PCT, control theory, and control of perception

[William_T_Powers2]

[From Bill Powers (931220.0915 MST)]
I think I send the wrong file... trying again

Osmo Eerola (931220.0830 GMT) --

I would say that PCTers have developed a certain way to model
behavior, a way they argue is different than control theory
(CT) since in PCT perceptions - not outputs - are controlled.

No, there is no difference between PCT and CT. You can stop
saying we claim that there is a difference, because we do not and
never have claimed that. There is merely a difference in the way
of breaking the whole system down into components, and describing
the behavior of the system. The mathematics is exactly the same
(although we simplify it somewhat). Do you think a theory is
altered simply because of choosing one way to represent the
physical situation rather than another entirely equivalent way?

What does it mean to "control" a variable? Doesn't that mean to
act on the variable to bring it to a desired state, and more
generally to maintain it in that state in the presence of
arbitrary disturbances?

If (using your diagram) we compare the external variable, S, and
the perceptual variable, P, we can ask which variable can be said
to be under control under the widest variety of conditions. Under
disturbances of S, and under changes in the parameters of the
output function G(s), both are equally well controlled. But if
H(s), the perceptual input function, changes its characteristics,
the external variable S will no longer be maintained in the same
reference state. However, the perceptual signal P will still be
maintained in a match with the reference signal. So P remains
under control over a wider range of system perturbations than
does S.

Another consideration is that when we're modeling organisms, the
perceptual signal P is the only variable that the system can know
it is controlling. It has no internal representation of the state
of S other than the signal P. So from the viewpoint of the
controlling system itself, P is the only variable it CAN control.

The main problem seems to be that 'engineers do not understand
behavior' and PCTers do not always understand control theory.

Why do you say that PCTers do not always understand control
theory? Are we doing our simulations wrong? Are we solving the
system equations erroneously? I think you find our
representations strange because they aren't laid out like the
diagrams you are accustomed to and we don't use the same verbal
identifications of the variables. I think you're confusing form
with substance.

Basic control theory is most likely not adequate for the
analysis of behavior, but I still do not see what radically new
PCT brings to basic control theory.

Basic control theory does a better job of explaining behavior
than any theory before it has ever done, including common sense.
You don't know what you're talking about. And you keep insisting
that we say PCT brings something new to control theory. It
doesn't. It's exactly the same theory. We have never said that
PCT is an improvement on control theory. It is not. It IS control
theory.

In my opinion the 'control of perceptions' is not such a
thing.

Is that just off the top of your head, or do you have some reason
for making that statement?

I do not buy this simple assumption [about controlling an
external plant] as a common engineering view. Control engineers
should be aware of the concepts like system, environment,
process, controller etc.

You seem to differ from the ideas we hear from other control
engineers. Most of them seem to consider that the control system
is a signal-handling organization that controls some external
plant, some aspect of the environment like pressure, temperature,
acceleration, chemical concentration, and so forth. When you say
that the organism is the system (plant), you seem to be putting
the plant inside the organism instead of in its environment.

In modelling (individual) behavior the organism is the system
(plant) and it has a clear interface and interactions to its
environment.

In English terminology, the "plant" means something like a
factory or an apparatus for carrying out some chemical or other
process. Some engineers use "system" to mean the combination of
the controlling elements and the environmental variables involved
in the control loop.

In PCT we choose a different way of subdividing the same elements
(we don't use the term "plant"). The "control system" means just
the part of the organism lying between the sensory receptors and
the muscular effectors. ONLY the nervous system is considered to
be the control system; even non-neural aspects of the body are
considered to be in the environment of the control system.

The environment, then, is everything that is affected by the
motor outputs, and that is sensed by the sensory inputs (as well
as everything else outside the organism). That includes internal
aspects of the body which are also sensed and affected by the
nervous system. The environment is governed by the laws of the
physical and chemical sciences.

Many control engineers use this same way of organizing models of
control. The sensors, electronics, and effectors are considered
the controlling part of the whole system; everything outside the
sensor-effector boundary is considered the environment (the
plant), containing whatever is to be controlled.

I prefer this way of subdividing the whole system, because it
makes our analysis standardized and orderly, as well as clearly
identifying what the control model must account for and what is
already accounted for by the physical and biological sciences.
The mathematical treatments are, of course, unaffected -- that
is, control theory itself remains exactly the same.

In many cases the organism wants to bring the environment or
part of it to a certain state. (Other organisms are then part
of the environment.) The organism can also bring itself to a
certain state (satisfied).

Well, what do you mean by satisfied? We would mean that the
difference between reference signals and perceptual signals is
made as small as possible. So "satisfied" means the same thing
whether we are talking about external variables or variables
physically inside the organism (but outside its nervous system).
Perceptions are altered by actions (outputs) until they are in
the state that the organism wants -- until the organism is
satisfied with its perceptions.

Problems will arise when the system starts to model its own
behavior! Turing and Church have some ideas about how well a
system can model itself.

Yes, and we have paid close attention to that problem. The
organism can understand the world, including itself, only in
terms of its perceptual signals. It has no other access to the
outside world. To understand how to model an organism, we must
take the point of view of the organism, not of an external
observer. Are there aspects of ourselves that we can't model?
Probably so, because there are aspects of ourselves that we don't
perceive. But we can model at least those processes that we are
aware of. This proves to be quite practical, and enough to keep
us busy for several generations at least.

There is no other way for a (complex) control system to follow
the effects of the control than via its sensory inputs from the
controlled process or environment. Is not that trivial?! It can
be named as the 'control of perceptions' but there is nothing
new in it.

I'm glad that "control of perception" appears now to be a self-
evident, trivial idea. That's an improvement over your previous
statement that there is no such thing. If the idea is so obvious
to you now, you have grasped the basic principle behind PCT, and
are ready to begin considering its implications. All right ideas
seem obvious once you understand them.

What you have to realize that in psychology and most other life
sciences, it has long been believed that perception controls
behavior: that is, that the environment acts on the nervous
system to cause behavior. Today, in cognitive science, it is
believed that mental activities like plans cause behavior. The
"control of perception" idea is new in both of these areas; it
says that output actions are actively varied in order to bring
perceptions to internally specified states. As a control
engineer, you know that external disturbances are met by changes
in the physical output of a control system, so those physical
outputs can't be controlled: they must be free to vary. As one
who now understands what control of perception means, you will
also see that the output variations are what bring perceptions to
internally-specified states, and maintain them there. And you
will see that "plans" can't specify the physical outputs; all
they can do is call for certain perceptions to appear, with the
control processes providing the actions needed to make those
perceptions come to the required states, even in the presence of
unpredictable disturbances. A goal is a specification for a
perception, not for an action.

In control engineering, the idea of control of perceptions, once
pointed out, seems obvious and trivial. That is as it should be.
It is trivial for the engineer, because the engineer doesn't care
what an artificial control system experiences through its
perceptions. The whole point of artificial systems is to control
an external variable, one that matters in the outside world to
other people (like the customer for whom the control system is
being designed). The perceptual signal is just part of what is
needed to achieve this result.

But in the study of living control systems, control of perception
becomes a central idea. The natural organism does not have the
educated engineer's knowledge of external processes. All it can
know must exist in the form of perceptions. For a living control
system, its perceptions ARE its environment. The emphasis
completely changes. A living control system controls what it
knows of its environment, not the actual environment itself.

The final step in this initial insight is to realize that you are
an organism, too. All that you know of the world is carried into
your brain as perceptual signals; if it were not for those
signals, you would be unaware of any environment. Everything you
see, hear, feel, taste, and smell is a perceptual signal. That
whole world in which you live and on which you act consists
entirely of perceptions. As the slogan goes (it is much more than
a slogan), "It's _all_ perception."

When we model control behavior, we use physics to provide a model
of the environment that connects our actions back to our
perceptions. That physics model, of course, was built up from our
perceptions just as everything else we know was built up. It
represents our best guess as to what might lie outside our
brains. That best guess changes over the decades and centuries;
some day it will be quite different from what it is now. But by
using the physics model, we can complete a model of behavior and
make it internally consistent both with the facts of neurology
(another model) and the facts of physics. The basis of this whole
model is control theory: the mathematical way of dealing with
systems in which causation runs around a closed loop.

···

---------------------------------------------------------------
Best,

Bill P.

[_Martin_Taylor]

[Martin Taylor 931220 1330]
(Osmo Eerola 931220.0830 GMT)

Following Bill Powers' posting (931220.0915 MST), here's one factor that
might help you see the difference in where you look what affects what
in biological control systems. You think it is unimportant where you
look in the loop, because it's all one loop. I used to think the same,
so I appreciate your position, though I now think it is wrong.

What you have to realize that in psychology and most other life
sciences, it has long been believed that perception controls
behavior: that is, that the environment acts on the nervous
system to cause behavior.

Perception->action->enviroment->perception. It's all one loop, right?
Bill pointed out one reason why this is not so. You might prefer another,
independent, reason.

We are NOT dealing with one control loop, even one with a very complex
multidimensional perceptual function, as some of our control engineers
seem to like to consider. We control many independent "things" (which
for the moment I will not call perceptual signals or states of the world).
How many? It is impossible to say, but we can find a limit to how many
could be independently controlled at any single moment (time multiplexing
allows more to be controlled, but at lower bandwidth, so that's a trade-off
that doesn't affect the argument). The limit is the number of degrees of
freedom we have at the narrowest part of the parallel set of control
functions.

We have some 600-800 different muscles, so the limit can be no greater than
that. But by my crude count, we have only around 100-150 different possible
movements of joints and "shaping" actions such as facial gestures and
speech gestures that don't involve joints. Let's say 100, as nice round
number.

If perception were to control behaviour, every perception would have to
be mappable onto some degree of freedom for behaviour. So there would
have to be no more degrees of freedom for perception than there are for
action. Is this the case?

We have some 10^6 fibres in each optic nerve, 3x10^4 in each auditory nerve,
and more for the other senses, including an enormous number of skin sensors.
In principle, though not in practice, each of these fibres could represent
an individual degree of freedom for perception. They do not in practice do
so because (a) they tend to act redundantly to enhance the robustness of the
signal, and (b) the world itself is highly correlated across space and time.
So a large part of the initial processing of these sensory signals is
probably done by passive statistical analysis. Nevertheless, it would be
extraordinary if the initial processing were to reduce the number of degrees
of freedom for perception by three orders of magnitude. And the bandwidths
of these sensory signals tend to be in the hundreds or thousands of Hz,
whereas the bandwidths of the action outputs are typically under 10 Hz
(I can't think of any single joint I can vibrate at 10 Hz, let alone one
I can move randomly with that bandwidth, so I'm being very generous here).

All of this means that the incoming sensory signals provide an enormously
greater variety of perceptual degrees of freedom than are available for
output, each varying much faster than the output actions can do. Yet, in
principle, almost all perceptions that we have (perhaps all) either are
themselves controllable or form part of a more complex perception that is
controllable. We do not (much) deal in useless input or output.

If there are ten, a hundred, or a thousand times more perceptions than
actions active at any moment, it is not logically possible that perception
determines action. It is logically possible that actions determine SOME
perceptions, and that is the core of PCT. Perception does not drive action,
even though the whole loop is action->perception->action.

Incidentally, in your earlier diagram (931216-09.10):

                                                       _______ Energy
                                                      I I flow
                                    ------------------I G2(s) I---->
                        _______ I _______ I_______I
              Comp I I I I I
  Ref. ------->X--------I G(s) I------I H(s) I---I
               I diff I______I S I_______I P I
               I I
               I----------------------------------I

The output to the right, through G2(s) labelled "Energy Flow", is what
we normally in PCT call a "side effect". Not being perceptible, it is
irrelevant to the organism doing the controlling. It may, of course,
have effects that return to haunt the actor later in time, but it is
irrelevant to the behaviour of the control system you have drawn.

So your comment (following) is irrelevant to control by a biological system,
though it may be relevant to an observer of that biological system:

See my figure above. It is not so seldom in artificial (man made)
control systems that you have to measure something else than the
final 'energy flow'.

Martin

[Osmo_Eerola]

From Osmo Eerola (931221.0810 GMT] to Bill Powers (931220.0915 MST)]

to [Martin Taylor 931220 1330]

No, there is no difference between PCT and CT. You can stop
saying we claim that there is a difference, because we do not and
never have claimed that.

Eureka. I will cease. Why do not you use just the term CT? Everything
would me much easier to stupid control engineers. The brand new
insight of controlling perceptions could well be included
in the framework of the (traditional) CT. But I guess that you
need a new term to make a high profile in behavioral sciences.

Do you think a theory is
altered simply because of choosing one way to represent the
physical situation rather than another entirely equivalent way?

Not at all. Vice versa, that is exactly what I have tried to say
to you, that PCT actually is CT - not a 'new science'.
Maybe everything is just a misunderstanding?

If (using your diagram) we compare the external variable, S, and
the perceptual variable, P, we can ask which variable can be said
to be under control under the widest variety of conditions.

...

>The main problem seems to be that 'engineers do not understand
>behavior' and PCTers do not always understand control theory.

The conclusions made from the block diagram are of course valid
(according to CT). In CT terminolgy (by definition) the control(ler)
part of the closed loop system comes after the subtraction of
reference and feedback i.e. the term diff = ref-P is used for
driving (controlling) the process. Therefore the control of
perception (P) sounds so strange to control engineers, but as
said many times, one can say so.

>Basic control theory is most likely not adequate for the
>analysis of behavior, but I still do not see what radically new
>PCT brings to basic control theory.

Basic control theory does a better job of explaining behavior
than any theory before it has ever done, including common sense.

Agree (Hm, a new term BCT), just doubt whether it is adequate.

>In my opinion the 'control of perceptions' is not such a
>thing.

Is that just off the top of your head, or do you have some reason
for making that statement?

Maybe just off the top of my head.

>I do not buy this simple assumption [about controlling an
>external plant] as a common engineering view. Control engineers
>should be aware of the concepts like system, environment,
>process, controller etc.

You seem to differ from the ideas we hear from other control
engineers. Most of them seem to consider that the control system
is a signal-handling organization that controls some external
plant, some aspect of the environment like pressure, temperature,
acceleration, chemical concentration, and so forth. When you say
that the organism is the system (plant), you seem to be putting
the plant inside the organism instead of in its environment.

As regards artificial control systems I agree with the common view.

In PCT we choose a different way of subdividing the same elements
(we don't use the term "plant"). The "control system" means just
the part of the organism lying between the sensory receptors and
the muscular effectors. ONLY the nervous system is considered to
be the control system; even non-neural aspects of the body are
considered to be in the environment of the control system.

So basically you are controlling with the computing element (neurons),
which derives information from the sensory receptors, the muscular
effectors (process, plant). You say, however, you control perception.
Okay, the whole system as described above, still seems to me very
conventional control approach - and there is NOTHING WRONG in it, just
fine!

>There is no other way for a (complex) control system to follow
>the effects of the control than via its sensory inputs from the
>controlled process or environment. Is not that trivial?! It can
>be named as the 'control of perceptions' but there is nothing
>new in it.

I'm glad that "control of perception" appears now to be a self-
evident, trivial idea. That's an improvement over your previous
statement that there is no such thing. If the idea is so obvious
to you now, you have grasped the basic principle behind PCT, and
are ready to begin considering its implications. All right ideas
seem obvious once you understand them.

It is selfevident in terms of CT. A different way of looking, not
a new science.

What you have to realize that in psychology and most other life
sciences, it has long been believed that perception controls
behavior: that is, that the environment acts on the nervous
system to cause behavior.

Old hat S-R system. Not totally wrong: when I see an angry dog
(stimulus) I react (response) by doing something. In that context
we can say that perception controls behavior, but of course
we can analyse the case deeper, in CT terms.

Today, in cognitive science, it is
believed that mental activities like plans cause behavior.

Not totally wrong, either. I can make plans, which cause behavior, but
again, we can go deeper in the analysis (i.e. how it will be done).

In control engineering, the idea of control of perceptions, once
pointed out, seems obvious and trivial. That is as it should be.
It is trivial for the engineer, because the engineer doesn't care
what an artificial control system experiences through its
perceptions.

Exactly.

But in the study of living control systems, control of perception
becomes a central idea. The natural organism does not have the
educated engineer's knowledge of external processes. All it can
know must exist in the form of perceptions. For a living control
system, its perceptions ARE its environment. The emphasis
completely changes. A living control system controls what it
knows of its environment, not the actual environment itself.

Exactly. An artificial control system controls what it knows
of its environment, too.

All that you know of the world is carried into
your brain as perceptual signals; if it were not for those
signals, you would be unaware of any environment>

I do not know anybody denying this.

···

-----------------------------------------------------------
to [Martin Taylor 931220 1330]

Following Bill Powers' posting (931220.0915 MST), here's one factor that
might help you see the difference in where you look what affects what
in biological control systems. You think it is unimportant where you
look in the loop, because it's all one loop. I used to think the same,
so I appreciate your position, though I now think it is wrong.

In traditional CT I do not think it is that important (It of course
helps you to understand the loop better, if you think what is going on
at the different blocks and nodes - but that is part of CT education).
It is important, when you 'try to think what the closed loop
itself is thinking about its functioning'. But I doubt any simple
closed loop can be aware of itself.

What you have to realize that in psychology and most other life
sciences, it has long been believed that perception controls
behavior: that is, that the environment acts on the nervous
system to cause behavior.

See my comment to Bill above on S-R.

If perception were to control behaviour, every perception would have to
be mappable onto some degree of freedom for behaviour. So there would
have to be no more degrees of freedom for perception than there are for
action. Is this the case?

I think so. It seems to be that the 'Shannonian' channel capacity of our
senses (input & receiving) is much higher than the channel capacity
needed for high quality outputs

i.e. for hearing & speech

channel capacity C = B log2(1+S/N)
[C]=1/s
B=channel bandwidth (Hz)
S=transmission power (Watts)
N=noise power (Watts)

HIFI Audio quality transmission (CD) means a C about 700 000 bps whereas
speech bandwidth (about 5000 Hz) and dynamics (about 50 dB) yield about
100 000 bps.

All of this means that the incoming sensory signals provide an enormously
greater variety of perceptual degrees of freedom than are available for
output, each varying much faster than the output actions can do. Yet, in
principle, almost all perceptions that we have (perhaps all) either are
themselves controllable or form part of a more complex perception that is
controllable. We do not (much) deal in useless input or output.

I think you forget one thing. Are not PCTers replacing the S-R behaviorism
by PCT behaviorism. Life is more than behaving. The inputs are needed to
make internal representations of the world. Once there is a representation
in memory (think elderly people) one can almost live without external
perceptions and outputs by just 'running neural programs' with the material
you have in memory.

If there are ten, a hundred, or a thousand times more perceptions than
actions active at any moment, it is not logically possible that perception
determines action.

Yes it is logicallly and in practice possible, but I have never argued,
that perceptions determine (directly) action (I think you refer
here to S-R behaviorusts). You forget that one of the main functions
of the nervous system is to reduce redundancy from the sensory information.
We can make a artificial system, which collects huge amounts of sensory
information just for controlling one on/off function (nuclear defense e.g.).
-----------------------------------

To Rick Marken (931220.1500)

Why the hostility to PCT. Osmo? I'm really curious.

Am I? As I mentioned earlier "I was criticizing the classification
of PCT as a new science - not the PCT itself, or its
developers or appliers." I have tried to keep (about) neutral
'scientific' argumentation.

I don't think PCt will neither "wipe out Finland nor put control engineers
on the unemployment rolls."

I just try to put PCT into its right place
in CT and in my mind! You fellows try to put it into its right
place in behavioral sciences.

In fact, it's kind of a nice little theory --

Sounds much better than 'new science'

Nearly everything you've said about PCT in your posts is incorrect.
From your PCT point of view, not from my CT point of view!

Do you care? Do you know what is wrong with what you said? Bill
Powers pointed out most of the gross errors;

Stupid men do not see their (gross) errors. I appreciate Bill's
patiency with me, but I do not admit that he has shown errors in
my thinking. He has shown only the different way how PCTers use CT.

Is someone saying bad things about PCT
over there in Finland? Do you think PCT advocates the sacrifice of
Finnish babies? Do you think PCTers are clannish? Is there a book
called "The Protocols of the Elders of PCT" circulating around up
there?

I do not understand what Finland and Finnish babies have to do with
'scientific' debate on PCT. I just used one (poor) example refering
to politics in this corner of world, but why using this kind of
argumentation? I apologize if I have hurt you.

I'm really interested in finding out what you dislike about PCT.
Why do you seem to have no interest in trying to understand PCT?

...

But I am completely puzzled by the response of control engineers to
PCT. PCT won't revolutionize control engineering or control theory.
PCT is BASED ON control theory. PCT just deals with HOW control theory
is applied to the behavior of living systems. So what's the problem
with PCT, Osmo? Inquiring minds want to know.

Well, I think after this long discussion I have learned to understand
your way of applying CT. But be more cautios in calling something
a 'new science'.

regards
osmo eerola

[_Martin_Taylor]

[Martin Taylor 931221 13:50]
(Osmo Eerola 931221.0810)

I'm not sure whether it is a question of language, but I think you entirely
inverted my point that was supposed to help you understand why PCT is
the control of perception rather than the control of output.

If perception were to control behaviour, every perception would have to
be mappable onto some degree of freedom for behaviour. So there would
have to be no more degrees of freedom for perception than there are for
action. Is this the case?

I think so. It seems to be that the 'Shannonian' channel capacity of our
senses (input & receiving) is much higher than the channel capacity
needed for high quality outputs

In English, the answer is "I think not." There are many more degrees of
freedom for perception than for action, as you go on to observe. And
from that fact follows the conclusion that perception cannot control
behaviour, but that behaviour could control at least some perception some
of the time.

I think you forget one thing. Are not PCTers replacing the S-R behaviorism
by PCT behaviorism. Life is more than behaving.

I believe that living organisms are physical systems, and need no magic,
mysticism, vital force or anything like that.

The inputs are needed to
make internal representations of the world. Once there is a representation
in memory (think elderly people) one can almost live without external
perceptions and outputs by just 'running neural programs' with the material
you have in memory.

If you have been following the discussions on CSG-L for more than a few
days, you will see that this is a very strong point of PCT. Look back
over the "feedforward" discussion in particular. How do you get to bed
in the dark?

If there are ten, a hundred, or a thousand times more perceptions than
actions active at any moment, it is not logically possible that perception
determines action.

Yes it is logicallly and in practice possible, but I have never argued,
that perceptions determine (directly) action (I think you refer
here to S-R behaviorusts). You forget that one of the main functions
of the nervous system is to reduce redundancy from the sensory information.
We can make a artificial system, which collects huge amounts of sensory
information just for controlling one on/off function (nuclear defense e.g.).

You might get to where you (seem to) want to go, an appreciation of PCT,
by asking yourself why it is true that "one of the main functions of the
nervous system is to reduce redundancy from the sensory information," rather
than simply citing it as something we might like to know. I tried to
provide you with an answer, in the post you to which you were responding.
Now I'll let you think it out for yourself. One understands wheels much
better by reinventing them than by being told how they work.

Martin

[Osmo_Eerola]

From Osmo Eerola [931228 08.45 GMT] to Martin Taylor 931221 13:50]

In English, the answer is "I think not."

Sorry for the slip of my pen. Always had problems with
the "inverse" expressions of English.

There are many more degrees of
freedom for perception than for action, as you go on to observe.
And
from that fact follows the conclusion that perception cannot control
behaviour, but that behaviour could control at least some perception some
of the time.

Cannot see why "multiple input" systems could not control "single output".
(see the last paragraph)

>I think you forget one thing. Are not PCTers replacing the S-R behaviorism
>by PCT behaviorism. Life is more than behaving.

I believe that living organisms are physical systems, and need no magic,
mysticism, vital force or anything like that.

Agree (I think so, too).
I mean that there are other cognitive processes than behaving
(e.g. reading and trying to memorize foreign language words or grammatic,
formulating abstract theories or plans in your mind, understanding spoken
mother tongue).
Or do you study these phenomena in PCT terms, too?
It is not extremly seldom that once a 'nice little theory' has been found,
just everything will be tried to explain by it.

If you have been following the discussions on CSG-L for more than a few
days, you will see that this is a very strong point of PCT. Look back
over the "feedforward" discussion in particular. How do you get to bed
in the dark?

Agree. Take a nice seat (your favourite chair), relax, close your eyes and
go back to the first childhood Christmas you remember. Stay there for a while.
How does PCT apply to your memories?

>>If there are ten, a hundred, or a thousand times more perceptions than
>>actions active at any moment, it is not logically possible that perception
>>determines action.
>
>Yes it is logicallly and in practice possible, but I have never argued,
>that perceptions determine (directly) action (I think you refer
>here to S-R behaviorusts). You forget that one of the main functions
>of the nervous system is to reduce redundancy from the sensory information.
>We can make a artificial system, which collects huge amounts of sensory
>information just for controlling one on/off function (nuclear defense e.g.).

You might get to where you (seem to) want to go, an appreciation of PCT,
by asking yourself why it is true that "one of the main functions of the
nervous system is to reduce redundancy from the sensory information," rather
than simply citing it as something we might like to know. I tried to
provide you with an answer, in the post you to which you were responding.
Now I'll let you think it out for yourself. One understands wheels much
better by reinventing them than by being told how they work.

Well, you are arguing above that since "there are thousand times more
perceptions...., its is not logically possible..." I cannot see the
rationale behind this statement: it is very well possible that multiple
inputs can control a few outputs - the output O(t) is a function of several
inputs I1(t), I2(t),...,In(t). There can even be several sensed (perceived)
variables of a process and all they control the ouput in combination in
a closed loop manner.

regards
Osmo Eerola

[_Martin_Taylor]

[Martin Taylor 931228 12:00]
(Osmo Eerola 931228 08.45)

Cannot see why "multiple input" systems could not control "single output".
(see the last paragraph).
...[last paragraph]
Well, you are arguing above that since "there are thousand times more
perceptions...., its is not logically possible..." I cannot see the
rationale behind this statement: it is very well possible that multiple
inputs can control a few outputs - the output O(t) is a function of several
inputs I1(t), I2(t),...,In(t). There can even be several sensed (perceived)
variables of a process and all they control the ouput in combination in
a closed loop manner.

It depends on what you mean by "control." In a world in which Ik(t) all
vary independently, there is very little relation between I2(t) and O(t),
and I would not think of I2(t) as controlling O(t). If you think it does,
then of course you are right.

In contrast, the PCT view is that O(t) is varies so as to bring I2(t) to
a specific value set by R2(t), ignoring the values of Ik(t) k != 2. Of
course, since these are all time functions and there are transport lags
and other time-based effects, control is imperfect. Nevertheless, I
would say that O(t) is controlling I2(t) in this example. It is not
controlling Ik(t) where k != 2.

It is important to note that I2(t) is a function of many variables, but
these are sensory inputs to the specific ECS in the hierarchy, and they
are NOT controlled by O(t). Only the value of the perceptual function
result is controlled, not the arguments to the perceptual function.
(This is one way of stating the PCT motto "many routes to the same result").

The point at issue is that the limit of control is determined by the
bottleneck in the degrees of freedom that an external observer can see
at different places in the loop. Whatever controls whatever else, there
can be no more controlling variables than the bottleneck degrees of
freedom. At any place in the loop where there are more degrees of freedom,
some are necessarily uncontrolled. In a mobile biological system, there
are more degrees of freedom in the perceptual inputs than in the mechanical
outputs, and therefore most of the perceptual signals are uncontrolled.
It is, I think, a matter of language as to whether the "loop" is controlled
or controlling, but to me, and to most PCT people, it seems natural to
talk of the variable that is stabilized as the one that is "controlled"
and the variable that alters its value to maintain that stability as the
one that is "controlling."

If you accept that use of language, then the stability of the perceptual
signal coupled with the variability of output needed for disturbance
compensation requires you to label perception as "controlled" and
output as "controlling."

···

===============

One point that leads to a lot of confusion when people come at first into
the PCT world is that there are several viewpoints on the control systems.
Two are sufficiently important that they are often the only two I consider:
the empathist who looks at what could be seen (known) at different places
in the control hierarchy, and the modeller or outside observer, who looks
at what can be seen in the outer world or in the set of signals in the
control system when it is in operation. These viewpoints are not easy
to separate, but it is important that one try to separate them.

The empathist sees perceptual control, because the perceptual signal is
the ONLY thing relating to the outer environment that can be affected
in a discoverable way by the organism. The outside observer cannot see
perceptual control. All the outside observer can see is that observer's
own perceptions, based on its own sensory input. The OO assumes that
the observed control system has sensory input from the same physical
world, and tests whether the observed system is controlling by means
of actions on the shared physical world, to see whether some variable
in the world seems to be stabilized by the actions of the observed system.
The OO can see the actions of the observed system, but not its perceptions.
So it is very natural for an outside observer to fall into the trap of
thinking that the observed system is controlling its actions rather than
its (unobservable) perceptions.

The genius of Bill Powers, in my view, was to take the empathist view,
and to notice that it gave a result different from the outside observer
view. There have been many who have come close (and continue to do so),
but apart from those influenced by Powers, I do not know of any who have
made this very important conceptual leap.

Once one recognizes the importance of the distinction between empathist
and OO views, one can become an Engineer or Theorist (Bob Clark's terms),
who can link the two views by imagining the signals that might occur
around the whole loop, both inside and outside the organism or the
individual ECS. They can develop or analyze control systems that might
look to the OO the same as does the real system whose internal operations
are unobservable. They should never imagine that their models ARE what
functions inside the real organism, though, and on CSG-L that mistake is
seldom made.

As you follow the discussions on CSG-L over time, you will probably observe
that most misunderstanding eventually come down to a question of which
viewpoint(s) are being used by the different protagonists. So far, you
have resolutely taken the OO viewpoint, which makes a discussion with
empathists (Marken, and often Powers) very difficult.

Osmo Eerola [931228 09:30

A deaf baby tries for a while to babble, but when he/she cannot hear
anything this behaving ceases. I think this can be used both for
and against the PCT. The baby ceases babbling, since there are no
perceptions to be controlled, but why does the baby start it anyway.

Try to get hold of Bill Powers' book "Behavior: the Control of Perception"
in which similar things (and your questions about imagination and rumination
in another post) are addressed. Also, there are collections of writings
available from Greg Williams. But for now, a quick answer:

The baby doesn't "know" much about the world except what evolution has
provided. There are certain control systems whose value has been demonstrated
to be stable over enough time that random mutations have developed them
and they stick because organisms that have them are more likely to
propagate their genes than organisms that have to develope them individually.

How can a baby find out about the world? You are a control engineer. How
do models get built? By varying the reference signals (inputs) to the
different control systems, randomly or nearly so. In the absence of
knowledge embodied in the linkage structure of the control systems, the
baby just acts. It waves its arms and legs, smiles and cries, and otherwise
exercises its various muscles. Sometimes those motions are followed by
improvements in perceptual states, sometimes not. In other words,
sometimes the baby feels good, sometimes not. According to PCT, a process
called "reorganization" tends to occur more rapidly when the baby does
not feel good. Reorganization takes two basic forms: changing what to
perceive (new or modified perceptual input functions) and changing what
to do (inverting the sense of output links, or changing the connections
of output linkage structures). These changes, in the absence of "knowledge"
about the way the world works, are necessarily random, but they stop when
the baby feels really good. How does the baby come to keep feeling "really
good?" It must be when the "right" perceptual variables are all being
effectively controlled. This never happens perfectly, but a reasonable
level of control is achieved quite quickly.

One perceptual complex that a deaf baby never acquires is that associated
with hearing things. It can't hear other people talk. It can't hear
what its randomly babbling vocal muscles have caused acoustically, when
it feels good or otherwise. A normal baby might cry, and that results
in food, or better, in the reduction of bad (hunger) feelings. Later,
there is more refinement in what babbling does, but if the babbling output
cannout be perceived as sound, it can't be part of reorganization. It is
wasted effort--side effects of other things the baby is "doing." It also
might conflict with some other possible controlled perceptions that the
baby has. So it stops, over time.

In PCT terms, then, your question is backwards. You should ask why the
baby stops babbling, rather than why it starts. Wasted effort in actions
that serve no control purpose is a very subtle reason for stopping those
actions. It involves quite disparate perceptions, in particular those
involved with tiredness and hunger (energy supply). The existence of
those feelings may well be associated with reorganization, and they are
less likely to be experienced when the side-effect actions are not
performed. So, reorganization that leads to their cessation will be
successful, whereas reorganization that augments them won't, and will
lead to further reorganization, until they are substantially reduced
or stopped.

One can say that all perception serves one function - survival.

Rather, it serves one function -- propagation of the genes, which does
not always require survival of the organism past mating, or even up to
mating, if the death of the organism assists the survival of enough
relatives.

Tom Bourbon has answered the rest of what I was intending to deal with.

Hope this helps you to find the empathist viewpoint, to segregate it
from the other viewpoints you will encounter, and to see each for what
it is.

Happy New Year

Martin