[spam] Re: Elements in the PCT model

[Richard_Marken2]

[From Rick Marken (2006.11.15.1900)]

Bjorn Simonsen (2006.11.15,11:30 EUST)

Here is an example about what I mean.

A pupil can�t just leave the classroom. He asks the teacher to leave. This question is a disturbance for the teacher. The disturbance results in a perceptual signal (perception), and an error. This error result in an Output variable (action), the words �Yes please�. The Feedback effect in this instance is that the pupil is allowed to leave the classroom.

A pupil can certainly "just" leave the classroom. But if the pupil is controlling for following a rule that says "You must get permission from the teacher before leaving the class" then he will ask the teacher to leave. The pupil's question will be a disturbance to the extent that it affects the state of a perception the teacher is controlling. For example, if the teacher is controlling for answering questions, then the question is a disturbance to this perception because it is an unanswered question. This disturbance doesn't cause a perceptual signal; it changes the state of an existing perception, the perception of the number of questions left to be answered. Before the pupil asked the question this perception was in the state "no questions left to answer"; when the pupil asked the question, the perception went to the state "one question to answer". Assuming the teacher's reference for this perception is "no questions left to answer", there is an error (the difference between the reference "no questions left to answer" and the perception "one question to answer") and the teacher acts to reduce the error by answering the question (by saying "Yes, please"), bringing the perception back to the reference state by countering the "question" disturbance. The feedback effect in this example is the teacher hearing herself answer the student's question; the teacher's answer bring the perception of number of questions left to be answered from one back to none.

In this case, by the way, the disturbance is perceived; it is part of the perception of the controlled perceptual variable "number of unanswered questions".

Now I'll try to say the same thing again in Norwegian....

Kidding;-)

Skol

Rick

···

---
Richard S. Marken Consulting
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400

[Bj_rn_Simonsen]

[From Bjorn Simonsen (20066.11.19, 13:15 EUST)]

Martin Taylor - received 2006.11.17, 23:33 EUST.

I think the one output quantity has different

effects on the outer world because of the outer
world, not because

of all the control systems through which the
output quantity works

its magic. Do me we disagree?

We don’t disagree, if you change “not
because” into “and because”. If

you leave it as “not because”, then we
disagree.

Reading this statement again, both you and Bill have encouraged
my exactness. If you think upon the control system controlling the perceptions
you control, at one or another level (which I read now), the output signal
certainly must have different effects on the outer world.

I’ll try to read more exact in the future and maybe we
will avoid some disagreements.

But it’s a mystery to me how this or the colour

comments relate to the question which was how to
describe in pure PCT

“trying to see my wife’s face” or
"wanting to leave the room with the

teacher’s permission".

The colour comments had nothing to o with the teacher’s
permission. My comments were related to your statement below.

I don’t think so. I grant you that we sense only
physical variables,

but the variations in the perceptual variables
don’t seem to

correspond in any meaningful way to physical units
measured in the

language of grams, metres, and seconds.

I also think that the perceptual signals at the level
we control are more dependent on different perceptual signals at lower levels
and of the input quantity. Still I think the feedback effect at the peripheral
first level is dependent on physical units in the language of grams, metres and
seconds.

Am I too far away, or shall we stop here?

bjorn

···

[Martin_Taylor2]

[From Bjorn Simonsen (20066.11.19, 13:15 EUST)]
Martin Taylor - received 2006.11.17, 23:33 EUST.

>... I grant you that we sense only physical variables,
>but the variations in the perceptual variables don't seem to
>correspond in any meaningful way to physical units measured in the
>language of grams, metres, and seconds.

I also think that the perceptual signals at the level we control are more dependent on different perceptual signals at lower levels and of the input quantity. Still I think the feedback effect at the peripheral first level is dependent on physical units in the language of grams, metres and seconds.

Am I too far away, or shall we stop here?

I think, to use an English metaphor (I hope it translates well) that you are barking up the wrong tree entirely. Maybe you are even in the wrong forest.

Consider the physical train of events leading from the person's muscular influences on the environment to the physical effects of the environment on the sensors. And let's confine ourselves to control of the perceived colour of something that is very readily influenced. Turning a knob that alters the frequency of a spectral line is the simplest means I can think of. You may think of some other, but I doubt it will affect the argument.

The person wants to control the perceived colour of the line to match some reference value. It doesn't matter how this reference value is derived.

The current state of the perception differs from the reference value, so the control unit emits some output that (eventually) results in muscle movements that change the knob angle. Changes in the knob angle cause (note "cause" rather than "influence", because I'm not even considering the possible existence of a disturbance at this point).

Now think of the physical effects and transformations in this part of the loop. Neural impulses might be measured in microwatts, an energy per second measure, so that the energy expended in neural impulses during the control movement can be measured in microwatt-seconds. These impulses, by various chemical means, cause microfibres in the muscles to contract. There's a gain here, because they use energy already chemically stored in the muscles. But is there a function relating neural microwatt-seconds to muscular gm-cm. I doubt it, becuase the effect will depend on how tired the muscle is, and on how tense the muscle was before the movement began. But for the sake of argument, let's say that at any instant during the move there is a gain factor. It would be a pure number, since we are comapring energy rates on both ends of the transform.

Now we have the muscles expending energy, some of which is dissipated as heat, some of which goes to turn the knob. Let's say that the knob rotates a diffraction grating, so it's very easy. Nevertheless, there is always loss. The diffraction grating has inertia, its bearings have friction, and in the end, we have a stationary diffraction grating, just as we did before. ALL the muscle energy has been dissipated into the environment. There is NO gain factor relating muscular output to changes in the position of the diffraction grating.

What does the diffraction grating do? It splits a broadband source into a spectrum. Any particular wavelength is directed in a different direction from any other wavelength. It doesn't influence the energy into or out of it. The different spectral wavelengths may have different powers, but that's a property of the source, unaffected by the rotational angle of the grating. The person's muscular output may affect the power being delivered in a particular direction, but there isn't any functional relation between hw much energy is output by the muscles and how much power is delivered to the beam that reaches the person's eye.

Now consider that last transformation before we dive beneath the person's skin. Photons from the beam directed toward the eye hit receptor cones in the retina. Some repsond more strongly to red, some to green, and some to blue. The output of each is in a power measure (by way of counting neural impulses) of microwatts, as is the power of the beam. There's a potential gain factor here. The power in the neural impulses is gained from energy stored in the cones, released by the effect of incoming photons. The amount of energy in an impulse isn't related to the power in the beam, though the number of impulses per second is. That relation changes very quickly, though, as the energy supply in a cone is depleted, and as the effects of neural interconnections take hold (e.g. lateral inhbition). It's not a gain factor to be relied on when calculating control loop gain.

So, what do we have when looking for a gain factor between output neural impulaes feeding the muscles and neural impulses up the optic nerve? Not a lot. The number of neural impulses up the optic nerve is likely to change very little as the spectrum changes, though the relationship between the numbers in the different colour channels will indeed change. Let's just use Red/Green as an example, and say that the perceived colour is a function of (rate of "Red" impules) / (rate of "Green" impulses). That's a raw (dimensionless) number. So the gain factor must have a dimension of "per microwatt". But we already saw that there could be at least one place in the feedback path where the output micorwatts have no relation at all to the sensory input. They don't have a functional relationship to the rotation of the grating.

So, in at least one case of a maximally peripheral and extremely simple (unbraided) control loop, we can't trace a gain function, and we don't seem to have even a relation between output energy and input energy before computing the ratio of "Red" to "Green" impulses.

···

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

All of that long analysis was in response to your " Still I think the feedback effect at the peripheral first level is dependent on physical units in the language of grams, metres and seconds."

I am saying that it isn't, even if you can measure the power of the light that enters the eye, or of the sound vibrations that move the eardrum, and even if the perception being controlled is at the lowest possible level in the hierarchy.

The units in which the gain can be measured are commensurate only inside the control unit: for example, the analyst could compute how much a unit change in the error value would alter the perceptual signal if the connection were broken between the perceptual signal and the comparator. And so forth.

Martin

[Bill_Powers1]

[From Bill Powers (2006.11.20.0810 MST)]

Martin Taylor (2006.11.17, 23:33 EUST) --

So, in at least one case of a maximally peripheral and extremely simple (unbraided) control loop, we can't trace a gain function, and we don't seem to have even a relation between output energy and input energy before computing the ratio of "Red" to "Green" impulses.

The problem with goal-oriented reasoning is that if you think you already know the right answer, the premises you choose in constructing an argument will be determined by whether they produce that answer. This leads to overlooking counterexamples as well as simple answers, and mixing qualitative with quantitative arguments.

There is a very simple way to measure the gain in the environmental part of a control loop. Calculate the partial derivative of the perceptual signal with respect to the first-order output quantity. If both the perceptual signal and the output quantity can vary, there will be such a partial derivative. Its value is the gain of the output function.

Gain is simply a logarithmic way of expressing a factor of proportionality. A negative gain implies a proportionality factor less than one. In the case you mention, there is a very large attenuation in the feedback loop, most of the output energy being used to overcome frictional losses and to operate the neural.detection equipment (sensing any input stimulus costs much more energy than is in the effective stimulus).

As I pointed out some years ago, at least 10 and perhaps 15, the energy budget of acting and sensing has nothing to do with the transmission of information. Information can easily move in a direction opposite to the direction of energy transfer. Information has nothing to do with energy (or entropy). Consider the ball that slowly falls during the countdown to the New Year in Times Square. Whoever or whatever is transmitting that signal is receiving, not sending, energy as the ball gives up its potential energy.

···

=========================
Up a level.

The point that you and Bjorn are both missing is that the physics of behavior is a proposal, a model, not a description of Real Reality. You're arguing about whether the physics model and the PCT model are being correctly applied. All arguments of this kind about "is" and "is not" are really arguments about who is following the rules of the model more correctly. No resolution can be reached through purely verbal arguments. There is simply no point in trying to settle any matter of this kind without setting up a formal argument, agreeing on what the data are, and following agreed-upon rules to derive conclusions. Then the only questions are whether you have considered all relevant data, measured accurately, and reasoned without blunders. If you have handled the preliminaries properly, there is no room left for disagreement. If there are still disagreements, then the participants are using different data, different means of measurement, or different rules of reasoning. In that case, they should be arguing about data selection, measuring methods, or methods of reasoning -- not about conclusions.

Best,

Bill P.

[Martin_Taylor2]

[From Bill Powers (2006.11.20.0810 MST)]

Martin Taylor (2006.11.17, 23:33 EUST) --

So, in at least one case of a maximally peripheral and extremely simple (unbraided) control loop, we can't trace a gain function, and we don't seem to have even a relation between output energy and input energy before computing the ratio of "Red" to "Green" impulses.

The problem with goal-oriented reasoning is that if you think you already know the right answer, the premises you choose in constructing an argument will be determined by whether they produce that answer. This leads to overlooking counterexamples as well as simple answers, and mixing qualitative with quantitative arguments.

There is a very simple way to measure the gain in the environmental part of a control loop. Calculate the partial derivative of the perceptual signal with respect to the first-order output quantity.

I think that's what I've been trying to get across. That you have to do it that way, and you can't use the energy budget of the physical effects in the feedback loop. However, just to niggle, this partial derivative has to be done open-loop.

As I pointed out some years ago, at least 10 and perhaps 15, the energy budget of acting and sensing has nothing to do with the transmission of information.

That's also something I've been trying point out, most recently by illustrating it throuhg a worked out example.

The point that you and Bjorn are both missing is that the physics of behavior is a proposal, a model, not a description of Real Reality.

I don't think either of us has missed that point. We have both mentioned it early in the thread, but we don't harp on it as the thread develops.

You're arguing about whether the physics model and the PCT model are being correctly applied. All arguments of this kind about "is" and "is not" are really arguments about who is following the rules of the model more correctly. No resolution can be reached through purely verbal arguments.

I'm afraid I have to dispute that. Mathematics is just another kind of verbal argument, after all. Moreover, in a situation in which the premises and relationships cover a wide range of possibilities, it is often vry difficult to create an intelligible mathematical description from which generic conclusions can be derived. I'd be interested, for example, in how you would presnet in an intelligble form, a mathematical demonstration of this:

As I pointed out some years ago, at least 10 and perhaps 15, the energy budget of acting and sensing has nothing to do with the transmission of information. Information can easily move in a direction opposite to the direction of energy transfer. Information has nothing to do with energy (or entropy).

which is a perfectly reasonable and easily understood statement. But for someone to understand it, they have to have an appreciation of what the different terms imply, and how they interrelate. To "mathematize" it, you have to include all those relevant definitions and explanations. Whose version of entropy? Which derivation of information? (And by the way, I think you will find that contemporary physics is coming back to the idea that there is a direct connection between entropy and information; They do go to the trouble of working out the mathematical formulation, but it takes a lot of education to understand what they do with the maths, education I don't have).

···

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

There is simply no point in trying to settle any matter of this kind without setting up a formal argument, agreeing on what the data are, and following agreed-upon rules to derive conclusions. Then the only questions are whether you have considered all relevant data, measured accurately, and reasoned without blunders. If you have handled the preliminaries properly, there is no room left for disagreement.

That really is the Bertrand Russell view, isn't it? How many hundred pages did it take him to demonstrate that you could say 1+1=2, starting with the NAND operator (or something like that). And then Goedel went and blew his program to pieces!

Martin

[Bill_Powers1]

[From Bill Powers (2006.11.21.0715 MST)]

Martin Taylor (2006.11.21) –

I think that’s what I’ve been
trying to get across. That you have to do it that way, and you can’t use
the energy budget of the physical effects in the feedback loop. However,
just to niggle, this partial derivative has to be done
open-loop.

That’s not necessary, because any component of the loop always has
the same input-output function (reorganization aside). However, you do
have to take disturbances into account in the case in question – even if
the loop is open.

The point that you and Bjorn are
both missing is that the physics of behavior is a proposal, a model, not
a description of Real Reality.

I don’t think either of us has missed that point. We have both mentioned
it early in the thread, but we don’t harp on it as the thread
develops.

My point is that your discussion hinges on the correct applications of
several models, and that can’t be settle verbally since the models are
quite specific.

I said:

You’re arguing about
whether the physics model and the PCT model are being correctly applied.
All arguments of this kind about “is” and “is not”
are really arguments about who is following the rules of the model more
correctly. No resolution can be reached through purely verbal
arguments.

and you reply:

I’m afraid I have to dispute
that. Mathematics is just another kind of verbal argument, after
all.

Yes, of course, and numbers are just another kind of written symbol.
However, in verbal arguments terms can change their definitions from one
use to the next, and premises can be constructed on the fly, as needed,
and all sorts of other logical perversions are permitted. A mathematical
argument requires agreement on definitions to start it, and adherence to
those definitions throughout. It requires that the means of reasoning be
known to all, and laid out in advance, and adhered to. This may be just
another kind of verbalizing, but the constraints are such that the result
is far more likely to have meaning than the results of the normal kind of
verbal reasoning have. When the PCT model and the physics model are
involved, and are the point, it would be better to stick to the
appropriate kind of argument (whether numbers and mathematical forms are
used or not)
For example, the many-to-one versus one-to-many issue: You and Bjorn are
talking about different parts of the models in different contexts, so
your disagreements are meaningless (as would be any apparent agreements).
If you speak of the effects of first-order output quantities in the
world-model, it is clear that one output quantity has many effects that
fan out into the physical world, only some of which are involved in the
local control system. If you speak of alternative paths to the same
result in the environment, a many-to-one effect, only some of them will
be in effect at a given time and the others will come into play only if
the first one fails, and then only after some reorganization to hook up
new output effects to the old error signal. If, on the other hand, you’re
speaking of output signals at higher levels, one set of connections to
lower systems has effects on only the reference inputs to lower orders
that are actually connected. And finally, the many-to-one connections in
the upgoing paths are subject to similar comments, with many inputs, some
controlled and some not controlled, contributing. There seems little room
for ambiguity here, so it’s hard to see what prevents agreement except
imprecise language and shifting meanings.
As I’ve explained elsewhere, I recommend using “quantity” to
refer to a physical measurement in one location, and “signal”
to refer to a variable inside the nervous system that measures an effect
at one location but also carries it to another location. So an output
quantity is, mostly, a force created by shortening of an elastic
muscle, while an output signal is the output of an output function
at one level which contributes to the net input to a comparator (or other
function) at a lower level of control system, still within the nervous
system.

Moreover, in a situation
in which the premises and relationships cover a wide range of
possibilities, it is often vry difficult to create an intelligible
mathematical description from which generic conclusions can be derived.
I’d be interested, for example, in how you would presnet in an
intelligble form, a mathematical demonstration of this:

As I pointed out some years ago,
at least 10 and perhaps 15, the energy budget of acting and sensing has
nothing to do with the transmission of information. Information can
easily move in a direction opposite to the direction of energy transfer.
Information has nothing to do with energy (or
entropy).

I assume that “this” refers to “Information can easily
move in a direction opposite to the direction of energy transfer”
rather than “I pointed out…” or “10 and perhaps 15 years
ago” or “information has nothing to do with energy or
entropy.” The first statement contains the kernel of my argument and
can be demonstrated.
A mathematical description of the reasoning in my statement is rather
easy to construct, though we might argue about how well the subject is
understood. I sketched in a case (the ball falling at midnight) in which
one could demonstrate that information is transmitted by means of
absorbing energy into the transmitter rather than sending it from
transmitter to receiver. I did not try to cover a “wide range of
possibilities”: disproofs of a general proposition require only
finding a single sound counterexample. I think you will agree that this
particular counterexample could be worked out if anyone seriously doubted
that it would work as suggested, so I wouldn’t want to go through all
that just to prove such a simple thing. I am not disputing that
information transfer can be accompanied by a forward transfer of
energy. I’m saying only that the direction of energy transfer is not, in
general, important. It may even be irrelevant, Shannon
notwithstanding.

There is simply no point in
trying to settle any matter of this kind without setting up a formal
argument, agreeing on what the data are, and following agreed-upon rules
to derive conclusions. Then the only questions are whether you have
considered all relevant data, measured accurately, and reasoned without
blunders. If you have handled the preliminaries properly, there is no
room left for disagreement.

That really is the Bertrand Russell view, isn’t it? How many hundred
pages did it take him to demonstrate that you could say 1+1=2, starting
with the NAND operator (or something like that). And then Goedel went and
blew his program to pieces!

I said, “a matter of this kind,” meaning a matter involving the
PCT model and the physics model. All you have to do is say what part of
which model at what level you are talking about, and I believe all the
disagreements will disappear. I can’t see what that has to do with
Principia or Godel.

Best,

Bill P.

[Bj_rn_Simonsen]

[From Bjorn Simonsen (20066.11.23, 11:10 EUST)]

Martin Taylor - received 2006.11.20, 17:44 EUST.

I had to read your mail more than twice. I have also
studied your last communication with Bill. I think Bill’s mail influenced my
comments below. It has also been a busy week.

I will explicit express that I appreciated your last
mail (received 2006.11.20, 17:44 EUST )(….most of your mails). You present
statements fill my days with content.

I think, to use an English metaphor (I hope it

translates well) that you are barking up the

wrong tree entirely. Maybe you are even in the

wrong forest.

It looks to me that we often bark trees in different
forests. I joined a forest more or less heedlessly when I wrote:

Still I think the feedback effect at the
peripheral

first level is dependent on physical units in the

language of grams, metres and seconds.

The last part of the sentence is a copy from one of
your earlier mails. What I wanted to write was: “Still I think the feedback
effect at the peripheral first level may lead to different values of the Input
quantity, caused most of changed disturbances”.

Here I confine myself to transmission of information
signals and variables.

Your extensive comments in your last mail show me the
problems that arise when we talk about the energy following a negative feedback
loop. This is really very demanding for me to follow, but I think I understand
what you say.

I confine myself to what Bill name transmission of
information.

Nevertheless I will comment your mail paragraph for
paragraph. If you want read it through, I agree in your energy thinking.

Consider the physical train of events leading from
the person’s

muscular influences on the environment to the
physical effects of the

environment on the sensors. And let’s confine
ourselves to control of

the perceived colour of something that is very
readily influenced.

Turning a knob that alters the frequency of a
spectral line is the

simplest means I can think of. You may think of
some other, but I

doubt it will affect the argument.

OK

The person wants to control the perceived colour
of the line to match

some reference value. It doesn’t matter how this
reference value is

derived.

I understand this as if the person wishes to perceive
e.g. blue light. He has experienced blue light earlier and he thinks he can
identify blue light. I also understand that the instrument is a knob tuning the
frequency of light and a screen showing the tuned light.

The current state of the perception differs from
the reference value,

so the control unit emits some output that
(eventually) results in

muscle movements that change the knob angle.
Changes in the knob

angle cause (note “cause” rather than
“influence”, because I’m not

even considering the possible existence of a
disturbance at this

point).

Your first sentence makes me stop. I think we think
the same way, but I will go in detail. And if you think I am wrong, please let
me know.

Your first sentence is about incidents in a negative
feedback loop, the PCT loop. The loop can be presented as a composed loop, as a
loop expressing what happens when we wish to perceive something or what happens
when we are imputed a disturbance e.g. in form of a light. The loop can also be
presented to show how a single disturbance variable is sensed and the way the
perceptual signal follow in a negative feedback loop (Bruce Nevin- watching
neurons connect, live).

My argument for all those lines are that when I talk
about a reference value as you do in your first sentence it is implied that I
talk about the certain loop which is in force for the perception I control.

If I wish to see blue light, control for blue light,
the loop is the one that is in force for blue light. There are other loops that
control for red light. Therefore the reference value doesn’t say anything about
which colour I control, the reference value tells me how much I wish to
perceive the yellow light.

Do you see it different?

The rest of your passage is OK.

Now think of the physical effects and
transformations in this part of

the loop. Neural impulses might be measured in
microwatts, an energy

per second measure, so that the energy expended in
neural impulses

during the control movement can be measured in
microwatt-seconds.

These impulses, by various chemical means, cause
microfibres in the

muscles to contract. There’s a gain here, because
they use energy

already chemically stored in the muscles. But is
there a function

relating neural microwatt-seconds to muscular
gm-cm. I doubt it,

becuase the effect will depend on how tired the
muscle is, and on how

tense the muscle was before the movement began.
But for the sake of

argument, let’s say that at any instant during the
move there is a

gain factor. It would be a pure number, since we
are comapring energy

rates on both ends of the transform.

As you describe, the perceptual signal is compared
with the reference signal and the error I think upon as an information signal.
It has its frequency and of course it represents energy. When this signal
reaches muscle fibres, I think upon the signal just as a switch that actuates
the contract of the muscle fibre. As you, I think upon the potential energy in
the muscle fibres as the only energy that contract the muscle fibre.

This energy has nothing to do with the energy
represented by the error signal (nothing is a strong word). The muscle fibre
has the energy it has.

The error signal may be very little and lead to the
consumption of just a part of the potential energy in the muscle fibre or it
may be very large and lead to consumption of all the energy (almost all) in the
muscle fibre. If the muscle was not so tired, it would have contracted more.

In a PCT information transmission loop I understand it
as if the error is multiplied with a gain. This gain I think is great because
the energy consumption that initiates the muscle contraction is usually much
smaller than the energy consumption that constitutes the muscle contraction. I
think it is a misrepresentation to multiply the error with a gain, a (often)
large number with the units output variables per error signal. But this is a
way to bring the information signal further.

If I am not under a delusion it is OK for me to start

argument, let’s say that at any instant during the move
there is a gain factor.”

But as you see here I have a unit (output variables
per error signal).

Now we have the muscles expending energy, some of
which is dissipated

as heat, some of which goes to turn the knob.
Let’s say that the knob

rotates a diffraction grating, so it’s very easy.
Nevertheless, there

is always loss. The diffraction grating has
inertia, its bearings

have friction, and in the end, we have a
stationary diffraction

grating, just as we did before. ALL the muscle
energy has been

dissipated into the environment. There is NO gain
factor relating

muscular output to changes in the position of the
diffraction grating.

This is OK in this example. But in other examples the
same muscle contraction could have great or very great effects on the
environment and the disturbance because of this change much or very much.

If the muscle contraction overturns a cup of coffee
the disturbance change. If the muscle contraction overturns a 5 meter high
bookshelf the disturbance changes very much.

What does the diffraction grating do? It splits a
broadband source

into a spectrum. Any particular wavelength is
directed in a different

direction from any other wavelength. It doesn’t
influence the energy

into or out of it. The different spectral
wavelengths may have

different powers, but that’s a property of the
source, unaffected by

the rotational angle of the grating. The person’s
muscular output may

affect the power being delivered in a particular
direction, but there

isn’t any functional relation between hw much
energy is output by the

muscles and how much power is delivered to the
beam that reaches the

person’s eye.

OK (your instrument is otherwise than mine, I think).

Now consider that last transformation before we
dive beneath the

person’s skin. Photons from the beam directed
toward the eye hit

receptor cones in the retina. Some repsond more
strongly to red, some

to green, and some to blue. The output of each is
in a power measure

(by way of counting neural impulses) of
microwatts, as is the power

of the beam. There’s a potential gain factor here.
The power in the

neural impulses is gained from energy stored in
the cones, released

by the effect of incoming photons. The amount of
energy in an impulse

isn’t related to the power in the beam, though the
number of impulses

per second is. That relation changes very quickly,
though, as the

energy supply in a cone is depleted, and as the
effects of neural

interconnections take hold (e.g. lateral
inhbition). It’s not a gain

factor to be relied on when calculating control
loop gain.

This is OK. I think your sentence: “The amount of
energy in an impulse

isn’t related to the power………” is wonderful. And I
agree with your last sentence. But there is an input gain and it has the unit
“perceptual signal per Input quantity unit”.

So, what do we have when looking for a gain factor
between output

neural impulaes feeding the muscles and neural
impulses up the optic

nerve? Not a lot.

To me this total gain in a way is dependent on factors
in the environment. I think we will experience that if we execute a walk on the
moon. (Excuse me for jumping away from colours).

The number of neural impulses up the optic nerve is

likely to change very little as the spectrum
changes, though the

relationship between the numbers in the different
colour channels

will indeed change.

This sentence is the reason for my detailed
description about loops above. I have thought that a loop controlling
perception of blue light doesn’t control perceptions of red light (here I think
upon a concrete loop where e.g. a B cone is influenced). You indicated that above:
“Some respond more strongly to red, some to green, and some to blue.”

Let’s just use Red/Green as an example, and say

that the perceived colour is a function of (rate
of “Red” impules) /

(rate of “Green” impulses). That’s a raw
(dimensionless) number. So

the gain factor must have a dimension of “per
microwatt”. But we

already saw that there could be at least one place
in the feedback

path where the output micorwatts have no relation
at all to the

sensory input. They don’t have a functional
relationship to the

rotation of the grating.

Maybe I misunderstand your Red/Green example. I perceive
it as if e.g. I look at two lamps, a red and a green lamp. Now I am out of the
loops and just analyse them.

I absolutely agree that the Gain factor producing
perceptual signals has no relation to the output microwatt when we talk about
energy.

As you understand I think the Input Gain factor in the
“information” loop has the unit perceptual signals per Input quantity.

So, in at least one case of a maximally peripheral
and extremely

simple (unbraided) control loop, we can’t trace a
gain function, and

we don’t seem to have even a relation between
output energy and input

energy before computing the ratio of
“Red” to “Green” impulses.

Yes, when we talk about energy.

···

from: “But for the sake of

---

All of that long analysis was in response to your
" Still I think the

feedback effect at the peripheral first level is
dependent on

physical units in the language of grams, metres
and seconds."

I am sorry I expressed myself as I did (look above).
But I appreciated and I will thank you for your long comment. I don’t think I
have followed the energy in the negative control feedback loop earlier.

I am saying that it isn’t, even if you can measure
the power of the

light that enters the eye, or of the sound
vibrations that move the

eardrum, and even if the perception being
controlled is at the lowest

possible level in the hierarchy.

OK

The units in which the gain can be measured are
commensurate only

inside the control unit: for example, the analyst
could compute how

much a unit change in the error value would alter
the perceptual

signal if the connection were broken between the
perceptual signal

and the comparator. And so forth.

OK

bjorn

[Martin_Taylor2]

[Martin Taylor 2006.11.23.20.10]

[From Bjorn Simonsen (20066.11.23, 11:10 EUST)]
Martin Taylor - received 2006.11.20, 17:44 EUST.

I confine myself to what Bill name transmission of information.

That area is technically much more difficult than is the discussion of energy transformation. I think we ought to leave it aside unless you want to go into some pretty detailed analyses. As I remember, the thread started with the question of whether the units of physical measurement (grams, cm, coulombs, seconds) of things happening in the environment were relevant when computing the gain around a feedback loop.

... let's confine ourselves to control of
>the perceived colour of something that is very readily influenced.
>Turning a knob that alters the frequency of a spectral line is the
>simplest means I can think of. ...
>The person wants to control the perceived colour of the line to match
>some reference value. It doesn't matter how this reference value is
>derived.

I understand this as if the person wishes to perceive e.g. blue light. He has experienced blue light earlier and he thinks he can identify blue light. I also understand that the instrument is a knob tuning the frequency of light and a screen showing the tuned light.

Yes.

>The current state of the perception differs from the reference value,
>so the control unit emits some output that (eventually) results in
>muscle movements that change the knob angle. Changes in the knob
>angle cause (note "cause" rather than "influence", because I'm not
>even considering the possible existence of a disturbance at this
>point).

Your first sentence makes me stop. I think we think the same way, but I will go in detail. And if you think I am wrong, please let me know.

I don't think we think the same way here. See below.

Your first sentence is about incidents in a negative feedback loop, the PCT loop.

It is about a starting state in a loop, not about incidents. It's the state when the analyst begins to consider what should then be expected to happen in the loop.

The loop can be presented as a composed loop, as a loop expressing what happens when we wish to perceive something or what happens when we are imputed a disturbance e.g. in form of a light.

I made a point that the loop I was considering was NOT subject to ANY disturbance. The error in the starting condition came from the fact that the screen was showing a colout (say red, for example) different from the reference colour (blue, you suggest).

The loop can also be presented to show how a single disturbance variable is sensed and the way the perceptual signal follow in a negative feedback loop (Bruce Nevin- watching neurons connect, live).

No. No disturbance. Only the light, which is part of the control unit's feedback loop. A disturbance, in this scenario, would be something, other than the person's muscular output, that altered the angle of the diffraction grating and thereby changed the colour of the light.

My argument for all those lines are that when I talk about a reference value as you do in your first sentence it is implied that I talk about the certain loop which is in force for the perception I control.

Which, in this case is the perception of the colour of the light, which can vary anywhere along the spectrum from deep red to indigo by way of green (but it can't show mixed colours such as purple or white).

If I wish to see blue light, control for blue light, the loop is the one that is in force for blue light.

No. It is the one that controls for colour, with a reference value of "blue".

There are other loops that control for red light.

This loop could equally have a reference value of "red". It's not a different loop.

Therefore the reference value doesn't say anything about which colour I control, the reference value tells me how much I wish to perceive the yellow light.
Do you see it different?

Yes, I do. The reference value says only which colour I want to see. That's all it says. It doesn't say anything about how much of the colour I want to see. That might be the business of a different control loop, but it's not the business of the one we are considering.

As you describe, the perceptual signal is compared with the reference signal and the error I think upon as an information signal.

Put very loosely, that's OK. But in normal PCT, it's a signal with a variable magnitude, and you can use that value in computations about how the loop behaves.

I think it is a misrepresentation to multiply the error with a gain, a (often) large number with the units output variables per error signal. But this is a way to bring the information signal further.

Whether it is a mistake depends on what you want to do with it. You probably could make statements about energy input and output. After all, any amplifier gets its output energy from a source separate from the control signal, so that's not really the point at issue.

Bill P. pointed out that once you get outside the control unit, you really can't talk again about gain factors until you get back into the control unit. The interfaces are (in the classic diagram) the output function and the perceptual input function. The input to the output function is the error signal. The output from the perceptual input function is the perceptual signal (the perception).

If (as an analyst) you disconnect the perceptual signal from the comparator and then change the value of the error signal by some known amount over some known time, you then will find a change in the perceptual signal, which also will be a function of time. The relation between the two changes is your feedback gain function. In the classical diagram, it is the combination of the output function, the environmental feedback path, and the perceptual input function. That gain is a dimensionless function of time (I mean a time-varying simple number).

The problems arise when we try to partition the path into the three components, and associate different gains and dimensions to each. In almost all cases, the perceptual signal relates to some kind of a pattern in the environment outside the control unit. As soon as we talk about "pattern" we are back in the realm of information theory, where things get rather tricky. I believe it might be possible to do an information-theoretic analysis of the control loop in which the partitioning might make sense, but when I tried to do this a dozen years ago, I couldn't make it work. I don't know if I could do any better now.

>... There is NO gain factor relating
>muscular output to changes in the position of the diffraction grating.

This is OK in this example. But in other examples the same muscle contraction could have great or very great effects on the environment and the disturbance because of this change much or very much.
If the muscle contraction overturns a cup of coffee the disturbance change. If the muscle contraction overturns a 5 meter high bookshelf the disturbance changes very much.

Disturbance to what? My perception of the value of the Persian carpet on which each thing fell? My desire to have a good taste in my mouth early in the morning? In either of those cases, I think the coffee spill would be a MUCH greater disturbance than the fall of the bookcase. When you talk about a disturbance you must ALWAYS be talking about a specific CONTROLLED perception. If I was a very messy person, it might be that neither the coffee spill nor the bookcase fall created any disturbance at all.

>So, what do we have when looking for a gain factor between output
>neural impulaes feeding the muscles and neural impulses up the optic
>nerve? Not a lot.

To me this total gain in a way is dependent on factors in the environment. I think we will experience that if we execute a walk on the moon. (Excuse me for jumping away from colours).

It is entirely dependent on factors in the environment, isn't it? The whole path between muscular output and phtons impinging on the retina to create neural impulses is in the environment (always, pace Bill P., recognizing that there may be no "real reality" out there).

>The number of neural impulses up the optic nerve is
>likely to change very little as the spectrum changes, though the
>relationship between the numbers in the different colour channels
>will indeed change.

This sentence is the reason for my detailed description about loops above. I have thought that a loop controlling perception of blue light doesn't control perceptions of red light (here I think upon a concrete loop where e.g. a B cone is influenced). You indicated that above: "Some respond more strongly to red, some to green, and some to blue."

A loop controlling the intensity of a blue light may well be independent of a loop controlling the intensity of a red light. But the loop we were discussing controls the colour of a ligh, which might be red, blue, or anything in between. To a very loose approximation, colour depends on the ratios of the outputs of the different colour channels, not to the intensity of any one of them.

>Let's just use Red/Green as an example, and say
>that the perceived colour is a function of (rate of "Red" impules) /
>(rate of "Green" impulses). That's a raw (dimensionless) number. So
>the gain factor must have a dimension of "per microwatt". But we
>already saw that there could be at least one place in the feedback
>path where the output micorwatts have no relation at all to the
>sensory input. They don't have a functional relationship to the
>rotation of the grating.

Maybe I misunderstand your Red/Green example. I perceive it as if e.g. I look at two lamps, a red and a green lamp.

No, just a lamp that changes colour as you turn a knob.

>The units in which the gain can be measured are commensurate only
>inside the control unit: for example, the analyst could compute how
>much a unit change in the error value would alter the perceptual
>signal if the connection were broken between the perceptual signal
>and the comparator. And so forth.

OK

And that, in essence, is what Bill P. said, too [From Bill Powers (2006.11.20.0810 MST)].

There is a very simple way to measure the gain in the environmental part of a control loop. Calculate the partial derivative of the perceptual signal with respect to the first-order output quantity. If both the perceptual signal and the output quantity can vary, there will be such a partial derivative. Its value is the gain of the output function.

Martin

[Bj_rn_Simonsen]

[From Bjorn Simonsen (2006…11.24,15:30 EUST)]

Martin Taylor 2006.11.23.20.10

(the transmission of information)

(That) area is technically much more difficult
than is the discussion

of energy transformation. I think we ought to
leave it aside unless

you want to go into some pretty detailed analyses.

I confine myself to the value of the frequency of
signals inside the brain.

I guess you among other factors think upon not linear
functions and time-dependent functions describing the gain in different functions.

As I remember, the thread started with the
question of whether the

units of physical measurement (grams, cm, coulombs, seconds) of

things happening in the environment were relevant when computing

the gain around a feedback loop.

Yes, but when the happenings in the environment are
sensed, we have a perceptual signal. And when I talk about PCT/HPCT I follow
the perceptual signal as an information transmission through the loop.

I know I do it in a very simple way (above). But that
is how I think upon PCT.

I am a curious person, therefore I continue studying
your comments. It looks like I am in another forest still. You talk about
controlling of light and I comment your statements as if you e.g. are
controlling blue light.

I have a tendency to deviate the theme that started
our discourse. Actually I should have stopped here and waited for your answer.
But you are so clear below saying:

Yes, I do. The reference value says only which
colour I want to see.

That’s all it says. It doesn’t say anything about
how much of the

colour I want to see. That might be the business
of a different

control loop, but it’s not the business of the one
we are considering.

I will continue commenting your comments below, but I
have to ask how the reference value can express which colour I want to see.

You expressed yourself: “Photons from the beam
directed toward the eye hit

receptor cones in the retina. Some respond more
strongly to red, some

to green, and some to blue.”

I will confine myself to take as my starting point
that some cones respond to red, some to green and some to blue. I skip “more
strongly” because those words make everything more difficult.

I think some signals from B cones are directed direct to
a second level comparator. Some signals are directed to the first level
comparator.

I think some signals from R cones are directed to
another second level comparator.

And so on.

Let us say that the knob/spectral line shows red
light.

When I wish to see blue light, the reference has a
value different from zero. It becomes compared with the perceptual signal.
There is no perceptual signal in the blue loop, because the knob shows red
light. The reference value minus the zero perceptual signal becomes an error
and an action turning the knob to another colour.

The signals from the R cones are directed to a
comparator with reference value zero because I don’t control for red light. We
don’t get any error because it becomes negative and negative signals don’t exist.

This is how I explain how I control for a certain
colour.

You say you control for light. The only way I think
you can control for light is to control the light at the first level
(intensity). If you just control for light, you wish to see light, and there is
a reference for your wish.

Signals from the B cones, the R cones and the G cones are
directed to different comparators at the first level. Because you don’t control
any perceptions at a higher level, the references there are zero.

At first level there are errors if you wish another
form for light. If you perceive the light OK the error is zero.

Which outputs make you turn the knob to another
colour?

It is about a starting state in a loop, not about
incidents. It’s the

state when the analyst begins to consider what
should then be

expected to happen in the loop.

Maybe the language fools me, but the starting state in
a loop is an incident for me.

If I wish to see blue light, control for blue
light, the loop is the

one that is in force for blue light.

No. It is the one that controls for colour, with a
reference

value of “blue”.

I have always thought that the reference always is an expression
for how much I wish to perceive of a certain quality. I also understand your
statement, if the same loop doesn’t control for different colours.

There are other loops that control for red
light.

This loop could equally have a reference value of
“red”. It’s not a

different loop.

Is it possible for you to paint such a loop or express
it with words. I have a problem here.

Therefore the reference value doesn’t say anything about which

colour I control, the reference value tells me
how much I wish to

perceive the yellow light.

Do you see it different?

Yes, I do. The reference value says only which
colour I want to see.

That’s all it says. It doesn’t say anything about
how much of the

colour I want to see. That might be the business
of a different

control loop, but it’s not the business of the one
we are considering.

Well I go for a new round. I still think using a diffraction
grating it is possible to simulate how to control for blue light when the
reference is a number expressing how strong I wish to perceive blue light.

As you describe, the perceptual signal is
compared with the

reference signal and the error I think upon as
an information signal.

Put very loosely, that’s OK. But in normal PCT,
it’s a signal with a

variable magnitude, and you can use that value in
computations about

how the loop behaves.

I still stand up. It is OK for me that the reference
signal may have a variable magnitude. It depends on the output signals at
higher levels how strong I wish to perceive blue light. Wishing to perceive
blue light is not an either – or condition.

I
think it is a misrepresentation to multiply the error with a

gain, a (often) large number with the units
output variables per

error signal. But this is a way to bring the
information signal

further.

Whether it is a mistake depends on what you want
to do with it. You

probably could make statements about energy input
and output. After

all, any amplifier gets its output energy from a
source separate from

the control signal, so that’s not really the point
at issue.

I can live with your statements, but I think you
misunderstood me.

To me gain is an amplifier expression. Here I think I have a quality and make the
same quality normally greater.

This is not what happens in an Output function. There
the quality called frequency lead to an action which is quite another quality.
Therefore I said:

I
think it is a misrepresentation to multiply the error with a

gain, a (often) large number with the units
output variables per

error signal.

Bill P. pointed out that once you get outside the
control unit, you

really can’t talk again about gain factors until
you get back into

the control unit. The interfaces are (in the
classic diagram) the

output function and the perceptual input function.
The input to the

output function is the error signal. The output
from the perceptual

input function is the perceptual signal (the
perception).

Yes, but if you go to his Live Block he has a Feedback
Gain. A logarithmic form for Gain.

If (as an analyst) you disconnect the perceptual
signal from the

comparator and then change the value of the error
signal by some

known amount over some known time, you then will
find a change in the

perceptual signal, which also will be a function
of time. The

relation between the two changes is your feedback
gain function. In

the classical diagram, it is the combination of
the output function,

the environmental feedback path, and the
perceptual input function.

That gain is a dimensionless function of time (I
mean a time-varying

simple number).

Yes this is OK because you say: “, it is the
combination of the output function,

the environmental feedback path, and the perceptual
input function”. But in Bill’s Live Block he has as I said a certain Feedback
gain. Here the output variable is multiplied with a Feedback gain to get an
Input quantity (a part of input quantity). And here the feedback gain must have
a unit named input quantities per output quantity. Are we both correct?

The problems arise when we try to partition the
path into the three

components, and associate different gains and
dimensions to each. In

almost all cases, the perceptual signal relates to
some kind of a

pattern in the environment outside the control unit.
As soon as we

talk about “pattern” we are back in the
realm of information theory,

where things get rather tricky. I believe it might
be possible to do

an information-theoretic analysis of the control
loop in which the

partitioning might make sense, but when I tried to
do this a dozen

years ago, I couldn’t make it work. I don’t know
if I could do any

better now.

I see no problems with your first sentence if we talk
about time independent gains.

Your second sentence I meet with respect. I know too
little about information theory.

As a comment to your last sentence, I will wait and
see.

… There is NO gain factor relating

muscular output to changes in the position of the diffraction grating.

This is OK in this example. But in other
examples the same muscle

contraction could have great or very great
effects on the

environment and the disturbance because of
this change much or very

much.

If the muscle contraction overturns a cup of
coffee the disturbance

change. If the muscle contraction overturns a
5 meter high bookshelf

the disturbance changes very much.

Disturbance to what? My perception of the value

of the Persian carpet on which each thing fell?

Are you fooling me with words here? How can a Persian
carpet be a value?

My desire to have a good taste in my mouth

early in the morning? In either of those cases, I
think the coffee

spill would be a MUCH greater disturbance than the
fall of the

bookcase. When you talk about a disturbance you
must ALWAYS be

talking about a specific CONTROLLED perception. If
I was a very messy

person, it might be that neither the coffee spill
nor the bookcase

fall created any disturbance at all.

Yes of course. In posterity I can say I hoped you read
my statements in your best intention.

So, what do we have when looking for a gain factor between output

neural impulaes feeding the muscles and neural impulses up the optic

nerve? Not a lot.

To me this total gain in a way is dependent on
factors in the

environment. I think we will experience that
if we execute a walk on

the moon. (Excuse me for jumping away from
colours).

It is entirely dependent on factors in the
environment, isn’t it? The

whole path between muscular output and phtons
impinging on the retina

to create neural impulses is in the environment
(always, pace Bill

P., recognizing that there may be no “real
reality” out there).

Well I think there is something outside the brain.
When I agree with one or more people we say the real reality out there is the
way we agree. It happens I disagree with some people about the real reality out
there. Most often I let them live and we disagree.

A loop controlling the intensity of a blue light
may well be

independent of a loop controlling the intensity of
a red light. But

the loop we were discussing controls the colour of
a light, which

might be red, blue, or anything in between. To a
very loose

approximation, colour depends on the ratios of the
outputs of the

different colour channels, not to the intensity of
any one of them.

Well this is what I have commented above.

bjorn

[Bill_Powers1]

[From Bill Powers (2006.11.24.1450 MST)]

Fred Nickols (2006.11.24.1431 EST) --

Bill's comment prompts a question. Setting aside the particulars of the discussion in question, I would like to deal with the situation in which one person says light is a disturbance and another says it isn't.

It seems to me that that statement is somewhat akin to one person saying that food is a reinforcer and another saying it isn't. Don't both statements depend on context?

If I'm trying to look at something and a bright light is causing me to squint or shade my eyes, it seems to me that this bright light is a disturbance to my goal of vewing something. On the other hand, if the light doesn't interfere with my ability to do that, then it's not a disturbance.

Yes, that is true. However. in the example Martin Taylor gave, it was the color of a light that was a controlled variable, affected by operating a knob that turns a diffraction grating. Since this variable is affected by the action of a control system, it can't qualify as a disturbance in the way Bjorn Simonsen is describing it. As I use the term, a disturbance is a variable that affects a controlled variable independently of the action of the control system.

I have pleaded for precision on this subject a number of times over the duration of CSGnet. Unfortunately, the word "disturbance" has two quite different meanings. It can mean the change in something that is caused by some other unnamed variable, or it can mean the variable that is causing the change. "There was a disturbance of the car's speed" (caused by a headwind, a tailwind, friction in a wheel bearing, or water in the gasoline). Or, "The tailwind was a disturbance that changed the speed of the car." I always use the term in the second sense: the disturbance is the variable that is responsible for the change in the input quantity. However, the amount of change that results when a disturbance is present depends on the other variable that also affects the same input quantity, the feedback from the controller's output action. If control is tight and the magnitude of the disturbance does not change too fast, there will be very little actual change in the input quantity even if the disturbance is large. So when a control system is reasonably effective, even quite a large disturbance causes hardly any disturbance. If that sentence sounds confusing, it is. So say what you mean if you want anyone to understand you.

In the Live Block Diagram, the "disturbing variable" (which is one way to be precise about what is meant) is shown as physically different from the input quantity, which is the controlled quantity. Just think of that diagram and there will be no confusion. Or less, anyway.

I won't get into the reinforcer thing. I've discussed that quite a lot, too. In my opinion, the theory of reinforcement is simply wrong and it should be abandoned entirely. An increase in reinforcement does not cause an increase in behavior. It is exactly the other way around: an increase in (the right kind of) behavior causes an increase in reinforcement, where "reinforcement" means nothing but whatever it is that is affected by the behavior. The change in the behavior remains completely unexplained by that theory.

Best,

Bill

[Martin_Taylor2]

Re: Elements in the PCT model
[Martin Taylor 2006.11.24.14.43]

Bill P. says our problem in communication hinges on a
disagreement about what constitutes a disturbance. So let’s get that
settled first. I don’t think that’s the main problem, but if it is a
problem then it should be fixed.

According to me:

A control system has two places where things happening in the
outside world can influence the values of its signals. There really
are only three such signals in the classic control loop (as in the
image here:

The three “inside” signals are P, R, and E. S comes
from outside, and O goes to the outside. S is the imput to the
perceptual input function, and ought to be shown as a bundle of lines
coming from different places. Likewise O should be shown as a bunch of
lines going to different places.

S is the “input quantity”, though for most control
systems it really should be the “input quantities”
(plural).

The value of the perceptual signal at any moment depends only on
S (and possibly the recent time course of S).

S has contributions from two places: (1) the effects of O on the
outside environment – in other words, anything that is perceptibly
influenced by the control system’s output – and (2) things that
happen in the outside environment independently of what the control
system does – called the disturbance when the perception is
controlled.

P, the perception, depends on both the output and the
disturbance.

Now we ask why a “disturbance” is called that. Without
the disturbance, the output would bring the perception very near its
reference value, and would remain unchanged until the reference value
changed. Unfortunately, variation among the influences from the outer
world “disturb” that stability. What the disturbance most
disturbs is the output signal value.

If there’s no control there is no output value to disturb. If
there were no control, the perceptual signal could vary all over the
place, without disturbing anything. Then all you would say is that
there is a signal called the perceptual signal that corresponds in
some way to a state of the outer world (the bundle of S signals). It’s
a perception, but not a disturbed one. The “disturbance”
contributes to the perceptual signal value, but most disturbs the
output value.

Always, the perceptual signal “reports” some state of
the outer world, whether it is controlled or not. Only if the
perception is controlled can we talk about the output signal being
disturbed. It makes sense to talk about a “disturbance”
signal that contributes to a controlled perception, but not about a
disturbance to a perceptual signal that is freely variable
(uncontrolled).

Now…

[From Bjorn
Simonsen (2006…11.24,15:30 EUST)]
Martin Taylor
2006.11.23.20.10

… It looks like I
am in another forest still. You talk about controlling of light and I
comment your statements as if you e.g. are controlling blue
light.

Why? I’m not talking about controlling blue light. I’m talking
about controlling the colour of a light. You started out dealing with
controlling for the colour of the light, and I’m staying with that.
Now you want to change that to controlling for the intensity of blue.
Why the shift?

I have a tendency
to deviate the theme that started our discourse. Actually I should
have stopped here and waited for your answer. But you are so clear
below saying:

Yes, I do. The
reference value says only which colour I want to
see.
That’s all it
says. It doesn’t say anything about how much of
the
colour I want
to see. That might be the business of a different
control loop,
but it’s not the business of the one we are
considering.

I will continue
commenting your comments below, but I have to ask how the reference
value can express which colour I want to see.

As with any controlled perception, the reference value is a
number between some lower and upper limit. In this case, we might say
the lowest number (let’s call it 0.0) corresponds to indigo, the
highest (say 10.0) corresponds to deep red. Maybe green is 5.0. Then a
reference value of about 1.5 might correspond to blue.

You expressed
yourself: “Photons from the beam directed toward the eye
hit
receptor cones in
the retina. Some respond more strongly to red,
some
to green, and some
to blue.”

I will confine
myself to take as my starting point that some cones respond to red,
some to green and some to blue. I skip “more strongly” because
those words make everything more difficult.
I think some
signals from B cones are directed direct to a second level comparator.
Some signals are directed to the first level
comparator.
I think some
signals from R cones are directed to another second level
comparator.

I don’t follow that. You are bringing in two new control systems,
a first level control system for blue intensity and a second level
comparator for blue intensity. Seems redundant to me (apart from the
fact that you need neither in order to control the colour of the
light.

And so
on.
Let us say that the
knob/spectral line shows red light.
When I wish to see
blue light, the reference has a value different from
zero.

Using my suggested numbers, the reference would be 1.5 (= blue)
and the perception would be around 9 (= red).

It becomes
compared with the perceptual signal. There is no perceptual signal in
the blue loop, because the knob shows red light.

So far, we haven’t introduced a mechanism for control. Your “blue
loop” is outside the control system for the perception of colour.
In the system I described (a moveable diffraction grating), no
“blue loop” is required for effective colour control.

The reference
value minus the zero perceptual signal becomes an error and an action
turning the knob to another colour.

Yes, but this happens in the colour control unit, not in any
independent “blue loop”.

The signals from
the R cones are directed to a comparator with reference value zero
because I don’t control for red light.

Why do you say you control for blue light, but not for red light?
If the red intensity changes, the perceived colour will change, won’t
it? If you are going to manipulate the colour by independently varying
the intensities of two lights that mix on the screen, rather than by
rotating a diffraction grating, wouldn’t you want to control the red
intensity as well as that of the blue?

This is how I
explain how I control for a certain colour.

You actually haven’t described a way that you might control for
any given colour. All you have described is how you would control for
the intensity of a single invariant colour (blue in this case).
Suppose the reference value for colour changed slightly, from blue to
turquoise, how would your mechanism deal with that? How would your
mechanism track a reference value that changed smoothly from blue
through gree to yellow and back again?

You say you control
for light.

No. I say that we are talking about controlling for the COLOUR of
the light.

Signals from the B
cones, the R cones and the G cones are directed to different
comparators at the first level. Because you don’t control any
perceptions at a higher level, the references there are
zero.

But we ARE controlling a perception at a higher level! We are
controlling for some function of the relative intensities of these
three independent channels. Furthermore, if the control mechanism is
limited to a rotation of a single knob, you cannot control the
intensities of the individual R, G, and B channels. You can’t control
B intensity, you can’t control R intensity, and you can’t control G
intensity. You might control one of them, but you couldn’t
independently control the others. You don’t have enough degrees of
freedom. That’s true whether the knob affects the rotation of a
diffraction grating, the placement of filters, or the intensities of
lights. You have only one degree of freedom for your output, and can
use it to control only one perceptual signal – in the case under
discussion, that is the colour of the light.

It is about a
starting state in a loop, not about incidents. It’s
the
state when the
analyst begins to consider what should then be
expected to
happen in the loop.

Maybe the language
fools me, but the starting state in a loop is an incident for
me.

Maybe it is the language. For me an incident is something that
changes. The starting state (or any other state) is just a state. To
have an incident you must have a change of state, not just a starting
state.

I have always
thought that the reference always is an expression for how much I wish
to perceive of a certain quality. I also understand your statement, if
the same loop doesn’t control for different
colours.

There are
other loops that control for red light.

This loop could
equally have a reference value of “red”. It’s not
a
different
loop.

Is it possible for
you to paint such a loop or express it with words. I have a problem
here.

I assume you mean “how could you describe a perceptual input
function that would output different values for different wavelengths
of light.”

The actual function in the eye is much more complex, but here’s a
possible one, where P is the perceptual value, and R, G, and B are the
logarithms of the intensities in the red, green, and blue channels
respecitvely.

Let’s try to build a crude function. P = f(R-B) would provide a
scale from one end of the spectrum to the other, and that might be
sufficient (f is just a monotonic function that provides soft limits
at 0 and 10, rather than allowing the value of P to go to plus or
minus infinity; zero and ten are chosen to match the numbers I used
above). This Perceptual input function wouldn’t discriminate between
green and purple, but that’s all right because we are asserting that
we are using only spectral wavelengths. If you want to make a second
discrimination, such as between green and purple, you’d have to have a
second function, such as P1 = G - (R+B)/2.

This function is nothing like what the physiological visual
system does, but I hope it illustrates the point. P corresponds to the
colour variation of the light, not its intensity, and not to the
intensity of either R or B.

Well I go for a new
round. I still think using a diffraction grating it is possible to
simulate how to control for blue light when the reference is a number
expressing how strong I wish to perceive blue
light.

Sure you could, but that would be independent of the control
system that controls the colour of the light. Gy the way, you might
find it a little difficult to control for the intensity of blue when
the perception is changing from green to yellow to red as the grating
rotates.

----separate topic------

To me gain is an
amplifier expression. Here I think I have a quality and make the same
quality normally greater.
This is not what
happens in an Output function. There the quality called frequency lead
to an action which is quite another quality. Therefore I
said:

I think it
is a misrepresentation to multiply the error with
a
gain, a
(often) large number with the units output variables
per
error
signal.

Perhaps so. There’s certainly a dimensional change between
“number per unit time” (number of neural impulses) and
“force applied to a lever”. There are a lot more if you work
through the effects of the actions involved in “seeing my wife’s
face”. But as I described earlier, those changes really can’t be
tracked through the environment, since we are normally dealing in
patterns and (eventually) with the effects pattern changes have on the
controlled perception. It’s pretty hard to think of what kind of
“gain” might b involved in changing a pattern … … into
… … …!

In simulations, it makes good sense to finesse those problems by
substituting a single feedback path to represent them all, and to keep
the units of measurement on that path the same as in the rest of the
loop. The only aspect of the feedback path that matters to the
simulation is the time course of the influence a momentary impulse
output has on the perceptual function, …

_ output

pulse

    x x  

influence on the

x x “input quantity”

x x

x x

···

---

time
—>

… and for that purpose, all the dimensional changes from the
output function to the environment have been reconverted back to
neural impulses per second.

But in
Bill’s Live Block he has as I said a certain Feedback gain. Here the
output variable is multiplied with a Feedback gain to get an Input
quantity (a part of input quantity). And here the feedback gain must
have a unit named input quantities per output quantity. Are we both
correct?

I suppose so, but in any realistic situation there is no one
single output quantity and no one single input quantity. The
simulations quite reasonably combine all the multiple effects into one
virtual effect called “output quantity” and one virtual
effect called “input qantity”. Only in exceptional
circumstances will you be able to measure either of those in the
environment accessible to an Observer. The Analyst would look for the
feedback gain inside the control system, from the error signal to the
perceptual signal.

The problems
arise when we try to partition the path into the
three
components, and
associate different gains and dimensions to each.
In
almost all
cases, the perceptual signal relates to some kind of
a
pattern in the
environment outside the control unit. As soon as
we
talk about
“pattern” we are back in the realm of information
theory,
where things
get rather tricky. I believe it might be possible to
do
an
information-theoretic analysis of the control loop in which
the
partitioning
might make sense, but when I tried to do this a
dozen
years ago, I
couldn’t make it work. I don’t know if I could do
any
better
now.

I see no problems
with your first sentence if we talk about time independent
gains.

Why time-independent? That’s a rare special case.

If the
muscle contraction overturns a cup of coffee the
disturbance
change. If
the muscle contraction overturns a 5 meter high
bookshelf
the
disturbance changes very much.

Disturbance to
what? My perception of the value

of the Persian carpet on which each thing
fell?

Are you fooling me
with words here? How can a Persian carpet be a
value?

Oh, perhaps $2000 before the coffee spill, $1200 after the spill.
Or so the owner might perceive it. If the user is controlling for a
perception of the carpet’s value, that would be quite a disturbance,
wouldn’t it? The output in payng for cleaning bills would be quite
different from what it would have been if the coffee had not been
spilled.

To me this
total gain in a way is dependent on factors in the
environment. I think we will experience that
if we execute a walk on
the moon.
(Excuse me for jumping away from colours).

It is entirely
dependent on factors in the environment, isn’t it?
The
whole path
between muscular output and phtons impinging on the
retina
to create
neural impulses is in the environment (always, pace
Bill
P., recognizing
that there may be no “real reality” out
there).

Well I think there
is something outside the brain.

Don’t we all? The problem Bill P keeps bringing up is that we
can’t prove it, which seems to be true. And since he called us both
out on the question a few days ago, I thought I had better put the
bracketed comment in there.

Martin

[Martin_Taylor2]

[Martin Taylor 2006.11.25.10.28]

[From Rick Marken (2006.11.24.2320)]

Martin Taylor (2006.11.24.20.26)

Fred Nickols (2006.11.24.1431 EST)] --

If I'm trying to look at something and a bright light is causing me to squint or shade my eyes, it seems to me that this bright light is a disturbance to my goal of vewing something. On the other hand, if the light doesn't interfere with my ability to do that, then it's not a disturbance.

Do I have the correct?

Quite succinct. Yes.

Succinct, yes, but not correct in the PCT sense of disturbance as "the variable that is responsible for the change in the input quantity".

Not for the first time, I'm afraid I have to disagree with Rick. I think he is looking at effects in a different control loop than the one Fred specified.

Fred is describing a disturbance as a change in the input quantity that is caused by some other unnamed variable. This is not the PCT meaning of disturbance.

The "something" Fred is looking at is reflected light from some object. The reflected light is an input quantity. When this input quantity is in stare x it is seen as "something", which is presumably the desired state of the input . The bright light represents a change in the input quantity -- when the value of this quantity goes way above x. Squinting is the output that brings the input quantity back toward x. When there is no "interfering" bright light there is no "disturbance" in the sense that there is no change in the input quantity; it remains at x.

Fred (as I interpret him) is talking about a control loop controlling for seeing some object. Before the "bright light" came on, he could see it well. When the light came on, he could see it less well -- maybe he couldn't even see it at all. So the control system controlling his perception of how well he could see produced output to counter the effect of the bright light. To me, that shows that the bright light constituted a disturbance. The example correctly and succinctly illustrates the meaning of the word.

I didn't understand Fred to be saying he cared about how bright the object was that he was examining, any more than the driver controlling for staying in lane cares about the angle of the steering wheel. Fred's squinting or eye shading are analogous to turning the wheel when the car is buffeted by wind.

Disturbances, in the PCT sense of "variables responsible for changes (variations) in an input quantity" are always there. They are not sometimes a disturbance and other times not.

This is technically correct (though see below), provided one keeps in mind a point that is often missed in talking about control: disturbances can be there with magnitude zero. There always _could be_ something coming from outer space that would disturb _any_ control system, however insulated it might be. It's the same kind of thing as saying that one continues to control even when the error is zero. You quite often hear people say that you aren't controlling because there's no reason to if there's no error. But you are, and in the same sense, any control system is always subject to disturbance.

There is another point, though, one that has been brought out in my interchanges with Bjorn: it's not "changes in the input quantity" that indicate the diturbance. In a well controlled system, the input quantity hardly changes. Thinking of a disturbance as something that changes the input quantity leads to the kind of confusion Bjorn and I have been trying to straighten out, between simply perceiving the world as it changes and disturbances to control.

A disturbance contributes to the input quantity, for sure, but it is "changes in the output quantity" that signal changes in the disturbance quantity. Change in the input quantity with no change in output means no disturbance.

Martin

[Bill_Powers1]

[From Bill Powers (2006.11.25.0930 MST)]

Rick Marken (2006.11.24.2320) --

Fred Nickols (2006.11.24.1431 EST)] --

If I'm trying to look at something and a bright light is causing me to squint or shade my eyes, it seems to me that this bright light is a disturbance to my goal of vewing something. On the other hand, if the light doesn't interfere with my ability to do that, then it's not a disturbance.

Do I have the correct?

Quite succinct. Yes.

Succinct, yes, but not correct in the PCT sense of disturbance as "the variable that is responsible for the change in the input quantity". Fred is describing a disturbance as a change in the input quantity that is caused by some other unnamed variable. This is not the PCT meaning of disturbance.

I don't see that he said what you say here in quotes. He said the bright light is a disturbance to my goal of viewing something else. He could have meant that it is a disturbance in the sense of a causal variable that would prevent his viewing something if he were not squinting. Of course his squinting keeps the bright light from having too much effect on his view of the something, so the effect of the disturbing variable, the bright light, is much less than it would have been if he had not squinted.

Is that what you meant, Fred?

Strictly speaking, Rick, I wonder if this is really clear:

Disturbances, in the PCT sense of "variables responsible for changes (variations) in an input quantity" are always there. They are not sometimes a disturbance and other times not.

What's hard to say clearly is that while the disturbing variable may always be there, contributing to the state of the input quantity, the output quantity is also there as long as the disturbance is there, contributing an opposing effect on the input quantity. So the net effect on the input quantity may be very small. The net effect is just large enough to produce enough error to produce enough output quantity to cancel all but that small amount of net effect. I'm ignoring onset and offset transients.

If we could wrap sentences around in circles all this might be easier to say. Unfortunately, sentences come out like beads on a string, one bead at a time. By the time the end of the sentence comes out, the beginning of it is already fading away.

Best.

Bill P.

[Richard_Marken2]

[From Rick Marken (2006.11.25.0915)]

Bill Powers (2006.11.25.0930 MST)]

Rick Marken (2006.11.24.2320) --

Fred Nickols (2006.11.24.1431 EST)] --

If I'm trying to look at something and a bright light is causing me to squint or shade my eyes, it seems to me that this bright light is a disturbance to my goal of vewing something. On the other hand, if the light doesn't interfere with my ability to do that, then it's not a disturbance.

Do I have the correct?

Quite succinct. Yes.

Succinct, yes, but not correct in the PCT sense of disturbance as "the variable that is responsible for the change in the input quantity". Fred is describing a disturbance as a change in the input quantity that is caused by some other unnamed variable. This is not the PCT meaning of disturbance.

I don't see that he said what you say here in quotes. He said the bright light is a disturbance to my goal of viewing something else. He could have meant that it is a disturbance in the sense of a causal variable that would prevent his viewing something if he were not squinting.

Yes. The telltale phrase that, in my mind, reveals the misconception is "if the light doesn't interfere with my ability to do that, then it's not a disturbance". I don't believe Fred would not have said this if he meant that the bright light is the momentary state of a disturbing variable.

Strictly speaking, Rick, I wonder if this is really clear:

Disturbances, in the PCT sense of "variables responsible for changes (variations) in an input quantity" are always there. They are not sometimes a disturbance and other times not.

What's hard to say clearly is that while the disturbing variable may always be there, contributing to the state of the input quantity, the output quantity is also there as long as the disturbance is there, contributing an opposing effect on the input quantity. So the net effect on the input quantity may be very small. The net effect is just large enough to produce enough error to produce enough output quantity to cancel all but that small amount of net effect. I'm ignoring onset and offset transients.

Yes, that adds clarity! I think this concept -- of a disturbance in the PCT sense -- is a difficult one to get. It's easy to see in a diagram but I think it's hard to get used to mapping it correctly to real situations. Maybe it would be a good exercise to analyze several different real life control situations and have people identify the variables that correspond to the output variable, disturbance variable and input quantity. Here's some examples off the top of my head:

1. Watching a pretty girl go by.

2. Watching a distant sailboat on the sea.

3. Putting books into a bookcase.

4. Making change when given a $5 for a $3.50 purchase.

5. Driving to the airport.

I'm sure you, Bill, can come up with some better ones. I think if we try to identify an input quantity, disturbance and output variable for examples like this it would be very instructive.

Best

Rick

Richard S. Marken Consulting
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400

[Fred_Nickols1]

[From Fred Nickols (2006.11.25.1425 EST)] --

[From Bill Powers (2006.11.25.0930 MST)]

Rick Marken (2006.11.24.2320) --

>>>Fred Nickols (2006.11.24.1431 EST)] --
>>>
>>>If I'm trying to look at something and a bright light is causing
>>>me to squint or shade my eyes, it seems to me that this bright
>>>light is a disturbance to my goal of vewing something. On the
>>>other hand, if the light doesn't interfere with my ability to do
>>>that, then it's not a disturbance.
>>>
>>>Do I have the correct?
>>
>>Quite succinct. Yes.
>
>Succinct, yes, but not correct in the PCT sense of disturbance as
>"the variable that is responsible for the change in the input
>quantity". Fred is describing a disturbance as a change in the input
>quantity that is caused by some other unnamed variable. This is not
>the PCT meaning of disturbance.

I don't see that he said what you say here in quotes. He said the
bright light is a disturbance to my goal of viewing something else.
He could have meant that it is a disturbance in the sense of a causal
variable that would prevent his viewing something if he were not
squinting. Of course his squinting keeps the bright light from having
too much effect on his view of the something, so the effect of the
disturbing variable, the bright light, is much less than it would
have been if he had not squinted.

Is that what you meant, Fred?

Yes, that is what I meant.

Regards,

Fred Nickols
nickols@att.net

[Martin_Taylor2]

[Martin Taylor 2006.11.25.17.16]

[From Fred Nickols (2006.11.25.1431 EST)] --

Let's assume a starting point with all elements present. The value of brightness is 1. With that level of brightness, I don't need to squint and I see the object just fine. The brightness of the light is not a disturbance. If the brightness exceeds 1, the object becomes more difficult to see but I can offset it by squinting.

In that case, I would have said that the disturbing variable was present initially, with a value of zero (even though the brightness was 1; the brightness is not itself a measure of the magnitude of the effect of the disturbing variable on the controlled perception -- seeing the object).

The value of squint goes up to offset a value of brightness greater than 1. The brightness is a disturbance but a manageable one. Ditto for decreasing the brightness of the light. At some point it's so dark I can't see the object and no amount of squinting will help. Again, brightness is a disturbance but because it's too little, not too much. So, I turn on the ceiling light and everything's fine.

The point is the one Rick was making: that having a variable x whose value at the moment is zero is different from not having a variable x.

I thought your original example was excellent. Your elaboration of it suggests I misinterpreted.

Martin

[Fred_Nickols1]

[From Fred Nickols (2006.11.25.1935 EST)] --

[Martin Taylor 2006.11.25.17.16]

<snip>

I thought your original example was excellent. Your elaboration of it
suggests I misinterpreted.

Okay, let's try a different one. This one mixes up the brightness of light with driving.

I'm headed west on US Hwy 36, going from Mount Vernon, OH (where I live) to see my granddaughter over in Cardington, OH (where she lives). It's late afternoon and the sun is low in the western sky. As I drive along OH Hwy 229, the sun is getting in my eyes, making it difficult for me to see the highway, the lane markers, oncoming traffic and so on. In plain talk, the sunlight is interfering with my ability to control the position of the car and to take into account oncoming traffic. So, I pull down the sun visor and don my sunglasses. So far so good. My visual acuity is greatly improved.

In my benighted grasp of PCT, I would say that my goal is one of driving safely, of having my car under suitable control. At some lower level, I might say my goal is one of maintaining visual acuity. Whatever the goal, I view the sun's rays in my eyes as a disturbance, as something that is upsetting my ability to keep my perceptions aligned with my goals or intentions.

Now I don't know if there's really a sun out there. Sure looks like it to me but what the heck, who knows? I do know that my ability to see is hampered by what looks like bright light coming from what most of us call the sun. It really gets wacky when you consider that Route 229 (or whatever it is I see out there and call Route 229 - if I can be said to see anything at all) twists and turns and goes up and down over hill and rise, which makes those pesky rays from the sun (as well as the sun itself) disappear from time to time behind hills and stands of trees. So, those darn rays (or is it the sun?) doesn't impede my vision in any kind of constant or continuing way; it's more or less intermittent.

So what's the disturbance (if any) here? Is it the sun or its rays or my perceptions of them (or it) or, as my granddaughter would say, "Whatever"! If you ask me, I would say something that is obviously stupid such as "the brightness of the sun's rays in my eyes impedes my ability to drive safely" - and I would therefore conclude that this brightness was a disturbance in relation to my goal of driving safely. That I am able to compensate for it doesn't alter its place in the PCT scheme of things - or does it?

I await enlightenment (no pun intended).

Regards,

Fred "Unenlightened" Nickols
nickols@att.net

[Martin_Taylor2]

Martin Taylor [2006.11.25.23.10]

[From Fred Nickols (2006.11.25.1935 EST)] --

[Martin Taylor 2006.11.25.17.16]

<snip>

I thought your original example was excellent. Your elaboration of it
suggests I misinterpreted.

Okay, let's try a different one. This one mixes up the brightness of light with driving.

I'm headed west on US Hwy 36, going from Mount Vernon, OH (where I live) to see my granddaughter over in Cardington, OH (where she lives). It's late afternoon and the sun is low in the western sky. As I drive along OH Hwy 229, the sun is getting in my eyes, making it difficult for me to see the highway, the lane markers, oncoming traffic and so on. In plain talk, the sunlight is interfering with my ability to control the position of the car and to take into account oncoming traffic. So, I pull down the sun visor and don my sunglasses. So far so good. My visual acuity is greatly improved.

In my benighted grasp of PCT, I would say that my goal is one of driving safely, of having my car under suitable control. At some lower level, I might say my goal is one of maintaining visual acuity. Whatever the goal, I view the sun's rays in my eyes as a disturbance, as something that is upsetting my ability to keep my perceptions aligned with my goals or intentions.

OK. Let's match this with the classic example of a wind gust that would push the car out of its lane if you didn't turn the steering wheel. What is being disturbed there? Isn't it the perception of the position of the car on the road? The output that compensates mappens to change the reference value for the angle of the steering wheel, but that angle is not disturbed by the wind gust. The driver could perfectly well keep the wheel at the same angle and would feel no effect from the wind gust if he did. But the car would drift sideways on the road.

In that example, the control system for the steering wheel is part of the feedback path external to the control system for perceiving the position of the car in its lane.

Now consider your example. Seeing the lane markers and oncoiming traffic are independent perceptions, which combine in a perception of your current safety. Seeing what road you are on, and turning at the appropriate place are other perceptions, which combine (with others) as part of the system that controls a perception of where you are going. That's different from the perception of road safety, though the road safety perception is in the feedback path for perceiving yourself to be reducing the error in your perception of where you are (as compared to its reference value, where you want to arrive).

Below the perception controlling for perception of driving safety are perceptions of how well you can see the lane markers and the oncoming traffic. You can judge which perception is importantly disturbed by the low sun by a thought experiment. Suppose the sun miraculously made it easier to see the traffic, while making it harder to see the lane markings. Would you then act to shade it? Probably yes if so doing would increase your perception of driving safety, no if it would make you perceive your driving as less safe. If that is so, then in the real case, what is being disturbed is the perception of safety. You pull down the visor and don sunglasses to increase your perception of safety. The mechanism is that you then can see the markings and the traffic better. But it's an output mechanism that counters a disturbance to your perception of safety.

It does happen to be true that the same sun would be a disturbance to your perception of how well you cna see lane markings, and if that were your only concern, you could legitimately argue that it is the disturbance to your clarity of sight that mattered. The hypothetical example in which the effects on the perception of traffic and on the perception of lane markings are opposite says that it the disturbance is primarily to the safety perception.

So what's the disturbance (if any) here? Is it the sun or its rays or my perceptions of them (or it) or, as my granddaughter would say, "Whatever"! If you ask me, I would say something that is obviously stupid such as "the brightness of the sun's rays in my eyes impedes my ability to drive safely" - and I would therefore conclude that this brightness was a disturbance in relation to my goal of driving safely. That I am able to compensate for it doesn't alter its place in the PCT scheme of things - or does it?

I await enlightenment (no pun intended).

So, your intuitive conclusion is exactly the same as the conclusion I arrive at by a slightly more analytic process.

I would add a caveat, however, which is that there's only one degree of freedom for output, which makes it impossible to control separately for seeing clearly the lane markings and the oncoming traffic. In my hypothetical situation, that would put the two control systems in conflict, and that, in turn, would make it considerably more difficult for an external observer to assert that "driving safely" was the variable being disturbed, rather than the clarity of seeing the lane markings and the traffic. You might simpy like to see things clearly, and not care about safety. To an external observer, the only real conclusion would be that the sun does disturb some controlled perception.

All of which isn't as elightening an answer as you might wish.

Martin

[Fred_Nickols1]

[From Fred Nickols (2006.11.26.0722 EST)] -

[Martin Taylor 2006.11.25.22.55]

<snip>

"disturbance" is a concept that applies only to a controlled
perception, not to external events that change an uncontrolled
perception.

OK. I think I've got that and I think I thought that all along. But those are just words and I don't know that what they mean to me is what they mean to you. In any event, that was part of what I was trying to get at with my original example - that a disturbance is only a disturbance in relation to something you're controlling for. Despite having gone round and round, I think that's borne out by the conversation so far.

The effective way to distinguish them is to see whether
the conditions indicating control are satisfied. The putative
disturbance is providing an instance of "the Test".

That is also kind of what I was trying to get at but from a slightly different angle. Let's see if I can convey it without muddying things up even more.

"The Test" - as I understand it - consists of disturbing some variable another person is thought to be controlling. If the person acts to counter the disturbance, there's some evidence that the person is indeed controlling for that variable.

Now, flip it around. Let's proceed on the assumption that the person is indeed controlling for variable X or whatever it is and the aim is to determine if some other factor (Y) is indeed a disturbance. If you remove what is believed to be the disturbance and behavior (output) changes too, is that evidence that Y was a disturbance but now no longer needs to be compensated for?

I'm thinking of the sun dipping below the horizon and now I no longer need my sunglasses or sun visor. The sun's rays have ceased to be a disturbance.

Regards,

Fred "dimly lit" Nickols
nickols@att.net

[Martin_Taylor2]

[Martin Taylor 2006.11.26.10.36]

[From Rick Marken (2006.11.25.2240)]

Martin Taylor (2006.11.25.22.55) --

I hope this is the last one.

Well, maybe one more.

[In answering this message, my thinking evolved. I've left the course of my changes of thought unedited. I hope it's not too confusing that what I say toward the end has a slant different from what I say at the start, and different from what I have said in earier messages.]

The real essential point, which you seem to be avoiding, is that "disturbance" is a concept that applies only to a controlled perception, not to external events that change an uncontrolled perception.

Gee, I thought this discussion was about the meaning of "disturbance" in PCT. How we got way over to wherever the heck it is that we are, I don't know. And how you can say that I'm avoiding the "real essential point" that "a disturbance is a concept that applies only to a controlled perception" is beyond me. This "real essential point" is part of the PCT definition of a "disturbance" that I just posted today: "Any variable in the environment of a control system that a) contributes to changes in the controlled input quantity and b) is not controlled by the same control system". The term "controlled input quantity" is equivalent to "controlled perception", by the way, in case that was confusing you.

So we are, an have been, in agreement. I'm not clear why you posted that part.

Now let's see if you can use a diagram like the one Bill provided (below) to show the disturbance, input quantity, output variable and reference involved in watching a pretty girl go by.

If you are old enough, there may be no disturbance at all. If you are a style-conscious female, there may be disturbance to your conrolled perception of self. Who knows?

Otherwise... What might you be controlling for that would be disturbed? That's the first question that must be answered before your question can be addressed. Let's suppose you are controlling for something that might be disturbed.

Just as an example, let's say you are trying to take a picture of an advertising sign, and the girl obscures part of it. You might act by asking her to move away from your sight-line. Here's Bill's picture as you posted it with the words changed to suit the example.

                                                                                                            r
                                 seeing an
                                unobscured
                                   sign

                                     >
                    --------------->Comp ------------->
                    > >
girl walked ----->visibility of object <------ request to
in front of sign move away
     d qi qo

Will that do?

You can write the same diagram for any case of a disturbance to a controlled variable.

In the case of the girl, if you are a young man I imagine that her walking by might disturb a whole lot of different controlled perceptions apart from the visibility of the sign, and might induce quite a few conflicts, the end result of which would most probably be no overt action at all.

This example leads me to retract my overstrong assertion that the indication of a disturbance is a countervailing change of output. If an external event disturbs two or more controlled perceptions in such a way that the control outputs oppose each other, the external observer may not see output, and therefore might conclude there had been no disturbance.

All of which makes the PCT definition rather more abstract than it might otherwise be. Morevover, it brings into question the usefulness of the distinction that you, Bill and I have all made: that the concept of "disturbance" applies only to a controlled perception.

Perhaps it remains useful to assert that "disturbance" applies only to a controlled variable when we are analyzing isolated control loops. An event that influences the input quantity of an isolated control loop can be observationally distinguished from both an event that does not influence that input quantity and an event that influences the quantity but the perception is not being controlled. But when the control loop is in a system of control loops the observational distinction is harder to make.

Even with an isolated control loop, when one brings in the concept of degree of control, the distinction between disturbing and non-disturbing changes in perception becomes unclear. When control is ideally perfect, the output exactly counters the disturbance. That condition is impossible to achieve in practice (like absolute zero in temperature). The output always fails to match the disturbance in some degree, and the input quantity is affected by the disturbance.

At the other extreme of control quality, the loop gain is zero. But zero is just a point on a continuum between a negative value that results in good control, and a positive value that results in an exponential increase in the effect of the "disturbance". Having zero loop gain is not the same as not having a loop, just as having zero error differs from not controlling.

Loop gain of zero can occur for different reasons. In everyday language, two possible ones are firstly that the person doesn't care about the change in perception induced by an external event, and doesn't try to alter it (is not controlling), and secondly that the person cares, produces output, but that output fails to influence the perception (controls ineffectively). Both cases indicate a loop gain of zero, but naively I would say that only in the second case is there a disturbance.

I'm thinking out loud here, and almost certainly contradicting things I've written earlier, but I'm beginning to think that whether a perceptible external event constitutes a disturbance cannot reliably be decided by an external observer.

Can it be decided by the "Observer" in the person that might be being disturbed?

There is more here than meets the eye, and my glib but long resposes to Bjorn may need rethinking.

"Maybe one more", indeed!

Martin