Signal, Schmignal?

CSGnet Archive CSG1997
1

[From Bruce Abbott (970829.0855 EST)]

Rick Marken (970828.2200) --

A suspicious person (who, me? ) might assume that substitution
of the term _perceptual signal_ for _ perception_ is sometimes
done intentionally to let the controlled _perception_ (say,
the rate at which food pellets appear) play another role,
such _stimulus_ or _reinforcer_. While I am not _that_
suspicious(;-), like you I do see considerable merit in
speaking about control as clearly as possible.

Dear PCT Police,

1. The _only_ signal that is under direct control in a control system is
    the perceptual signal.

2. A stimulus is that which affects the rate of firing of a sensory
    neuron, usually via a specialized structure called a sensory receptor.
    In a PCT control system, the controlled environmental variable is a
    stimulus. The role of the CEV does not change merely by noting that it
    is a stimulus, i.e., something that can be transduced by a sensory
    receptor and represented thereafter in the nervous system as a time-
    varying series of neural impulses.

3. A controlled perception does not correspond to a reinforcer. A
    reinforcer is something that affects the CV in such a way as to reduce
    error in the controlled perception.

I hope you won't be issuing citations for proper use of these terms. The
jails are over-crowded as it is.

Sincerely,

A Concerned Citizen

2

[From Rick Marken (970829.0810)]

Bruce Abbott (970829.0855 EST) --

1. The _only_ signal that is under direct control in a control
system is the perceptual signal.

Irrelevant. The violation was using the term _perceptual signal_
where _perception_ was appropriate. Although it is, indeed, the
perceptual signal that is controlled in a control loop, we don't
test to determine which _perceptual signal_ is under control (as
you implied in your description of the test); we test to determine
which _perception_ (or, better, which perceptual _variable_)
is under control.

2. A stimulus is that which affects the rate of firing of a sensory
    neuron

In a control loop, the correlate of a "stimulus" is a disturbance to
the controlled variable; the controlled variable itself, the variable
that actually affects the rate of firing of a sensory neuron, is not
a stimulus any more than it is a response -- because it is _both;
the controlled variable is a variable in a _closed loop_.

3. A controlled perception does not correspond to a reinforcer.
A reinforcer is something that affects the CV in such a way as to
reduce error in the controlled perception.

Then there is no such thing as a reinforcer because there in no
independent variable in a control loop that "affects the CV in
such a way as to reduce error in the controlled perception".
The only thing that can affects the CV in this way is the
_combined_ effect of disturbance and output.

I hope you won't be issuing citations for proper use of these
terms.

I think a few weeks in PCT traffic school should cool your heels.

Best

A Concerned PCT Policeman

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

3

[From Bruce Abbott (970829.1100)]

Rick Marken (970829.0810) --

Bruce Abbott (970829.0855 EST)

1. The _only_ signal that is under direct control in a control
system is the perceptual signal.

Irrelevant. The violation was using the term _perceptual signal_
where _perception_ was appropriate. Although it is, indeed, the
perceptual signal that is controlled in a control loop, we don't
test to determine which _perceptual signal_ is under control (as
you implied in your description of the test); we test to determine
which _perception_ (or, better, which perceptual _variable_)
is under control.

In other words, you wish to distinguish between the perceptual signal (a
neural current) and some particular aspect of that perceptual signal, such
as its value, rate of change, or integral? Is this what you mean?

2. A stimulus is that which affects the rate of firing of a sensory
    neuron

In a control loop, the correlate of a "stimulus" is a disturbance to
the controlled variable; the controlled variable itself, the variable
that actually affects the rate of firing of a sensory neuron, is not
a stimulus any more than it is a response -- because it is _both;
the controlled variable is a variable in a _closed loop_.

This sort of difference often lies at the heart of our disagreements. You
want to define "stimulus" as something like "a goad to action,"
(reflexology) whereas I want to define it as that which "stimulates," i.e.,
affects the rate of firing of impulses in sensory nerves (sense physiology).
Confusion arises when I define a term in one way and you then respond as if
I intended some other definition. A disturbance generally is not directly
sensed, therefore it is not a stimulus as I have defined the term.

3. A controlled perception does not correspond to a reinforcer.
A reinforcer is something that affects the CV in such a way as to
reduce error in the controlled perception.

Then there is no such thing as a reinforcer because there in no
independent variable in a control loop that "affects the CV in
such a way as to reduce error in the controlled perception".
The only thing that can affects the CV in this way is the
_combined_ effect of disturbance and output.

There is such a thing as a reinforcer because there are actions that affect
the CV in such a way as to reduce error in the controlled perception. My
definition does not require a reinforcer to be an independent variable
within the loop, any more than "output" must be an independent variable
within the loop.

Regards,

Concerned Citizen

4

[From Bill Powers (970829.0907 MDT)]

71 years old today! Incredible.

Bruce Abbott (970829.0855 EST)--

Dear PCT Police,

Only a warning this time.

1. The _only_ signal that is under direct control in a control system is
   the perceptual signal.

The only time this matters is when the perceptual input function changes
its form. The CV is always the inverse input function of the perceptual
signal, by definition. If the input function changes, the CV becomes a
different variable, but if the observer doesn't know it has changed,
confusion may result. When we observe the CV, we presumably are using input
functions similar to those in the observed system. So the state of the CV
always corresponds exactly to the state of the perceptual signal we assume
to be representing it (observation noise aside). If the perceptual signal
is controlled, the CV is controlled.

Generally, the CV is what we _observe_; the perceptual signal is inferred
in terms of a particular model. The Test can be applied only to the
observed CV, not to the (hypothetical) perceptual signal.

2. A stimulus is that which affects the rate of firing of a sensory
   neuron, usually via a specialized structure called a sensory receptor.
   In a PCT control system, the controlled environmental variable is a
   stimulus.

What is usually called "the stimulus" is a distal variable which we would
identify as a disturbance. For example, if you turn on a light and move it,
you can observe an organism's eyes tracking it. So it appears that the
motion of the light is a stimulus, with the eye movement being a response
to it. However, if we calculate or observe the position of the retinal
image, we will see that the actual proximal stimulus is an image that is
maintained almost stationary near the center of vision: that is the "CV."
The movement of the light tends to disturb the CV, but the eye movements
counteract the effects of the disturbance almost completely.

The role of the CEV does not change merely by noting that it
   is a stimulus, i.e., something that can be transduced by a sensory
   receptor and represented thereafter in the nervous system as a time-
   varying series of neural impulses.

Not all CVs are stimuli in this sense. For example, suppose you are in a
tracking experiment and decide to keep the distance between target and
cursor at 2 inches. While the images of target and cursor are visual
stimuli, the distance between them is not: it causes no sensory receptors
to fire.

3. A controlled perception does not correspond to a reinforcer. A
   reinforcer is something that affects the CV in such a way as to reduce
   error in the controlled perception.

I don't understand what you mean here. In the general model, there are two
things that affect the CV: the output of the system (via the feedback
function) and disturbances (the sum of all effects independent of the
system's output). Are you saying that there is something else that affects
the CV that doesn't belong in either category?

You might want to reconsider your wording when you say "in such a way as to
reduce error in the controlled perception." There is no error in the
controlled perception, is there? The controlled perception simply
corresponds at all times to the state of the observed CV (if we have
defined it right). The error is a relationship between the controlled
perception and another signal, the reference signal. You seem to be
implying, I assume unintentionally, that the perception and the CV can vary
in different ways. Just for the record, they cannot: if the CV is
(correctly) known, the perceptual signal is known. They always correspond
_exactly_.

Best,

Bill P.

5

[From Rick Marken (970829.0930)]

Bruce Abbott (970829.1100)--

In other words, you wish to distinguish between the perceptual
signal (a neural current) and some particular aspect of that
perceptual signal, such as its value, rate of change, or
integral? Is this what you mean?

No. I mean to distinguish between a neural signal (perceptual
signal) and the aspect of experience that is represented by
that signal (eg, rate of food pellet delivery, degree of influence
of the Florentine school on Raphaeal's "Transfiguration")

There is such a thing as a reinforcer because there are actions
that affect the CV in such a way as to reduce error in the
controlled perception.

See officer Powers' report [Bill Powers (970829.0907 MDT)].

Regards,

Concerned Citizen

Not of PCTland. If you can't help thumbing your nose at the rules
over here you might consider going back to Behaviorismland, where
the difference between perceptions and perceptual signals doesn't
make any difference, where "stimuli" cause neural impulses, and
where the reinforcers grow on every vine.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

6

[From Rick Marken (970828.2200)]

Me:

But I do have a question: How do I apply a disturbance to a
perceptual signal in order to Test whether or not that signal
is under control?

Bill Powers (970828.2009 MDT) --

Seems to me we often speak of controlling perceptions when
what we mean is controlling something that we presume is
being perceived as we perceive it.

Yes, indeed. And we speak of testing to determine the perception
that is under control. But we rarely (if we are careful) speak
of testing to determine the perceptual _signal_ that is under
control.

When we speak of testing to determine the _perception_ that is
under control, it is clear that we are trying to discover the
aspect of our own perceptual experience (such as the rate at
which food pellets appear) that corresponds to the perception
being controlled by the organism. On the other hand, when we
speak (carelessly) of testing to determine the perceptual
_signal_ that is under control, it implies that we are testing
to determine which of the many afferent neural signals in the brain
is under control. I think this harmless looking little change
in locution subtley points our attention away from the most
important (and least noticed) aspect of the control process --
the perceptual variable controlled by the organism - and toward
an anonymous surrogate of that perception. It suggests (by
omission) that it is not a perceptual aspect of, say, the food
pellets that is controlled; it's just an anonymous neural signal.

A suspicious person (who, me? ) might assume that substitution
of the term _perceptual signal_ for _ perception_ is sometimes
done intentionally to let the controlled _perception_ (say,
the rate at which food pellets appear) play another role,
such _stimulus_ or _reinforcer_. While I am not _that_
suspicious(;-), like you I do see considerable merit in
speaking about control as clearly as possible.

But maybe I can lighten up a bit and just not make a big
thing out of it; perception, perceptual signal, whatever.

Best

The PCT Police

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

7

[From Bruce Abbott (970829.1310 EST)]

Bill Powers (970829.0907 MDT)]

71 years old today! Incredible.

Congratulations, Bill, and Happy Birthday! Still a quarter century or more
left on the meter, if your Dad's longevity is any guide . . . (:->

Bruce Abbott (970829.0855 EST)

1. The _only_ signal that is under direct control in a control system is
   the perceptual signal.

The only time this matters is when the perceptual input function changes
its form. The CV is always the inverse input function of the perceptual
signal, by definition. If the input function changes, the CV becomes a
different variable, but if the observer doesn't know it has changed,
confusion may result. When we observe the CV, we presumably are using input
functions similar to those in the observed system. So the state of the CV
always corresponds exactly to the state of the perceptual signal we assume
to be representing it (observation noise aside). If the perceptual signal
is controlled, the CV is controlled.

I agree in general, but we have no guarantee that the perceptual signal we
perceive is identical to the one being perceived by someone else; for
example, the other guy may be colorblind or have poor visual acuity.

Generally, the CV is what we _observe_; the perceptual signal is inferred
in terms of a particular model. The Test can be applied only to the
observed CV, not to the (hypothetical) perceptual signal.

Yes, although one hopes that the perceptual signal will be capable of direct
measurement someday. (It already is in some limited cases.)

2. A stimulus is that which affects the rate of firing of a sensory
   neuron, usually via a specialized structure called a sensory receptor.
   In a PCT control system, the controlled environmental variable is a
   stimulus.

What is usually called "the stimulus" is a distal variable which we would
identify as a disturbance. For example, if you turn on a light and move it,
you can observe an organism's eyes tracking it. So it appears that the
motion of the light is a stimulus, with the eye movement being a response
to it. However, if we calculate or observe the position of the retinal
image, we will see that the actual proximal stimulus is an image that is
maintained almost stationary near the center of vision: that is the "CV."
The movement of the light tends to disturb the CV, but the eye movements
counteract the effects of the disturbance almost completely.

See my earlier response to Rick. I merely point out that the term
"stimulus" need not connote something that elicits a response; in sense
physiology it is simply something that affects the output of a sensory
receptor, i.e. a source of stimulation, the input to a sensory transducer.

The role of the CEV does not change merely by noting that it
   is a stimulus, i.e., something that can be transduced by a sensory
   receptor and represented thereafter in the nervous system as a time-
   varying series of neural impulses.

Not all CVs are stimuli in this sense. For example, suppose you are in a
tracking experiment and decide to keep the distance between target and
cursor at 2 inches. While the images of target and cursor are visual
stimuli, the distance between them is not: it causes no sensory receptors
to fire.

True, although some are. CVs are often derived from stimuli via neural
"computation," as in your distance example. Perceptions are neural signals;
stimuli are not.

3. A controlled perception does not correspond to a reinforcer. A
   reinforcer is something that affects the CV in such a way as to reduce
   error in the controlled perception.

I don't understand what you mean here. In the general model, there are two
things that affect the CV: the output of the system (via the feedback
function) and disturbances (the sum of all effects independent of the
system's output). Are you saying that there is something else that affects
the CV that doesn't belong in either category?

No. The output of a control system affects the CV via the feedback
function, which translates the output into some physical effect on the CV.
Example: Consuming a food pellet (output) increases the level of nutrients
in the bloodstream, the CV. If the current level of the CV is below its
reference, the food pellet is a reinforcer because swallowing it reduces the
error.

You might want to reconsider your wording when you say "in such a way as to
reduce error in the controlled perception." There is no error in the
controlled perception, is there? The controlled perception simply
corresponds at all times to the state of the observed CV (if we have
defined it right). The error is a relationship between the controlled
perception and another signal, the reference signal. You seem to be
implying, I assume unintentionally, that the perception and the CV can vary
in different ways. Just for the record, they cannot: if the CV is
(correctly) known, the perceptual signal is known. They always correspond
_exactly_.

Error is defined as a difference between the current level of the CV and its
reference level. Error in a CV is that deviation. It is not, of course, a
property of the CV, but of the relationship between the CV and its
reference. There is nothing wrong with this language; if my mechanic tells
me that my wheels are misaligned, I know perfectly well what he means; I
don't inform him that misalignment is not a property of wheels.
Misalignment, like error, is understood as relative to some standard.

if the CV is
(correctly) known, the perceptual signal is known. They always correspond
_exactly_.

Trivially true by definition: to "know a CV correctly" means to know what
corresponds exactly to the perceptual signal. But the CV we observe and the
"correct" CV may differ. For example, a participant may be instructed to
keep a spot of light at a certain constant intensity by moving a slider up
and down. If the sensitivity to light of the participant's retinas is
improving during this experiment, the "correct" CV will not correspond to
the light intensity measured by an electronic light meter. In that case we
may observe the light intensity to be gradually decreasing, although the
participant would report that she is keeping it relatively constant.

Regards,

Bruce

8

[From Bill Powers (970829.1229 MDT)]

Bruce Abbott (970829.1310 EST)--

I agree in general, but we have no guarantee that the perceptual signal we
perceive is identical to the one being perceived by someone else; for
example, the other guy may be colorblind or have poor visual acuity.

In that case, where the perceptual functions of the controller and the
observer differ, the observer will find disturbances that result in actions
by the controller but do not seem to disturb what the observer is seeing,
and vice versa. This is equivalent to a misidentification of the CV, and
the Test will reveal it. In fact something similar to this was involved in
the discovery of color-blindness.

For purposes of this discussion, let's assume that the CV has passed the
Test, and continues to do so.

Generally, the CV is what we _observe_; the perceptual signal is inferred
in terms of a particular model. The Test can be applied only to the
observed CV, not to the (hypothetical) perceptual signal.

Yes, although one hopes that the perceptual signal will be capable of
direct measurement someday. (It already is in some limited cases.)

True, but not relevant to the current discussion.

2. A stimulus is that which affects the rate of firing of a sensory
   neuron, usually via a specialized structure called a sensory >>>

receptor.

   In a PCT control system, the controlled environmental variable is a
   stimulus.

What is usually called "the stimulus" is a distal variable which we would
identify as a disturbance.

See my earlier response to Rick. I merely point out that the term
"stimulus" need not connote something that elicits a response; in sense
physiology it is simply something that affects the output of a sensory
receptor, i.e. a source of stimulation, the input to a sensory transducer.

Yes, but most perceptual signals are functions of multiple sensory signals,
meaning that there is no way to know what is being perceived except to
apply a similar input function to similar intensity signals (actual or
inferred).
Your designation of a "stimulus" as something that excites a sensory ending
rules out most perceptions above the intensity level. It also includes some
environmental variables that actually function as disturbances; even though
they can be sensed, it is not the perception of the disturbance that leads
to opposition, but the change in the CV that is involved in the control
loop. We have demonstrated that showing the magnitude of the disturbance on
the screen, using another movable bar or a number, detracts from
performance in a tracking task.

Not all CVs are stimuli in this sense. For example, suppose you are in a
tracking experiment and decide to keep the distance between target and
cursor at 2 inches. While the images of target and cursor are visual
stimuli, the distance between them is not: it causes no sensory receptors
to fire.

True, although some are. CVs are often derived from stimuli via neural
"computation," as in your distance example. Perceptions are neural
signals; stimuli are not.

I agree; stimuli are environmental variables. However, since the same set
of multiple stimuli can be inputs to many different perceptual function
computations, there is no way to tell what "the stimulus" is by examining
environmental variables. Most often, "the stimulus" is defined by what the
observer finds salient in his own experiences. And until proven otherwise,
we have to consider that most things designated as stimuli are remote from
the actual sensory processes, and could just as well be disturbances.

···

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

I don't understand what you mean here. In the general model, there are two
things that affect the CV: the output of the system (via the feedback
function) and disturbances (the sum of all effects independent of the
system's output). Are you saying that there is something else that affects
the CV that doesn't belong in either category?

No. The output of a control system affects the CV via the feedback
function, which translates the output into some physical effect on the CV.
Example: Consuming a food pellet (output) increases the level of nutrients
in the bloodstream, the CV. If the current level of the CV is below its
reference, the food pellet is a reinforcer because swallowing it reduces
the error.

You're offering two different definitions of possible CVs. One is consuming
food pellets (individually or at some rate). The other is the level of
nutrients in the bloodstream. Suitable applications of the Test can tell us
if one or both of these variables is a controlled variable. A hierarchical
relation would seem possible; the rate at which pellets are consumed is
controlled at a reference level set by the system that controls the level
of nutrients in the bloodstream.

If the level of nutrients in the bloodstream is the only controlled
variable, then we would not expect a disturbance of the rate of pellet
consumption to be resisted, provided we maintained the nutrient level
constant by some other means. If, however, at constant nutrient level we
were to find that disturbing the rate of pellet consumption resulted in an
opposing change in the output that affects pellet consumption, we would
conclude that pellet consumption is itself a CV. in addition to nutrient
level.

What you seem to be proposing is that a reinforcer is some variable in the
loop just prior to the point where we measure (or deduce) the true CV. If
the CV is nutrient level, then the reinforcer is whatever raises or lowers
the nutrient level. Is that correct? In short, are you saying that the
reinforcer is simply part of the environmental feedback function?

Note that by this definition, the only physical effect of the reinforcer is
its effect on the CV.

You might want to reconsider your wording when you say "in such a way as
to
reduce error in the controlled perception." There is no error in the
controlled perception, is there?

Error is defined as a difference between the current level of the CV and
its reference level. Error in a CV is that deviation. It is not, of
course, a property of the CV, but of the relationship between the CV and
its reference. There is nothing wrong with this language; if my mechanic
tells me that my wheels are misaligned, I know perfectly well what he
means; I don't inform him that misalignment is not a property of wheels.
Misalignment, like error, is understood as relative to some standard.

I thought we were going to try to speak as rigorously as we can. This is
going to be difficult if you want to rely on "knowing perfectly well" what
is meant, as opposed to what is said. If what you say can't be taken
literally, and depends on informal knowledge of what you mean, then I have
to be a mind-reader to know what you intend by your words. More to the
point, so does a general reader seeing your words for the first time. This
will not take us very far toward a scientific theory. If you're going to
ignore formal definitions of terms, I won't be able to hold you (or myself)
to exact definitions, and we just won't get anywhere with this.

I know you resent having your language criticized, but all I can say to
that is "get over it." I don't really think you're arguing against
precision of usage in a scientific discussion. If you don't want such
criticisms, edit before you publish.

if the CV is
(correctly) known, the perceptual signal is known. They always correspond
_exactly_.

Trivially true by definition: to "know a CV correctly" means to know what
corresponds exactly to the perceptual signal. But the CV we observe and
the "correct" CV may differ.

Make up your mind. Either it's trivially true, meaning true, or it's not.
If the CV we observe differs from the actual CV, we have misdefined the CV,
and need to apply the Test some more to refine the definition. Otherwise,
if we can't think of any more tests, the CV corresponds exactly to the
perceptual signal; that is what we must assume. We could be wrong, but if
so we will find that out later when the wrongness makes a difference.

For example, a participant may be instructed to
keep a spot of light at a certain constant intensity by moving a slider up
and down. If the sensitivity to light of the participant's retinas is
improving during this experiment, the "correct" CV will not correspond to
the light intensity measured by an electronic light meter. In that case we
may observe the light intensity to be gradually decreasing, although the
participant would report that she is keeping it relatively constant.

We would conclude that (a) the reference intensity is drifting, (b) the
loop gain is changing, or (c) the input sensitivity is changing. Some
additional work would be needed to see which is the case, but it would be
possible to isolate the reason. If the input sensitivity is changing, we
could redefine the CV in terms of an input function with a decreasing
sensitivity. To maintain despite the evidence that the CV is simply a
function of light intensity would be a mistake; it is a function of light
intensity and time.

Let's keep focussed on the definition of the reinforcer, which is what we
were talking about before the hiatus. Is it correct to say that the
reinforcer is whatever immediately prior variable that the CV depends upon,
and that it is part of the environmental feedback function?

Best,

Bill P.

Oh, Yes, P.S. Rick, what is the difference between a perceptual signal and
a perception? In B:CP I said that a perception is the existence of a
perceptual signal in the appropriate pathway (so consciousness doesn't
enter the definition). Have we changed that?

9

[From Bruce Abbott (970829.1720 EST)]

Bill Powers (970829.1229 MDT) --

Bruce Abbott (970829.1310 EST)

Your designation of a "stimulus" as something that excites a sensory ending
rules out most perceptions above the intensity level. It also includes some
environmental variables that actually function as disturbances; even though
they can be sensed, it is not the perception of the disturbance that leads
to opposition, but the change in the CV that is involved in the control
loop. We have demonstrated that showing the magnitude of the disturbance on
the screen, using another movable bar or a number, detracts from
performance in a tracking task.

Yes, disturbances can be stimuli, CVs can be stimuli, and stimuli can be
sensed that are neither. I am simply pointing out that the term "stimulus"
can be defined in a meaningful and useful way that does not imply that which
triggers a response.

. . . CVs are often derived from stimuli via neural
"computation," as in your distance example. Perceptions are neural
signals; stimuli are not.

I agree; stimuli are environmental variables. However, since the same set
of multiple stimuli can be inputs to many different perceptual function
computations, there is no way to tell what "the stimulus" is by examining
environmental variables. Most often, "the stimulus" is defined by what the
observer finds salient in his own experiences. And until proven otherwise,
we have to consider that most things designated as stimuli are remote from
the actual sensory processes, and could just as well be disturbances.

Yes, the word "stimulus" can be assigned other meanings. That is why, if
one is going to use the term, one should be careful to define what one means
by it.

The output of a control system affects the CV via the feedback

function, which translates the output into some physical effect on the CV.
Example: Consuming a food pellet (output) increases the level of nutrients
in the bloodstream, the CV. If the current level of the CV is below its
reference, the food pellet is a reinforcer because swallowing it reduces
the error.

You're offering two different definitions of possible CVs. One is consuming
food pellets (individually or at some rate). The other is the level of
nutrients in the bloodstream.

I defined the CV in my example as "level of nutrients in the bloodstream."
How is this "two different definitions"? To be precise, what you mean is
not that I _offer_ two definitions, but that you _can think of_ two
different definitions, the one I offered and "consuming food pellets." This
is a hypothetical example in which I am free to proclaim what the CV is, for
the purpose of illustrating what I mean by "reinforcer."

Suitable applications of the Test can tell us

if one or both of these variables is a controlled variable. A hierarchical
relation would seem possible; the rate at which pellets are consumed is
controlled at a reference level set by the system that controls the level
of nutrients in the bloodstream.

Yes, certainly, but irrelevant here.

What you seem to be proposing is that a reinforcer is some variable in the
loop just prior to the point where we measure (or deduce) the true CV. If
the CV is nutrient level, then the reinforcer is whatever raises or lowers
the nutrient level. Is that correct? In short, are you saying that the
reinforcer is simply part of the environmental feedback function?

Yes, it is that which links output to reduction in the error between the CV
and its reference level.

Note that by this definition, the only physical effect of the reinforcer is
its effect on the CV.

Yes indeed. It is not an independent variable, but only one variable within
the closed loop of circular causality.

Error is defined as a difference between the current level of the CV and
its reference level. Error in a CV is that deviation. It is not, of
course, a property of the CV, but of the relationship between the CV and
its reference. There is nothing wrong with this language; if my mechanic
tells me that my wheels are misaligned, I know perfectly well what he
means; I don't inform him that misalignment is not a property of wheels.
Misalignment, like error, is understood as relative to some standard.

I thought we were going to try to speak as rigorously as we can. This is
going to be difficult if you want to rely on "knowing perfectly well" what
is meant, as opposed to what is said. If what you say can't be taken
literally, and depends on informal knowledge of what you mean, then I have
to be a mind-reader to know what you intend by your words. More to the
point, so does a general reader seeing your words for the first time. This
will not take us very far toward a scientific theory. If you're going to
ignore formal definitions of terms, I won't be able to hold you (or myself)
to exact definitions, and we just won't get anywhere with this.

Bill, I'm all in favor of using precise language (which is why I'm going to
the trouble of defining my terms). But sometimes your criticisms of
language usage are a little unfair. By standard usage "error" implies a
comparison, and yet you go out of your way to suggest that I may be using
the term as if it were a property of the CV itself. I could play the same
game with your writing, if I wished. For example, if you were to write
something like "there was a blue cursor on the screen," I might insist, in
the interest of precise communication, that you rephrase your statement,
since "blue" is not a property of the cursor on the screen but only of your
own conscious perception of that cursor. I won't do this, though, because I
know what you mean when you say that the cursor is blue, and to pretend
otherwise would be simply to raise criticism for the sake of criticism.

I don't believe you had any real difficulty understanding what I meant by
"error in the CV." You are just looking for ways to find fault.

if the CV is
(correctly) known, the perceptual signal is known. They always correspond
_exactly_.

Trivially true by definition: to "know a CV correctly" means to know what
corresponds exactly to the perceptual signal. But the CV we observe and
the "correct" CV may differ.

Make up your mind. Either it's trivially true, meaning true, or it's not.

I haven't equivocated, so I have no idea what you are talking about.
"Trivially true" means that logically it must be true (no need for
evidence); in this case it must be true because a "correct" CV is _defined_
as one that corresponds to the perceptual signal.

If the CV we observe differs from the actual CV, we have misdefined the CV,
and need to apply the Test some more to refine the definition. Otherwise,
if we can't think of any more tests, the CV corresponds exactly to the
perceptual signal; that is what we must assume. We could be wrong, but if
so we will find that out later when the wrongness makes a difference.

Of course.

For example, a participant may be instructed to
keep a spot of light at a certain constant intensity by moving a slider up
and down. If the sensitivity to light of the participant's retinas is
improving during this experiment, the "correct" CV will not correspond to
the light intensity measured by an electronic light meter. In that case we
may observe the light intensity to be gradually decreasing, although the
participant would report that she is keeping it relatively constant.

We would conclude that (a) the reference intensity is drifting, (b) the
loop gain is changing, or (c) the input sensitivity is changing. Some
additional work would be needed to see which is the case, but it would be
possible to isolate the reason. If the input sensitivity is changing, we
could redefine the CV in terms of an input function with a decreasing
sensitivity. To maintain despite the evidence that the CV is simply a
function of light intensity would be a mistake; it is a function of light
intensity and time.

Yes, the variable we observe (and take to be the CV) may differ from the
actual CV. We cannot take for granted that what _we_ perceive will be
identical to what the participant perceives. A proper series of tests can
establish what the actual CV is. Sounds like we agree.

Let's keep focussed on the definition of the reinforcer, which is what we
were talking about before the hiatus. Is it correct to say that the
reinforcer is whatever immediately prior variable that the CV depends upon,
and that it is part of the environmental feedback function?

Yes, within the control loop. It does not function there as an independent
variable. In particular, I am not giving it a linear causal role in the
production of "responses" (output).

Regards,

Bruce

10

[From Rick Marken (970829.1500)]

Bill Powers (970829.1229 MDT) --

Oh, Yes, P.S. Rick, what is the difference between a perceptual
signal and a perception? In B:CP I said that a perception is
the existence of a perceptual signal in the appropriate pathway
(so consciousness doesn't enter the definition). Have we changed
that?

No. The difference between a perception and a perceptual signal
is a difference in "point of view". When I see the word "idiot"
on the screen, that word is a perception (for me). It is also (in
theory) a perceptual signal in my brain. The perceptual signal is
(in theory) the perception "idiot". So, from the point of view
of me (a brain) the perception is "idiot"; from the point of
view of you (a theorist) or Dr. Penfield (a neurophysiologist)
the perception is electrical activity in some neuron.

A person doing The Test wants to see what an actor is controlling
_from the actor's point of view_. The Tester wants to see, therefore,
which of his own _perceptions_ (which theoretically exist as
perceptual signals in himself) corresponds to the perception
controlled by the actor (which theoretically exists as a perceptual
signal in the actor).

The Tester could be said to be testing to see which of his own
perceptual signals corresponds to the perceptual signal controlled
by the actor. But that would be _very_ confusing to me because it
would suggest that the Tester has a point of view relative to his
own perceptions (that he can see his perceptions from the point
of view of an observer of his brain) that he cannot actually have
(easily;-)). It also suggests that what we want to know from The
Test is whether a particular neural signal is under control. That
seems to me like saying that The Test aims to determine whether
the sensory current in a thermostat is under control while ignoring
the perception to which this sensory current corresponds (temperature,
humidity, pressure, etc.).

Therefore, I find it much _clearer_ to have The Test described
as an attempt to discover controlled _perceptions_. rather than
controlled perceptual _signals_. Even better would be to say that
The Test is an attempt, by the Tester, to discover which of his
own perceptions correspond to the perceptions controlled by an
actor.

Best

Rick

···

--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

11

[From Bill Powers (970829.1643 MDT)]

Bruce Abbott (970829.1720 EST)--

I'm going to try to avoid striking at every bit of bait dangled in front of
me.

Note that by this definition, the only physical effect of the reinforcer
is its effect on the CV.

Yes indeed. It is not an independent variable, but only one variable
within the closed loop of circular causality.

Let's keep focussed on the definition of the reinforcer, which is what we
were talking about before the hiatus. Is it correct to say that the
reinforcer is whatever immediately prior variable that the CV depends
upon, and that it is part of the environmental feedback function?

Yes, within the control loop. It does not function there as an independent
variable. In particular, I am not giving it a linear causal role in the
production of "responses" (output).

Fine. The next step, I presume, is to follow the effect of the reinforcer
on through the system to the output, assuming that the reference signal is
held constant.

Best,

Bill P.

12

[From Bruce Abbott (970829.2200 EST)]

Bill Powers (970829.1643 MDT) --

Bruce Abbott (970829.1720 EST)

Let's keep focussed on the definition of the reinforcer, which is what we
were talking about before the hiatus. Is it correct to say that the
reinforcer is whatever immediately prior variable that the CV depends
upon, and that it is part of the environmental feedback function?

Yes, within the control loop. It does not function there as an independent
variable. In particular, I am not giving it a linear causal role in the
production of "responses" (output).

Fine. The next step, I presume, is to follow the effect of the reinforcer
on through the system to the output, assuming that the reference signal is
held constant.

Each reinforcer consumed will reduce the amount of error remaining between
the CV and its reference; as error diminishes, output (food consumption)
will diminish; ultimately, average food intake will be just sufficient to
balance nutrient losses due to metabolic activity. (I am simplifying; the
real situation with a real animal is more complex, involving bouts of eating
("meals") separated by intervals of not eating. But I want to keep the
example simple and clear). The observed reduction in food consumption as
nutrient levels approach their reference level is sometimes described as an
effect of "satiation." Satiation undoes what deprivation establishes -- the
latter, in this case, being the lowering of the CV below its reference level.

Comments? Questions?

Regards,

Bruce

13

[From Bill Powers (970830.0500 MDT)]

Rick Marken (970829.1500)--

No. The difference between a perception and a perceptual signal
is a difference in "point of view". When I see the word "idiot"
on the screen, that word is a perception (for me). It is also (in
theory) a perceptual signal in my brain. The perceptual signal is
(in theory) the perception "idiot". So, from the point of view
of me (a brain) the perception is "idiot"; from the point of
view of you (a theorist) or Dr. Penfield (a neurophysiologist)
the perception is electrical activity in some neuron.

A very, very nice exposition, and you are right. We start with experience.
I was speaking theoretically, and that, of course, is not experience -- or
else it's a different kind of experience. The term "perceptual signal" is
part of a model, while "perception" means what we experience without a
model. I'll try to remember that.

Best,

Bill P.

···

A person doing The Test wants to see what an actor is controlling
_from the actor's point of view_. The Tester wants to see, therefore,
which of his own _perceptions_ (which theoretically exist as
perceptual signals in himself) corresponds to the perception
controlled by the actor (which theoretically exists as a perceptual
signal in the actor).

The Tester could be said to be testing to see which of his own
perceptual signals corresponds to the perceptual signal controlled
by the actor. But that would be _very_ confusing to me because it
would suggest that the Tester has a point of view relative to his
own perceptions (that he can see his perceptions from the point
of view of an observer of his brain) that he cannot actually have
(easily;-)). It also suggests that what we want to know from The
Test is whether a particular neural signal is under control. That
seems to me like saying that The Test aims to determine whether
the sensory current in a thermostat is under control while ignoring
the perception to which this sensory current corresponds (temperature,
humidity, pressure, etc.).

Therefore, I find it much _clearer_ to have The Test described
as an attempt to discover controlled _perceptions_. rather than
controlled perceptual _signals_. Even better would be to say that
The Test is an attempt, by the Tester, to discover which of his
own perceptions correspond to the perceptions controlled by an
actor.

Best

Rick
--
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken

14

[From Bill Powers (970830.0455 MDT)]

Bruce Abbott (970829.2200 EST)--

The next step, I presume, is to follow the effect of the reinforcer
on through the system to the output, assuming that the reference signal is
held constant.

Each reinforcer consumed will reduce the amount of error remaining between
the CV and its reference; as error diminishes, output (food consumption)
will diminish; ultimately, average food intake will be just sufficient to
balance nutrient losses due to metabolic activity.

You're going a little too fast. Each reinforcer consumed will increase the
magnitude of the CV (which we are assuming to be lower than its reference
level). Metabolic drains will tend to decrease the CV. If the rate of
consumption of the pellets generates a greater increase in the CV than the
losses between pellet intakes, the CV will increase toward the reference
level. As the CV increases, the error will decrease. Since error is driving
behavior (lever pressing, for example), behavior will decrease.

Are we together still?

Best,

Bill P.

15

[From Bruce Abbott (970830.0840 EST)]

Bill Powers (970830.0455 MDT) --

Bruce Abbott (970829.2200 EST)

Each reinforcer consumed will reduce the amount of error remaining between
the CV and its reference; as error diminishes, output (food consumption)
will diminish; ultimately, average food intake will be just sufficient to
balance nutrient losses due to metabolic activity.

You're going a little too fast. Each reinforcer consumed will increase the
magnitude of the CV (which we are assuming to be lower than its reference
level). Metabolic drains will tend to decrease the CV. If the rate of
consumption of the pellets generates a greater increase in the CV than the
losses between pellet intakes, the CV will increase toward the reference
level. As the CV increases, the error will decrease. Since error is driving
behavior (lever pressing, for example), behavior will decrease.

Are we together still?

Yes. Some reinforcers require an "establishing operation", deprivation of
food, for example, to make food an effective reinforcer. This requirement
is explained by the control model: the establishing operation is simply a
manipulation that produces a given amount of error between the CV and its
reference level. Food consumption cannot reduce error between the CV and
its reference unless error is present to begin with.

Other stimuli do not require a separate establishing operation; the stimulus
itself is said to do so. Rats will press a lever to turn off a footshock
that is being delivered through the floor of the chamber. The shock itself
is described as a "negative" reinforcer, but according to the control model,
turning on the shock is an establishing operation, if it is assumed that the
rat's reference for experiencing shock is zero. Reduction of shock is the
reinforcer, according to my definition. In this case, pressing the lever
reduces the shock to zero, but it has been shown that responding can be
maintained under reductions other than reduction to zero.

So far so good?

Regards,

Bruce

16

[From Bruce Abbott (970830.1020 EST)]

Bill Powers (970830.0832 MDT) --

Bruce Abbott (970830.0840 EST)

. . . Some reinforcers require an "establishing operation", deprivation of
food, for example, to make food an effective reinforcer. This requirement
is explained by the control model: the establishing operation is simply a
manipulation that produces a given amount of error between the CV and its
reference level. Food consumption cannot reduce error between the CV and
its reference unless error is present to begin with.

I agree, but the "establishing operation" has to be done before the
contingency is enabled, doesn't it? We're talking about the situation as it
exists after the establishing operation and just after the contingency is
turned on. This is the time when the CV is zero (or at its lowest), and the
error between the CV and the reference setting is maximum, producing the
maximum rate of behavior. Reinforcers then begin to occur, the CV rises,
the error falls, and the behavior decreases. Are you still in agreement
with this?

Typically the "establishing operation" is done before the contingency is
enabled, as when the rat's body weight is brought down to 80% of its ad
libitum weight. The rat is given no means of correcting the error between
CV and reference until it is placed in the chamber. But it is possible to
place the undeprived rat into the chamber full time, with contingency
between lever and food already established, and simply wait for normal
metabolic losses to create enough error to produce responding. The reason
for doing the establishing operation (food deprivation) prior to enabling
the contingency is that one can thereby establish a specific initial level
of error between the CV and its reference. This level of error is (in
systems other than Skinner's) referred to as the level of "drive" or
motivation. By systematically examining performance under different levels
of "drive," one can relate the "intensity" of behavior to the level of drive.

Comments? Questions?

Regards,

Bruce

17

[From Bill Powers (970830.0832 MDT)]

Bruce Abbott (970830.0840 EST)--

As the CV increases, the error will decrease. Since error is driving
behavior (lever pressing, for example), behavior will decrease.

Are we together still?

Yes. Some reinforcers require an "establishing operation", deprivation of
food, for example, to make food an effective reinforcer. This requirement
is explained by the control model: the establishing operation is simply a
manipulation that produces a given amount of error between the CV and its
reference level. Food consumption cannot reduce error between the CV and
its reference unless error is present to begin with.

I agree, but the "establishing operation" has to be done before the
contingency is enabled, doesn't it? We're talking about the situation as it
exists after the establishing operation and just after the contingency is
turned on. This is the time when the CV is zero (or at its lowest), and the
error between the CV and the reference setting is maximum, producing the
maximum rate of behavior. Reinforcers then begin to occur, the CV rises,
the error falls, and the behavior decreases. Are you still in agreement
with this?

Best,

Bill P.

18

[From Bill Powers (970830.1009 MDT)]

Bruce Abbott (970830.1020 EST)--

Typically the "establishing operation" is done before the contingency is
enabled, as when the rat's body weight is brought down to 80% of its ad
libitum weight. The rat is given no means of correcting the error between
CV and reference until it is placed in the chamber.

Right. This is the usual procedure, so let's stick with that. I'm asking if
we are agreed that when the contingency is first established, we have a
condition of minimum CV, maximum error, and maximum behavior, and that as
reinforcements are delivered, the CV increases, the error decreases, and
the behavior decreases. Is this correct?

Best,

Bill P.

19

[From Bruce Abbott (970830.1845 EST)]

Bill Powers (970830.1009 MDT)] --

Bruce Abbott (970830.1020 EST)

Typically the "establishing operation" is done before the contingency is
enabled, as when the rat's body weight is brought down to 80% of its ad
libitum weight. The rat is given no means of correcting the error between
CV and reference until it is placed in the chamber.

Right. This is the usual procedure, so let's stick with that. I'm asking if
we are agreed that when the contingency is first established, we have a
condition of minimum CV, maximum error, and maximum behavior, and that as
reinforcements are delivered, the CV increases, the error decreases, and
the behavior decreases. Is this correct?

Correct. I sense that you have something you wish to pursue here, so I'll
hold off elaborating further until I hear from you.

Regards,

Bruce

20

[From Bill Powers (970830.1812 MDT)]

Bruce Abbott (970830.1845 EST)]

I'm asking if
we are agreed that when the contingency is first established, we have a
condition of minimum CV, maximum error, and maximum behavior, and that as
reinforcements are delivered, the CV increases, the error decreases, and
the behavior decreases. Is this correct?

Correct. I sense that you have something you wish to pursue here, so I'll
hold off elaborating further until I hear from you.

Ok, so we're agreed on the PCT predictions. Now, what does reinforcement
theory say about the effect of reinforcements on behavior, under the same
circumstances?

Best,

Bill P.