richard Marken

wrote:

[From Rick Marken (2014.01.11.1340)]

        > Martin Taylor (2014.01.11.15.01) --



        >MT:  The unquoted part of your message also seems

technically correct, but since

        > it is quite irrelevant to the preceding messages in the

thread, I didn’t

        > check the math carefully.



        RM: Here's the unquoted part of my message:
          RM: But what Rick knows that you don't

know is that if x = y + z then

          x will change if y changes and z remains constant, _or_ if

z changes

          and y remains constant _or_ if both x and y change by

different

          amounts. For example, if x changes from 5 to 7, then it

could be

          because y increased by 2 while z remained constant or

because z

          increased by 2 while y remained constant or because y

increased by 4

          while z decreased by 2 or because y increased by 1 while z

increased

          by 1.



          Substitute q.o for x, d for y and o for z (q.o = d + o)

and you can

          see that a change in the value of q.o (the controlled

variable) could

          have resulted from an infinite number of different changes

in the

          disturbance, d, and output, o. A system that is

controlling q.o has no

          way of "knowing" what is causing changes in q.o; and it

doesn’t have

          to. All it has to know is the state of q.o and produce

outputs, o, in

          proportion to the difference between the intended

(reference) and

          current state of that variable (perception).
      Could you please tell me how this is "quite irrelevant to the

preceding messages in the thread"? I thought it was relevant
to what you said, which was:

        MT: Part 2 demonstrates that Rick can't tell the difference

between the value of a variable and a change in the value of
that variable.

        I don't propose to answer in detail, but I would just like

Rick to note that if x = y+z, and y changes while z stays
the same, x changes by (strange coincidence) the value of
the change in y, no matter what the value of z.

[From Rick Marken (2014.01.12.1140)]

image00148.jpg

···

Martin Taylor (2014.01.12.09.51)–

Here’s an example in graphic form…

RM: This is a perfect example of a open-loop analysis of a closed-loop process. By looking only at a 7 second time interval during which the output is constant for the first 3 seconds, the increase in the disturbance variable that occurs at time 0 appears to be the start of a causal chain that cause an increase in output 3 seconds later.

This is pure S-R – disturbance causes output – and it gives a completely misleading picture of how control works. It’s the picture one can get when control is described as “compensation for disturbance” (rather than “protecting a CV from disturbance”) since that phrase can suggest that the compensating output is caused by the disturbance. It’s an S-R picture of control.

What’s wrong with this picture can be seen by considering a different 7 second time segment of the control process, one with the same disturbance but with the output going negative rather than being constant during the first 3 seconds of the time period. The negative output at the start of the time period is the result of changes in the loop variables like those shown in the last 4 seconds of the 7 second segment shown in the diagram above. In this case, with the output starting off moving negative, you will see that the apparent “response” to the disturbance after a 3 second delay is not output starting to go negative but not changing at all. In this different time segment, where the output begins the segment going negative, the increase int he disturbance e at the start of the interval will result in no change in output 3 seconds later. In other words, the “stimulus” (disturbance) now causes no “response”.

If the S-R analysis in the diagram above were correct – if, that is, it were appropriate to analyze control under the assumption that disturbances only have their effects when the output variable is constant – then we should be able to observe a delayed relationship (strong negative correlation) between the state of the controlled variable (v in the diagram) and system output, o. But Bill has shown (as have I in several papers) that no such relationship exists, a least when control is good.

So why does this matter? I guess I think it matters because it seems to take one’s eye – particularly the eyes of PCT researchers – off the main “ball” of PCT: the controlled perceptual variables around which observed behavior is organized. I think this was the point Bill was making in his “Bucket of Beans” paper (and talk) that is reprinted in LCS II. If you believe disturbances, or the effects thereof on controlled variables, to be the cause of the outputs, then it becomes perfectly ok to continue to study behavior in S-R terms. This is basically what “conventional” or “manual” control theorists have been doing since Kenneth Craik introduced conrol theory into the study of behavior back in 1947 (Craik,
K. J. W. (1947) Theory of the human operator in control systems. British. J. of Psychology, 38, 56-61). It’s not PCT and I’m hoping to get people to join me (help me out, really) in doing PCT research.

Best

Rick.


Richard S. Marken PhD
www.mindreadings.com
The only thing that will redeem mankind is cooperation.
– Bertrand Russell

[Martin Taylor 2014.

Absolute nonsense. It is a moment by moment description of what

happens in ANY control loop when the disturbance changes. It is, of
course, in every signal added to everything consequent on what
happened earlier, but that’s NOT what we were talking about. We were
talking about what happens when a control system in balance, with
all signals zero, is disturbed by a “bump” in the disturbance. The
diagram shows what happens with a particular bump that was chosen to
be longer than the loop transport lag.
If you want to represent the entire dynamic set of waveforms for an
arbitrary disturbance waveform over a long time, you can do just the
same, but use one “bump” per sample of the input and add all the
results together. Then you will see the output changing and
contributing to the perceptual signal value along with the
disturbance, but the trace of each individual “bump” will be just
like the above.
Yes, each link in a control loop IS an S-R subsystem. No, the
control analysis is not an S-R analysis.
Bill P. tried to explain to you why you were wrong about this years
ago. He’s not here to correct you now, unfortunately. I do the best
I can, but I think I stop now, as wilful ignorance is a strongly
controlled perception, and I don’t have his Papal authority over the
loose canon.
Martin

image00148.jpg

···

On 2014/01/12 2:39 PM, Richard Marken
wrote:

[From Rick Marken (2014.01.12.1140)]

              Martin Taylor

(2014.01.12.09.51)–

Here’s an example in graphic form…

        RM: This is a perfect example of a

open-loop analysis of a closed-loop process. By looking only
at a 7 second time interval during which the output is
constant for the first 3 seconds, the increase in the
disturbance variable that occurs at time 0 appears to be the
start of a causal chain that cause an increase in output 3
seconds later.

        This is pure S-R -- disturbance causes output --  and it

gives a completely misleading picture of how control works.
It’s the picture one can get when control is described as
“compensation for disturbance” (rather than “protecting a CV
from disturbance”) since that phrase can suggest that the
compensating output is caused by the disturbance. It’s an
S-R picture of control.

[From Bruce Abbott (2014.01.12.1725 EST)]

I can’t resist jumping in here.

RM: Rick Marken (2014.01.12.1140)]

Martin Taylor (2014.01.12.09.51)–

Here’s an example in graphic form…

image00148.jpg

RM: This is a perfect example of a open-loop analysis of a closed-loop process. By looking only at a 7 second time interval during which the output is constant for the first 3 seconds, the increase in the disturbance variable that occurs at time 0 appears to be the start of a causal chain that cause an increase in output 3 seconds later.

BA: That is because it IS the start of a causal chain that causes an increase in output 3 seconds later. The scenario assumed here is that we have steady conditions prior to the change in the value of the disturbing variable (not just in the preceding 7 seconds, but in the long term) – constant input (v), perception (p), error (e), and output (o). (Note that this does NOT mean that the output is zero; it would be nonzero if the output were currently acting to oppose a steady nonzero value of the disturbing variable. In that case there would also be a small amount of steady error, the amount depending on the loop gain of the system.)

In Martin’s example, there is a 1-s delay between each “station” around the loop. At time t=0 the value of the disturbing variable changes. (It doesn’t matter which direction; Martin’s example has it increasing, so I’ll use an increase here.) The disturbance adds some amount to the value of the CV. This increase is transmitted at t=1 s to the input function, causing (in S-R fashion) an increase in the input function’s output, the perception p. One second later this increase in p reaches the comparator. Given that the reference value is constant, the error e increases, S-R fashion. One second after that, the increase in e reaches the output function, causing an increase in o, in S-R fashion. One second later the change in output reaches v, completing the first transit around the loop. Martin assumes that the output function includes an leaky integrator, so the output decreases (the sign of change must be negative if the output is to oppose the disturbance) slowly and only a portion of the full output initially affects the controlled variable, v as we begin the second cycle around the loop. If the increased value of the disturbing variable remains constant, the slowly increasing output will offset more and more of the increase in the disturbance until the system again reaches steady state with only a small, constant residual error, just enough to maintain output at its the final value.

RM: This is pure S-R – disturbance causes output – and it gives a completely misleading picture of how control works. It’s the picture one can get when control is described as “compensation for disturbance” (rather than “protecting a CV from disturbance”) since that phrase can suggest that the compensating output is caused by the disturbance. It’s an S-R picture of control.

BA: And it’s a completely accurate one under the assumed conditions.

RM: What’s wrong with this picture can be seen by considering a different 7 second time segment of the control process, one with the same disturbance but with the output going negative rather than being constant during the first 3 seconds of the time period. The negative output at the start of the time period is the result of changes in the loop variables like those shown in the last 4 seconds of the 7 second segment shown in the diagram above. In this case, with the output starting off moving negative, you will see that the apparent “response” to the disturbance after a 3 second delay is not output starting to go negative but not changing at all. In this different time segment, where the output begins the segment going negative, the increase int he disturbance e at the start of the interval will result in no change in output 3 seconds later. In other words, the “stimulus” (disturbance) now causes no “response”.

BA: The analysis presented was designed to show how a step-change in the disturbance value propagates around the loop under initially steady-state conditions. It’s just a particularly clear example showing how changes propagate around the loop. It was not designed to model the general case.

BA: When you bring in the case in which the system is not at equilibrium (as in your example of the output “going negative”), then of course it is not possible to ignore the effects of an ongoing, changing output on those variables. The effect of the disturbance on the controlled variable is now mixed with the effect of the output on that same variable, and what propagates around the loop is that combined effect. But the variables involved around the loop, from v to p to e to o and back to v, are still changing in “S-R” fashion, one after the other according to the loop delays, those changes propagating around the loop in that order, not simultaneously. If the changes in output happen to exactly match, in the opposing direction, the changes in v that would have been produced by an unopposed disturbance, then I suppose you could say that v was “protected” from the disturbance, but such an outcome is basically a matter of chance. Most of the time the effects of the disturbance on v are merely reduced, although in the case of fast changes in the disturbance variable, there are times when the system will be producing an output that is in the wrong direction (the change in output hasn’t caught up with the change in disturbance) and actually will augment the disturbance rather than attenuate it.

RM: If the S-R analysis in the diagram above were correct – ** if, that is, it were appropriate to analyze control under the assumption that disturbances only have their effects when the output variable is constant** – then we should be able to observe a delayed relationship (strong negative correlation) between the state of the controlled variable (v in the diagram) and system output, o. But Bill has shown (as have I in several papers) that no such relationship exists, a least when control is good.

BA: I’ve bolded and underlined the red herring in the above-quoted paragraph. Neither Martin nor I have suggested any such thing. Martin’s analysis was valid under the conditions he stated for it: A system in which no changes are occurring at time t=0. Neither Martin nor I have ever suggested that this same analysis applies when the system is initially In some non-equilibrium state. If the output is going negative at the beginning, then the changes in v during the first transit of the loop are no longer purely a function of the changes in the disturbance. Similarly, if the disturbance takes the form of continuously varying changes, then it becomes difficult to analyze what happens during subsequent transits when the combined effects of varying disturbance and varying output must be tracked. The simplifying assumption of a step function provides a clear example of the process under discussion: the transmission of changes around the loop in forward order, as opposed to all these particular changes induced by the changed disturbance value taking place simultaneously.

RM: So why does this matter? I guess I think it matters because it seems to take one’s eye – particularly the eyes of PCT researchers – off the main “ball” of PCT: the controlled perceptual variables around which observed behavior is organized. I think this was the point Bill was making in his “Bucket of Beans” paper (and talk) that is reprinted in LCS II. If you believe disturbances, or the effects thereof on controlled variables, to be the cause of the outputs, then it becomes perfectly ok to continue to study behavior in S-R terms. This is basically what “conventional” or “manual” control theorists have been doing since Kenneth Craik introduced conrol theory into the study of behavior back in 1947 (Craik, K. J. W. (1947) Theory of the human operator in control systems. British. J. of Psychology, 38, 56-61). It’s not PCT and I’m hoping to get people to join me (help me out, really) in doing PCT research.

Martin and I have been using the term “S-R” to refer to the way in which variables and signals are transformed in their respective functions within the control loop. You put a specific value in (S), you get a specific value out (R). This is a far cry from asserting that closed loop control can be analyzed correctly as a stimulus-response open-loop system. It’s an assertion that neither of us are making.

Bruce

[From Rick Marken (2014.01.12.1430)]

···

[Martin Taylor 2014.

        RM: This is a perfect example of a

open-loop analysis of a closed-loop process…

MT: Absolute nonsense.

RM: It doesn’t seem that absolute to me;-)

MT: It is a moment by moment description of what

happens in ANY control loop when the disturbance changes.

RM: Actually, no. What you have presented is a moment to moment description of what happens in a PARTICULAR control loop during a period during which the output happens to be constant during the first 3 seconds of that period.

MT: It is, of

course, in every signal added to everything consequent on what
happened earlier, but that’s NOT what we were talking about. We were
talking about what happens when a control system in balance, with
all signals zero,

RM: Right, we are talking about that PARTICULAR control loop, during a period when all signals start at zero.

MT: If you want to represent the entire dynamic set of waveforms for an

arbitrary disturbance waveform over a long time, you can do just the
same, but use one “bump” per sample of the input and add all the
results together. Then you will see the output changing and
contributing to the perceptual signal value along with the
disturbance, but the trace of each individual “bump” will be just
like the above.

RM: Yes, you can do this; but the control system can’t. The control system’s output is based on the deviation of the controlled variable from the reference specification. The state of the controlled variable is at all times the combined result of disturbance and output. It’s the state of the controlled variable that matters to the control system; the relative degree to which that state depends on the effects of disturbance(s) and output is unknown to the control system; all it knows – and all it needs to know – is the state of the controlled variable (pacem John Keats).

MT: Yes, each link in a control loop IS an S-R subsystem. No, the

control analysis is not an S-R analysis.

RM: The control analysis is not an S-R analysis but your analysis of the behavior of the control system is. Although each link in a control loop is a causal (S-R) link, the causal links are arranged in a circle. So when you do the control analysis properly you find that the relationship between disturbance and output does not reflect a causal path through the organism, as your analysis implies, but rather reflects the inverse of the feedback (environmental) connection between output and controlled variable. That is, in a control loop, disturbances to a controlled variable cannot be considered the cause, via the organism, of the outputs that oppose their effect on that variable. Taking the disturbance to be a cause of outputs (as it appears to be due to it’s very high negative correlation with output) is succumbing to the behavioral illusion.

MT: Bill P. tried to explain to you why you were wrong about this years

ago. He’s not here to correct you now, unfortunately. I do the best
I can, but I think I stop now, as wilful ignorance is a strongly
controlled perception, and I don’t have his Papal authority over the
loose canon.

RM: Since the behavioral illusion was one of Bill’s most important discoveries – and surely his most devastating for conventional psychology, I find it hard to believe that Bill would have corrected my analysis. But whether he would or not is irrelevant to me. I think Bill’s greatest gift to the world (to me anyway) is not a set of canon law but, rather, a way of thinking, which I guess I would call the “scientific” approach to evaluating ideas. The most important aspect of this way of thinking was carefully demonstrating things to oneself to see if the (often verbal but sometimes mathematical) claims pan out in fact. The claim you make in your analysis, which is that the effect of the disturbance on the CV is a cause of output (albeit a cause that lags behind its effect) in a control task such as the compensatory tracking task, is one that was the subject of my very first research on PCT (Marken,
R. S. (1980) The Cause of Control Movements in a Tracking Task. Perceptual and Motor Skills, 51,
755-758) and it turns out to be demonstrably false.

But maybe I’m jumping to conclusions about what your conclusion is based on your analysis. It looks to me that your analysis suggest that there is a causal connection between disturbance and output going through the organism (with the output delayed relative to disturbance). That is, you analysis suggests a causal path as follows:

Environment | System | Environment

d -->cv -->|-->p-->C-->e-->|-->o

This suggests that you can study characteristics of the organism (System) by manipulating disturbances, d, and seeing their effects on observed behavior, o. Is this a correct conclusion based on your analysis, or am I misunderstanding you?

Best

Rick

Richard S. Marken PhD
www.mindreadings.com
The only thing that will redeem mankind is cooperation.

                                               -- Bertrand Russell