Perception in PCT (was Re: When Control Systems Collide)

rsmarken · September 29, 2018, 5:26am

[Rick Marken 2018-09-28_22:26:02]

[Martin Taylor 2018.09.28.16.12]

  MT: Indeed, what we know of neurophysiology tells us that there is no
single place within the brain where that [perceptual] value exists for some
future technology to measure.

Â RM: The work of Hubel & Wiesel suggests that perceptions, in the form of neural firing rates, are located in afferent neurons and other single neurons in the brain.Â

  MT: The "perception" in Bill's model is
a rather vague average rate of firings of nerves in a “bundle”
with a fuzzy boundary.

RM: I don’t see what’s vague about it. In Bill’s model a perception – or, more properly, the state of a perceptual variable – is the average firing rate in an afferent nerve (a bundle of neurons) at any instant. The average firing rate is a variable called the perceptual signal, p, and it varies along with variations in in the aspect of the environment that is defined by the perceptual function that produces the perceptual signal as an output. This seems very precise to me.Â

RM: In PCT models, p is allowed to go negative, which, of course, neural firing rates can’t do. So in order to make our models true to the facts of physiology, every control system should really be two control systems controlling the same variable, one system acting when the perceptual signal goes above the reference signal and the other acting when the perceptual signal goes below the reference signal. I’ve attached a simple hierarchical model that implements the control systems in this way – the physiologically correct way. If you run this model note that the none of the signals in the model – all of which are presumed to be neural firing rates – go negative. But while this parallel loop model is truer to the physiology than is the more familiar single loop version where signals can go negative (as in the Live Block Diagram: https://www.dropbox.com/s/sizvbwso44mastu/LiveBlock.exe?dl=0 ) it’s a lot more tedious to create and gives the same result as the single loop model.Â

RM: But the fact that perception is modeled as a neural current in PCT makes it difficult to conceive of how control of complex perceptions, like sequences and programs – perceptions that are defined over a fairly long period of time – can work. That was why I posted my post entitled “Control of Higher Level Perceptions”.Â I think it’s important that we develop models that can control such perceptions to show how control of these higher level perceptions could work.Â

Â

BestÂ

Rick

2 level parallel control2.xlsx (18.9 KB)

···

–
Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery

MartinT · September 29, 2018, 3:39pm

[Martin Taylor 2018.09.29.10.47]

[Rick Marken 2018-09-28_22:26:02]

[Martin Taylor 2018.09.28.16.12]

              MT: Indeed, what we know of neurophysiology tells us
that there is no single place within the brain where
that [perceptual] value exists for some future
technology to measure.

          RM: The work of Hubel & Wiesel suggests that
perceptions, in the form of neural firing rates, are
located in afferent neurons and other single neurons in
the brain.

It's interesting that you should mention H&W, since their work

was published when I was in graduate school, transitioning from
being an engineer to a perception psychologist. When it was
reported, the effect on the group I associated with was that finally
the physiologists acknowledged that neurons could do
quasi-mathematical operations that psychologists believed they had
to be able to do. But the implications of their work was not as you
describe. It was mainly that they proved it was possible for neurons
to recode very large numbers of inputs into an informationally
efficient form that kept the important contingencies among the
inputs (important meaning frequently occurring) while losing the
contingencies that were probably (on statistical grounds) due to
irregular fluctuations in the visual field.

At that time, a colleague and I had been asked to produce a report

for a weekly seminar group on the Bush-Mosteller theory of memory.
We used a Hebbian approach, but I can’t remember whether that was
because B&M had done so or whether we just thought it was a
neato idea. Regardless, that was our expectation for the synaptic
variation in the inputs to visual neurons (and their distribution of
outputs to the main part of the brain in the visual cortex). This
was a good part of the reason the H&W findings came as a relief
rather than as a surprise.

The result of this was always that it demonstrated that the

perceptions we form depended on recoding the mix of
on-centre-off-surround, oriented edges, corners, and so on signals
that were reported along the optic nerve to the brain, and were
never located in any single place thereafter, any more than any
particular component of a Fourier Transform locates a value of a
signal at any one moment.

In other words, what you say is simply not true. Whether at some

future time some perception or other is found to be represented in
one and only one place in the brain, it is not true that the
findings of H&W suggest that it is.

              MT: The "perception" in Bill's model is a rather
vague average rate of firings of nerves in a “bundle”
with a fuzzy boundary.

          RM: I don't see what's vague about it. In Bill's model
a perception – or, more properly, the state of a
perceptual variable – is the average firing rate in an
afferent nerve (a bundle of neurons) at any instant.

Which does directly contradict your earlier assertion, does it not?

Incidentally, would you like to tell us what the average firing rate

in a nerve bundle is at an instant when none of the neurons in the
bundle are firing?

Would you like also to discuss how it is determined which neurons

are deemed to belong in any particular “bundle” and which do not,
for any particular perceptual signal? Remember, when you do this,
that most, if not all, effective neurons have thousands of synaptic
inputs, and probably no two have all their inputs coming from the
same set of source neurons, which means that each has a different
pattern of inputs that produces its maximum firing rate. Put another
way, the firing rate for any input pattern will produce a wide
distribution of firing rates across all the neurons that respond at
all to it, rather than there being a sharp cut-off between neurons
that do and neurons that don’t respond to that pattern.

          The average firing rate is a variable called the
perceptual signal, p, and it varies along with variations
in in the aspect of the environment that is defined by the
perceptual function that produces the perceptual signal as
an output.

Well, at least you wrote something on which we can agree.

This seems very precise to me.

But not to me.

          RM: In PCT models, p is allowed to go negative, which,
of course, neural firing rates can’t do. So in order to
make our models true to the facts of physiology, every
control system should really be two control systems
controlling the same variable, one system acting when the
perceptual signal goes above the reference signal and the
other acting when the perceptual signal goes below the
reference signal. I’ve attached a simple hierarchical
model that implements the control systems in this way –
the physiologically correct way. If you run this model
note that the none of the signals in the model – all of
which are presumed to be neural firing rates – go
negative. But while this parallel loop model is truer to
the physiology than is the more familiar single loop
version where signals can go negative (as in the Live
Block Diagram: https://www.dropbox.com/s/sizvbwso44mastu/LiveBlock.exe?dl=0
) it’s a lot more tedious to create and gives the same
result as the single loop model.

Well, in many respects it does, but there's more to it, as we

discussed a year or two ago. The opposition between the two
+positive and -positive sides of the balance introduces a new degree
of freedom because of the slowing effect of the leaky integrator
output. I chose to call this effect “stiffness” by analogy with the
case of two opposing control systems that allow a parade flag-bearer
to keep the flagstick vertical in a cross-wind by tightening her
grasp with each hand to increase the oppositional conflict-induced
gain of the virtual controller for deviations in that direction, and
at the same time in the orthogonal (stiffness) direction.

          RM: But the fact that perception is modeled as a neural
current in PCT makes it difficult to conceive of how
control of complex perceptions, like sequences and
programs – perceptions that are defined over a fairly
long period of time – can work.

Ah, yes. That's an issue I discussed with Bill at some length when I

first began to learn about PCT. I don’t remember ever becoming clear
about it at that time, though he may have been, but I think I am a
little clearer now, though not perfectly clear. Probably another
thread would b a better place to discuss it. Maybe we can get it
straight with the participation of the general CSGnet readership.

          That was why I posted my post entitled "Control of
Higher Level Perceptions". I think it’s important that we
develop models that can control such perceptions to show
how control of these higher level perceptions could work.

Yes.

Martin

rsmarken · September 29, 2018, 9:44pm

[Rick Marken 2018-09-29_14:43:56]

[Martin Taylor 2018.09.29.10.47]

          Â RM: The work of Hubel & Wiesel suggests that
perceptions, in the form of neural firing rates, are
located in afferent neurons and other single neurons in
the brain.

MT: It's interesting that you should mention H&W,...But the implications of their work was not as you
describe. It was mainly that they proved it was possible for neurons
to recode very large numbers of inputs into an informationally
efficient form…

RM: This may be the implication that an information theorist could get from the results of the Hubel & Weisel work. But I see their results as being consistent with the PCT assumption that neural firing rates are perceptual signals that represent the states of perceptual variables computed by perceptual functions that correspond to the neural networks that make up the neuron’s receptive field.

RM: Here’s an example of the kind of data obtained by Hubel & Weisel:

RM: As the orientation of the image of the bar (left panel) varies, neural firing rate (middle panel and right panel) varies. This result could be seen as showing the “selectivity” of the neuron to orientation, this neuron being selectively sensitive to a horizontal line. But it could also be seen as showing a neuron that perceives orientation; the firing rate of the neuron could be seen as a perceptual signal whose firing rate is a measure of the orientation of the line that falls on its receptive field; zero firing rate being vertical, moderate firing beingÂ a 45 degree angle relative to vertical and maximum firing being horizontal. So setting a reference of 0 for the firing rate of this neuron, a control system could keep the line oriented vertically, as you can do in my “Control of Perception” demo: https://www.mindreadings.com/ControlDemo/ControlOfPerception.html.

RM: So, again, it looks to me like the Hubel & Weisel results are quite consistent with the PCT model of perception and they locate perceptual signals at the level of single cells (neurons).Â

Best

Rick

···

At that time, a colleague and I had been asked to produce a report
for a weekly seminar group on the Bush-Mosteller theory of memory.
We used a Hebbian approach, but I can’t remember whether that was
because B&M had done so or whether we just thought it was a
neato idea. Regardless, that was our expectation for the synaptic
variation in the inputs to visual neurons (and their distribution of
outputs to the main part of the brain in the visual cortex). This
was a good part of the reason the H&W findings came as a relief
rather than as a surprise.
The result of this was always that it demonstrated that the
perceptions we form depended on recoding the mix of
on-centre-off-surround, oriented edges, corners, and so on signals
that were reported along the optic nerve to the brain, and were
never located in any single place thereafter, any more than any
particular component of a Fourier Transform locates a value of a
signal at any one moment.
In other words, what you say is simply not true. Whether at some
future time some perception or other is found to be represented in
one and only one place in the brain, it is not true that the
findings of H&W suggest that it is.
              MT: The "perception" in Bill's model is a rather
vague average rate of firings of nerves in a “bundle”
with a fuzzy boundary.

          RM: I don't see what's vague about it. In Bill's model
a perception – or, more properly, the state of a
perceptual variable – is the average firing rate in an
afferent nerve (a bundle of neurons) at any instant.

Which does directly contradict your earlier assertion, does it not?



Incidentally, would you like to tell us what the average firing rate
in a nerve bundle is at an instant when none of the neurons in the
bundle are firing?
Would you like also to discuss how it is determined which neurons
are deemed to belong in any particular “bundle” and which do not,
for any particular perceptual signal? Remember, when you do this,
that most, if not all, effective neurons have thousands of synaptic
inputs, and probably no two have all their inputs coming from the
same set of source neurons, which means that each has a different
pattern of inputs that produces its maximum firing rate. Put another
way, the firing rate for any input pattern will produce a wide
distribution of firing rates across all the neurons that respond at
all to it, rather than there being a sharp cut-off between neurons
that do and neurons that don’t respond to that pattern.
          The average firing rate is a variable called the
perceptual signal, p, and it varies along with variations
in in the aspect of the environment that is defined by the
perceptual function that produces the perceptual signal as
an output.

Well, at least you wrote something on which we can agree.
This seems very precise to me.

But not to me.

          RM: In PCT models, p is allowed to go negative, which,
of course, neural firing rates can’t do. So in order to
make our models true to the facts of physiology, every
control system should really be two control systems
controlling the same variable, one system acting when the
perceptual signal goes above the reference signal and the
other acting when the perceptual signal goes below the
reference signal. I’ve attached a simple hierarchical
model that implements the control systems in this way –
the physiologically correct way. If you run this model
note that the none of the signals in the model – all of
which are presumed to be neural firing rates – go
negative. But while this parallel loop model is truer to
the physiology than is the more familiar single loop
version where signals can go negative (as in the Live
Block Diagram: https://www.dropbox.com/s/sizvbwso44mastu/LiveBlock.exe?dl=0
) it’s a lot more tedious to create and gives the same
result as the single loop model.

Well, in many respects it does, but there's more to it, as we
discussed a year or two ago. The opposition between the two
+positive and -positive sides of the balance introduces a new degree
of freedom because of the slowing effect of the leaky integrator
output. I chose to call this effect “stiffness” by analogy with the
case of two opposing control systems that allow a parade flag-bearer
to keep the flagstick vertical in a cross-wind by tightening her
grasp with each hand to increase the oppositional conflict-induced
gain of the virtual controller for deviations in that direction, and
at the same time in the orthogonal (stiffness) direction.

          RM: But the fact that perception is modeled as a neural
current in PCT makes it difficult to conceive of how
control of complex perceptions, like sequences and
programs – perceptions that are defined over a fairly
long period of time – can work.

Ah, yes. That's an issue I discussed with Bill at some length when I
first began to learn about PCT. I don’t remember ever becoming clear
about it at that time, though he may have been, but I think I am a
little clearer now, though not perfectly clear. Probably another
thread would b a better place to discuss it. Maybe we can get it
straight with the participation of the general CSGnet readership.
          That was why I posted my post entitled "Control of
Higher Level Perceptions".Â I think it’s important that we
develop models that can control such perceptions to show
how control of these higher level perceptions could work.Â

Â

Yes.

Martin

–
Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery

_Boris_Hartman · October 2, 2018, 2:04pm

RM: In PCT models, p is allowed to go negative, which, of course, neural firing rates can’t do. So in order to make our models true to the facts of physiology, every control system should really be two control systems controlling the same variable, one system acting when the perceptual signal goes above the reference signal and the other acting when the perceptual signal goes below the reference signal. I’ve attached a simple hierarchical model that implements the control systems in this way – the physiologically correct way.

HB : It seems that I don’t understand what you meant by “two control systems controlling the same variable”. Obviously it’s not physiological correct way if you “located” perceptual signal at the level of single cell (neuron) ???

I think that comparator in PCT is well enough dealing with changes in potential of nerv cell to produce “error” signal. That’s the only important thing about “mismatch” in neuron, no matter how reference and perceptual nerv signal vary.

Bill P :

COMPARATOR : The portion of control system that computes the magnitude and direction of mismatch between perceptual and reference signal.

HB : “Mismatch” is quite well describing what happens in neuron and how “error” signal is produced. You don’t need two comparators for different reference or perceptual signal. Whatever you are proposing with 2 control units has nothing to do with physiology neither with PCT.

Can you show me some literature where you found physiological plausability with “two control systems controlling the same variable” ??? It can be PCT literature ?

I don’t understand how what you wrote above is connected to :

RM earlier : So, again, it looks to me like the Hubel & Weisel results are quite consistent with the PCT model of perception and they locate perceptual signals at the level of single cells (neurons).

Boris

···

From: Richard Marken (rsmarken@gmail.com via csgnet Mailing List) csgnet@lists.illinois.edu
Sent: Saturday, September 29, 2018 7:26 AM
To: csgnet csgnet@lists.illinois.edu
Cc: Richard Marken rsmarken@gmail.com
Subject: Perception in PCT (was Re: When Control Systems Collide)

[Rick Marken 2018-09-28_22:26:02]

[Martin Taylor 2018.09.28.16.12]

MT: Indeed, what we know of neurophysiology tells us that there is no single place within the brain where that [perceptual] value exists for some future technology to measure.

RM: The work of Hubel & Wiesel suggests that perceptions, in the form of neural firing rates, are located in afferent neurons and other single neurons in the brain.

MT: The “perception” in Bill’s model is a rather vague average rate of firings of nerves in a “bundle” with a fuzzy boundary.

RM: I don’t see what’s vague about it. In Bill’s model a perception – or, more properly, the state of a perceptual variable – is the average firing rate in an afferent nerve (a bundle of neurons) at any instant. The average firing rate is a variable called the perceptual signal, p, and it varies along with variations in in the aspect of the environment that is defined by the perceptual function that produces the perceptual signal as an output. This seems very precise to me.

RM: In PCT models, p is allowed to go negative, which, of course, neural firing rates can’t do. So in order to make our models true to the facts of physiology, every control system should really be two control systems controlling the same variable, one system acting when the perceptual signal goes above the reference signal and the other acting when the perceptual signal goes below the reference signal. I’ve attached a simple hierarchical model that implements the control systems in this way – the physiologically correct way.

HB : It’s good attempt Rick, but can you show me literature where you found physiological plausability with “two control systems controlling the same variable” ???

RM : If you run this model note that the none of the signals in the model – all of which are presumed to be neural firing rates – go negative. But while this parallel loop model is truer to the physiology than is the more familiar single loop version where signals can go negative (as in the Live Block Diagram: https://www.dropbox.com/s/sizvbwso44mastu/LiveBlock.exe?dl=0 ) it’s a lot more tedious to create and gives the same result as the single loop model.

RM: But the fact that perception is modeled as a neural current in PCT makes it difficult to conceive of how control of complex perceptions, like sequences and programs – perceptions that are defined over a fairly long period of time – can work. That was why I posted my post entitled “Control of Higher Level Perceptions”. I think it’s important that we develop models that can control such perceptions to show how control of these higher level perceptions could work.

Best

Rick

–

Richard S. Marken

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.”
–Antoine de Saint-Exupery

FredNickols · October 2, 2018, 2:09pm

From Fred Nickols (2018.10.02.1007 ET)

I’m no modeler and everyone know that but, with respect to Rick’s closing paragraph about control of higher level perceptions, a term came to mind "incremental progress."Â A program can be parsed into steps, phases and other increments bearing other labels.Â Presumably, the program itself isn’t controlled except by way of controlling its component parts.

So, to control higher level perceptions, we have to control the relevant (i.e., contributing) lower level ones.Â Isn’t that consistent with what Bill was getting at with his hierarchy?

···

Regards,

Fred Nickols

Managing Partner

Distance Consulting LLC

“Assistance at A Distance”

MartinT · October 2, 2018, 3:20pm

[Martin Taylor 2018.10.02.10.02]

[Rick Marken 2018-09-29_14:43:56]

[Martin Taylor 2018.09.29.10.47]

                      RM: The work of Hubel & Wiesel
suggests that perceptions, in the form of
neural firing rates, are located in afferent
neurons and other single neurons in the brain.

            MT: It's interesting that you should mention
H&W,…But the implications of their work was not as
you describe. It was mainly that they proved it was
possible for neurons to recode very large numbers of
inputs into an informationally efficient form…

          RM: This may be the implication that an information
theorist could get from the results of the Hubel &
Weisel work. But I see their results as being consistent
with the PCT assumption that neural firing rates are
perceptual signals that represent the states of perceptual
variables computed by perceptual functions that correspond
to the neural networks that make up the neuron’s receptive
field.

Sure. What does this have to do with my comment?

Or with anything I wrote there or in the postscript to my main

message, that the lowest level of the Powers hierarchy “Intensity”
is produced from a multitude of inputs of the many types discovered
by H&W and their successors, some of which involve velocity? Or
with the point that a “neural bundle” central to the concept of a
“neural current” (a.k.a. “perceptual signal”) consists of neurons
with only a fuzzy membership value in the bundle, since no two
neurons are likely to have exactly the same pattern of synaptic
inputs and weights?

If you were to argue that the bottom level of the control hierarchy

actually consisted of control of whether a particular spot on the
retina reports a bright spot with a darker surround, or a line
moving northwest, I would have no complaint, other than that it
would seem rather implausible, given the enormous ratio between the
degrees of freedom per second for variation at that level and the
degrees of freedom per second for our ability to influence the
environment. Neither would I have an argument if you were to then
say that control of those variables was internal to the retina, or
to some internal feedback loop. I don’t know enough of the neural
structure to know whether such control loops exist in the retina or
whether the optic nerve might carry enough reference values to the
retina to allow them to function.

···

[Rick Marken 2018-09-29_13:35:13]

[Martin Taylor 2018.09.29.12.01]

          MT: Bottom line: If we are going to
try to figure out working models for, say, program
control, it is probably worth reconsidering the actual
levels in the control hierarchy.

    RM: The exercise I am proposing requires no assumptions about
the hierarchy at all. What I would like to see is a control
model that can do what a person can do in my “Program Control”
demo (https://www.mindreadings.com/ControlDemo/ProgramControl.html ),
which is to control a program, such as the one in the demo: “if the shape is circle, the next
color is blue; else, the next color is red”.

Best

Bruce Nevin has made one answer to this, but I don't think either

that or Marken’s comment have any relationship to the point to which
Marken was responding. In my agreement with [Rick Marken
2018-09-28_22:26:02]:

            RM: But the fact that perception is modeled as a
neural current in PCT makes it difficult to conceive of
how control of complex perceptions, like sequences and
programs – perceptions that are defined over a fairly
long period of time – can work.

  Ah, yes. That's an issue I discussed with Bill at some length when
I first began to learn about PCT. I don’t remember ever becoming
clear about it at that time, though he may have been, but I think
I am a little clearer now, though not perfectly clear. Probably
another thread would b a better place to discuss it. Maybe we can
get it straight with the participation of the general CSGnet
readership.
            That was why I posted my post entitled "Control of
Higher Level Perceptions". I think it’s important that
we develop models that can control such perceptions to
show how control of these higher level perceptions could
work.

Yes.

I simply agreed with Rick that in order to make a working model of,

say, program control, we need to know how the variables that are
inputs to its perceptual signals, that are extended over time
(Rick’s original point) are produced. Rick’s response that we do in
fact control perceptions, whether correctly demonstrated or not, is
totally irrelevant.

... other than as a demonstration that if I agree with Rick on

something, he must find a way to make it appear that he no longer
agrees with what he said before.

Martin

rsmarken · October 2, 2018, 8:10pm

[Rick Marken 2018-10-02_13:09:37]

From Fred Nickols (2018.10.02.1007 ET)

FN: I’m no modeler and everyone know that but, with respect to Rick’s closing paragraph about control of higher level perceptions, a term came to mind "incremental progress."Â A program can be parsed into steps, phases and other increments bearing other labels.Â Presumably, the program itself isn’t controlled except by way of controlling its component parts.

RM: Yes, that’s generally true in real life; when carrying out a program, like a recipe, we have to control the components of the program (the preparation of the ingredients) in order to keep the program under control. But in my demo the whole program perception can be kept under control with just a bar press.Â

BP: So, to control higher level perceptions, we have to control the relevant (i.e., contributing) lower level ones.Â Isn’t that consistent with what Bill was getting at with his hierarchy?

RM: Again, yes. But my demo is constructed so that higher level perceptions (a sequence and a program) can be controlled without controlling the relevant lower level perception – shapes, sizes and colors. The sequence and program perceptions themselves are presumably constructed from these lower level perceptions but, again, the demo has been set up so that you don’t have to control the lower level perceptions in order to control the higher level one’s.Â

RM: Try it and see. https://www.mindreadings.com/ControlDemo/ProgramControl.html

BestÂ

Rick

···

Regards,

Fred Nickols

Managing Partner

Distance Consulting LLC

“Assistance at A Distance”

On Tue, Oct 2, 2018 at 10:04 AM “Boris Hartman” csgnet@lists.illinois.edu wrote:

RM: In PCT models, p is allowed to go negative, which, of course, neural firing rates can’t do. So in order to make our models true to the facts of physiology, every control system should really be two control systems controlling the same variable, one system acting when the perceptual signal goes above the reference signal and the other acting when the perceptual signal goes below the reference signal. I’ve attached a simple hierarchical model that implements the control systems in this way – the physiologically correct way.

Â

HB : It seems that I don’t understand what you meant by “two control systems controlling the same variable”. Obviously it’s not physiological correct way if you “located” perceptual signal at the level of single cell (neuron) ???

Â

I think that comparator in PCT is well enough dealing with changes in potential of nerv cell to produce “error” signal. That’s the only important thing about “mismatch” in neuron, no matter how reference and perceptual nerv signal vary.

Â

Bill P :

COMPARATOR : The portion of control system that computes the magnitude and direction of mismatch between perceptual and reference signal.

Â

HB : “Mismatch” is quite well describing what happens in neuron and how “error” signal is produced. You don’t need two comparators for different reference or perceptual signal. Whatever you are proposing with 2 control units has nothing to do with physiology neither with PCT.

Â

Can you show me some literature where you found physiological plausability with “two control systems controlling the same variable” ??? It can be PCT literature ?

Â

I don’t understand how what you wrote above is connected to :

Â

RM earlier : So, again, it looks to me like the Hubel & Weisel results are quite consistent with the PCT model of perception and they locate perceptual signals at the level of single cells (neurons).Â

Â

Boris

Â

From: Richard Marken (rsmarken@gmail.com via csgnet Mailing List) csgnet@lists.illinois.edu
Sent: Saturday, September 29, 2018 7:26 AM
To: csgnet csgnet@lists.illinois.edu
Cc: Richard Marken rsmarken@gmail.com
Subject: Perception in PCT (was Re: When Control Systems Collide)

Â

[Rick Marken 2018-09-28_22:26:02]

[Martin Taylor 2018.09.28.16.12]

MT: Indeed, what we know of neurophysiology tells us that there is no single place within the brain where that [perceptual] value exists for some future technology to measure.

Â RM: The work of Hubel & Wiesel suggests that perceptions, in the form of neural firing rates, are located in afferent neurons and other single neurons in the brain.Â

MT: The “perception” in Bill’s model is a rather vague average rate of firings of nerves in a “bundle” with a fuzzy boundary.

RM: I don’t see what’s vague about it. In Bill’s model a perception – or, more properly, the state of a perceptual variable – is the average firing rate in an afferent nerve (a bundle of neurons) at any instant. The average firing rate is a variable called the perceptual signal, p, and it varies along with variations in in the aspect of the environment that is defined by the perceptual function that produces the perceptual signal as an output. This seems very precise to me.Â

Â

RM: In PCT models, p is allowed to go negative, which, of course, neural firing rates can’t do. So in order to make our models true to the facts of physiology, every control system should really be two control systems controlling the same variable, one system acting when the perceptual signal goes above the reference signal and the other acting when the perceptual signal goes below the reference signal. I’ve attached a simple hierarchical model that implements the control systems in this way – the physiologically correct way.

Â

HB : It’s good attempt Rick, but can you show me literature where you found physiological plausability with “two control systems controlling the same variable” ???

Â

RM : If you run this model note that the none of the signals in the model – all of which are presumed to be neural firing rates – go negative. But while this parallel loop model is truer to the physiology than is the more familiar single loop version where signals can go negative (as in the Live Block Diagram: https://www.dropbox.com/s/sizvbwso44mastu/LiveBlock.exe?dl=0 ) it’s a lot more tedious to create and gives the same result as the single loop model.Â

Â

RM: But the fact that perception is modeled as a neural current in PCT makes it difficult to conceive of how control of complex perceptions, like sequences and programs – perceptions that are defined over a fairly long period of time – can work. That was why I posted my post entitled “Control of Higher Level Perceptions”.Â I think it’s important that we develop models that can control such perceptions to show how control of these higher level perceptions could work.Â

Â

BestÂ

Â

Rick

Â

–

Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery

–
Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery

FredNickols · October 2, 2018, 8:28pm

From Fred Nickols (2018.10.02.1623 ET)

Rick: Â I have no doubt that what you say about your model is true. However, if true, it also seems to me that what youâ€™ve done is disproved a basic tenet of Billâ€™s theory about the hierarchy, namely, that higher level perceptions are controlled based on feedback from lower level systems.Â If thatâ€™s the case, then there is no HPCT.

···

Fred Nickols
Distance Consulting LLC
â€œAssistance at A Distanceâ€?
www.nickols.us

MartinT · October 2, 2018, 8:45pm

[Martin Taylor 2018.10.02.16.32]

From Fred Nickols (2018.10.02.1623 ET)

    Rick: Â I have no doubt that what you say about
your model is true. However, if true, it also seems to me that
what youâ€™ve done is disproved a basic tenet of Billâ€™s theory
about the hierarchy, namely, that higher level perceptions are
controlled based on feedback from lower level systems.Â If
thatâ€™s the case, then there is no HPCT.

I have to agree with Rick here. His demo neither proves nor

disproves anything about HPCT, other than that it is possible to
control the perceived sequence by hitting the space bar. What is
necessary for the demo is not that the control at lower levels must
be understood, or that we can determine how they are controlled (or
if they are) but that the controller has available some action that
allows the controlled variable to be changed toward its reference
value. So long as that happens, the demo works. Bruce Nevin thinks
the controlled variable isn’t actually a program, but it doesn’t
matter to the principle that what matters is that some available
action can influence the controlled perception in such a way as to
maintain its similarity to its reference value.

My earlier comment, which led Rick to refer to the demo, was about

Rick’s statement that he thought we should try to develop working
models for the higher levels of control, such as program control. If
Rick thinks that his demo is such a working model, he has a
different definition of “working model” than I do. My definition
would include questions of how the intervening level perceptions are
generated and controlled, as you suggest.

Martin

···

Fred Nickols

    Distance Consulting LLC

    â€œAssistance at A Distanceâ€?

    [www.nickols.us](http://www.nickols.us)

rsmarken · October 3, 2018, 1:00am

[Rick Marken 2018-10-02_18:00:22]

[Martin Taylor 2018.10.02.10.02]

            MT: It's interesting that you should mention
H&W,…But the implications of their work was not as
you describe. It was mainly that they proved it was
possible for neurons to recode very large numbers of
inputs into an informationally efficient form…

          RM: This may be the implication that an information
theorist could get from the results of the Hubel &
Weisel work. But I see their results as being consistent
with the PCT assumption that neural firing rates are
perceptual signals that represent the states of perceptual
variables computed by perceptual functions that correspond
to the neural networks that make up the neuron’s receptive
field.

MT: Sure. What does this have to do with my comment?

RM: It’s relevant to your comment regarding the implication you draw from the Hubel & Weisel work. My comment points out that the implication you draw – thatÂ

Hubel & WeiselÂ proved it was possible for neurons to recode very large numbers of inputs into an informationally efficient form – is not an implication that is relevant to the possible neurological basis of perception in the PCT model. In PCT neurons don’t “recode” (or “code”) inputs.; they simply carry neural impulses that vary in rate depending on the nature of the stimulus or sensory input to perceptual functions. So according to PCT, variations in neural firing rate are an analogÂ of variations in the aspect of the environment that is perceived by the perceptual function.Â

RM:Â Hubel & Weisel found that the rate of firing of a single neuron varies depending on variations in the aspect of the environment that is “detected” by the neuron’s receptive field. But these results can be seen to be consistent with the PCT view of perception, as shown in the little figure I posted, and post again:

RM: Here you can see that the neuron’s rate of firing (center panel) can be seen to be an analog of the angle of the bar presented to the receptive field of the neuron (left panel). So the receptive field can be consideredÂ equivalent to the PCT perceptual function and the rate of firing can be considered equivalent to the PCT perceptual signal.Â

BestÂ

Rick

···

Or with anything I wrote there or in the postscript to my main
message, that the lowest level of the Powers hierarchy “Intensity”
is produced from a multitude of inputs of the many types discovered
by H&W and their successors, some of which involve velocity? Or
with the point that a “neural bundle” central to the concept of a
“neural current” (a.k.a. “perceptual signal”) consists of neurons
with only a fuzzy membership value in the bundle, since no two
neurons are likely to have exactly the same pattern of synaptic
inputs and weights?
If you were to argue that the bottom level of the control hierarchy
actually consisted of control of whether a particular spot on the
retina reports a bright spot with a darker surround, or a line
moving northwest, I would have no complaint, other than that it
would seem rather implausible, given the enormous ratio between the
degrees of freedom per second for variation at that level and the
degrees of freedom per second for our ability to influence the
environment. Neither would I have an argument if you were to then
say that control of those variables was internal to the retina, or
to some internal feedback loop. I don’t know enough of the neural
structure to know whether such control loops exist in the retina or
whether the optic nerve might carry enough reference values to the
retina to allow them to function.
------------
[Rick Marken 2018-09-29_13:35:13]

[Martin Taylor 2018.09.29.12.01]

          MT: Bottom line: If we are going to
try to figure out working models for, say, program
control, it is probably worth reconsidering the actual
levels in the control hierarchy.

    RM: The exercise I am proposing requires no assumptions about
the hierarchy at all.Â What I would like to see is a control
model that can do what a person can do in my “Program Control”
demo (https://www.mindreadings.com/ControlDemo/ProgramControl.html ),
which is to control a program, such as the one inÂ the demo:Â “if the shape is circle, the next
color is blue; else, the next color is red”.Â

Best

Bruce Nevin has made one answer to this, but I don't think either
that or Marken’s comment have any relationship to the point to which
Marken was responding. In my agreement with [Rick Marken
2018-09-28_22:26:02]:

            RM: But the fact that perception is modeled as a
neural current in PCT makes it difficult to conceive of
how control of complex perceptions, like sequences and
programs – perceptions that are defined over a fairly
long period of time – can work.

  Ah, yes. That's an issue I discussed with Bill at some length when
I first began to learn about PCT. I don’t remember ever becoming
clear about it at that time, though he may have been, but I think
I am a little clearer now, though not perfectly clear. Probably
another thread would b a better place to discuss it. Maybe we can
get it straight with the participation of the general CSGnet
readership.
            That was why I posted my post entitled "Control of
Higher Level Perceptions".Â I think it’s important that
we develop models that can control such perceptions to
show how control of these higher level perceptions could
work.Â

Â

Yes.
I simply agreed with Rick that in order to make a working model of,
say, program control, we need to know how the variables that are
inputs to its perceptual signals, that are extended over time
(Rick’s original point) are produced. Rick’s response that we do in
fact control perceptions, whether correctly demonstrated or not, is
totally irrelevant.
... other than as a demonstration that if I agree with Rick on
something, he must find a way to make it appear that he no longer
agrees with what he said before.
Martin

–
Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery

rsmarken · October 3, 2018, 2:16am

[Rick Marken 2018-10-02_19:16:48]

From Fred Nickols (2018.10.02.1623 ET)

FN: Rick: Â I have no doubt that what you say about your model is true. However, if true, it also seems to me that what youâ€™ve done is disproved a basic tenet of Billâ€™s theory about the hierarchy, namely, that higher level perceptions are controlled based on feedback from lower level systems.Â If thatâ€™s the case, then there is no HPCT.

RM: It’s not my “model” that’s true; it’s the reality of the situation. The reality is that a bar press can be used to keep a particular sequence (or program) occurring; that is, a simple bar press can be used to control a sequence or program. I set the situation up this way because I am measuring the ability to control different kinds of perceptions depending on the rate at which the components of these perceptions occur. The idea was that, according to HPCT, it should be possible to control lower level perceptions when their components occur at a fast rate and to control higher level perceptions only when their components occur at a slower rate. This is because the control loop is presumably longer for controlling higher level as compared to lower level perceptions. Actually, the rational for this approach to studying the control hierarchy is described inÂ Marken, R. S., Khatib, Z. and Â Mansell, W. (2013) Motor Control as the
Control of Perception, Perceptual and
Motor Skills, 117, 236-247, which is attached.Â

RM: My “Hierarchical behavior of perception” Demo (https://www.mindreadings.com/ControlDemo/Hierarchy.html) shows that this is true for configurations, transitions and sequences: configurations (low in the hierarchy) can be controlled at a faster rate than transitions (a bit higher up) and transitions can be controlled at aÂ higher rate than sequences (the highest up of the group). My “Program control” demo (https://www.mindreadings.com/ControlDemo/ProgramControl.html) just takes it one more level up and shows that sequences, which are presumed to be lower in the hierarchy than programs, can be controlled at a faster rate than programs.Â

RM: So my demo doesn’t disprove anything about HPCT. In fact, it provides evidence in favor of the hierarchical model. The results are consistent with much of what Bill proposed as the hierarchy of perceptual variables; configuration perceptions are lower in the control hierarchy than transition perceptions which are lower than sequence perceptions which are lower than program perceptions.

RM: The fact that these different types of perceptions can all be controlled with a simple bar press doesn’t invalidate HPCT. There is nothing in PCT that says that control of a higher level perception must be based on control all the lower level perceptions that are components of that perception. The model just says that we control perceptions using whatever outputs are available to us that affect the state of that perception. I set up my “Hierarchy” demos so that all the different perceptions can be controlled by the same output so that any differences in the rate at which different perceptions could be controlled could not be attributed to the possible “confounding” effects on the timing of the need to control other lower level variables. Any difference in the rate at which the different perceptions could be controlled could be due only to the difference in the level in the nervous system at which the different perceptions are controlled.Â

Best regards

Rick

Hierarchical Control1.pdf (262 KB)

···

–
Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery

rsmarken · October 3, 2018, 2:24am

[Rick Marken 2018-10-02_19:23:56]

[Martin Taylor 2018.10.02.16.32]

    FN: Rick: Â I have no doubt that what you say about
your model is true. However, if true, it also seems to me that
what youâ€™ve done is disproved a basic tenet of Billâ€™s theory
about the hierarchy, namely, that higher level perceptions are
controlled based on feedback from lower level systems.Â If
thatâ€™s the case, then there is no HPCT.

MT: I have to agree with Rick here. His demo neither proves nor
disproves anything about HPCT, other than that it is possible to
control the perceived sequence by hitting the space bar.

RM: Thanks for the agreement. But, while it’s true that my demo neither proves nor disproves anything about HPCT (proof is a mathematical, not a scientific, concept), it certainly provides strong evidence for the some of the types of controlled perceptual variables and the hierarchical arrangement between them that was hypothesized by Bill as the HPCT model.Â

Best

Rick

Â

···

Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery

MartinT · October 3, 2018, 2:25pm

[Martin Taylor 2018.10.03.10.24]

[Rick Marken 2018-10-02_19:23:56]

[Martin Taylor 2018.10.02.16.32]

              FN: Rick:  I have no doubt that what you
say about your model is true. However, if true, it
also seems to me that what you’ve done is disproved a
basic tenet of Bill’s theory about the hierarchy,
namely, that higher level perceptions are controlled
based on feedback from lower level systems. If that’s
the case, then there is no HPCT.

          MT: I have to agree with Rick here. His demo neither
proves nor disproves anything about HPCT, other than that
it is possible to control the perceived sequence by
hitting the space bar.

        RM: Thanks for the agreement. But, while it's true that
my demo neither proves nor disproves anything about HPCT
(proof is a mathematical, not a scientific, concept), it
certainly provides strong evidence for the some of the types
of controlled perceptual variables and the hierarchical
arrangement between them that was hypothesized by Bill as
the HPCT model.

Again I agree!

Martin

MartinT · October 3, 2018, 2:46pm

[Martin Taylor 2018.10.03.10.25]

[Rick Marken 2018-10-02_18:00:22]

[Martin Taylor 2018.10.02.10.02]

                      MT: It's interesting
that you should mention H&W,…But the
implications of their work was not as you
describe. It was mainly that they proved it
was possible for neurons to recode very large
numbers of inputs into an informationally
efficient form…

                    RM: This may be the implication that an
information theorist could get from the results
of the Hubel & Weisel work. But I see their
results as being consistent with the PCT
assumption that neural firing rates are
perceptual signals that represent the states of
perceptual variables computed by perceptual
functions that correspond to the neural networks
that make up the neuron’s receptive field.

MT: Sure. What does this have to do with my comment?

        RM: It's relevant to your comment regarding the
implication you draw from the Hubel & Weisel work.

Which was that if the basic elements of vision provided to the

visual system consist of configurations and events, then
“intensity”, which is taken to be the base of the Powers set of
eleven levels of perception, must itself be an inferred or computed
variable in the same way as are the higher-level types of
perceptions.

Your comments are entirely irrelevant to this point.

My comment points out that the implication you draw – that
Hubel & Weisel proved it was
possible for neurons to recode very large numbers of
inputs into an informationally efficient form – is not an
implication that is relevant to the possible neurological
basis of perception in the PCT model.

Agreed.

          In PCT neurons don't "recode" (or
“code”) inputs.; they simply carry neural impulses that
vary in rate depending on the nature of the stimulus or
sensory input to perceptual functions. So according to
PCT, variations in neural firing rate are an analog of
variations in the aspect of the environment that is
perceived by the perceptual function.

Exactly. They encode the variations in the aspect of the environment

that they do. They implement the perceptual function, but they do
not individually produce a “neural current”. It takes a bundle of
them to do that, and each member of the bundle has its own
idiosyncratic set of synaptic connections that determine precisely
which perceptual function it implements. The neural current in the
bundle is the result of putting together all these slightly
different “per neuron” perceptual functions.

The fact that you have a personal antipathy for certain mathematical

ways of analyzing the world does not invalidate them. In order to
avoid using the words, you use long-winded circumlocutions to say
what the words say more succinctly. " variations
in neural firing rate are an analog of variations in the aspect of
the environment" is said more easily as “neural firing rates
encode aspects of the environment”. It’s the same thing. You are
happy with “variance”, but not with its generalization
“uncertainty”, which, if a distribution is Gaussian, is directly
proportional to variance with a proportionality factor of 0.5log₂2πe. Your unhappiness
with the words or the concepts does not affect their usefulness.

RM: Hubel & Weisel found
that the rate of firing of a single neuron varies depending
on variations in the aspect of the environment that is
“detected” by the neuron’s receptive field. But these
results can be seen to be consistent with the PCT view of
perception, as shown in the little figure I posted, and post
again:

Agreed.

        RM: Here you can see that the neuron's rate of firing
(center panel) can be seen to be an analog of the angle of
the bar presented to the receptive field of the neuron (left
panel). So the receptive field can be considered equivalent
to the PCT perceptual function and the rate of firing can be
considered equivalent to the PCT perceptual signal.

Not agreed. The "PCT perceptual signal" is always taken to be a

neural current. However, as Bill pointed out early in B:CP, he
didn’t expect simulations based on neural currents to be as accurate
as they have turned out to be. We always must distinguish between
the “perceptual signal” that is the firing rate of a single neuron,
and the “PCT perceptual signal”, which exchanges spatial averaging
over many neurons for the time averaging that is necessary to
determine the recent firing rate of a neuron.

Martin

rsmarken · October 6, 2018, 8:55pm

[Rick Marken 2018-10-06_13:54:52]

[Martin Taylor 2018.10.03.10.25]

                      MT: It's interesting
that you should mention H&W,…But the
implications of their work was not as you
describe. It was mainly that they proved it
was possible for neurons to recode very large
numbers of inputs into an informationally
efficient form…

                    RM: This may be the implication that an
information theorist could get from the results
of the Hubel & Weisel work. But I see their
results as being consistent with the PCT
assumption that neural firing rates are
perceptual signals that represent the states of
perceptual variables computed by perceptual
functions that correspond to the neural networks
that make up the neuron’s receptive field.

MT: Sure. What does this have to do with my comment?

        RM: It's relevant to your comment regarding the
implication you draw from the Hubel & Weisel work.

MT: Which was that if the basic elements of vision provided to the
visual system consist of configurations and events, then
“intensity”, which is taken to be the base of the Powers set of
eleven levels of perception, must itself be an inferred or computed
variable in the same way as are the higher-level types of
perceptions.

RM: Sorry about the delay in the response but I spent the last few days trying, with no success, to figure out what this means.Â So when you say:

MT: Your comments are entirely irrelevant to this point.

RM I guess I have to agree.Â

Best

RickÂ

Â

···

My comment points out that the implication you draw – thatÂ
Hubel & WeiselÂ proved it was
possible for neurons to recode very large numbers of
inputs into an informationally efficient form – is not an
implication that is relevant to the possible neurological
basis of perception in the PCT model.

Agreed.
          In PCT neurons don't "recode" (or
“code”) inputs.; they simply carry neural impulses that
vary in rate depending on the nature of the stimulus or
sensory input to perceptual functions. So according to
PCT, variations in neural firing rate are an analogÂ of
variations in the aspect of the environment that is
perceived by the perceptual function.

Exactly. They encode the variations in the aspect of the environment
that they do. They implement the perceptual function, but they do
not individually produce a “neural current”. It takes a bundle of
them to do that, and each member of the bundle has its own
idiosyncratic set of synaptic connections that determine precisely
which perceptual function it implements. The neural current in the
bundle is the result of putting together all these slightly
different “per neuron” perceptual functions.
The fact that you have a personal antipathy for certain mathematical
ways of analyzing the world does not invalidate them. In order to
avoid using the words, you use long-winded circumlocutions to say
what the words say more succinctly. " variations
in neural firing rate are an analogÂ of variations in the aspect of
the environment" is said more easily as “neural firing rates
encode aspects of the environment”. It’s the same thing. You are
happy with “variance”, but not with its generalization
“uncertainty”, which, if a distribution is Gaussian, is directly
proportional to variance with a proportionality factor of 0.5log₂2Ï€e. Your unhappiness
with the words or the concepts does not affect their usefulness.

RM:Â Hubel & Weisel found
that the rate of firing of a single neuron varies depending
on variations in the aspect of the environment that is
“detected” by the neuron’s receptive field. But these
results can be seen to be consistent with the PCT view of
perception, as shown in the little figure I posted, and post
again:

Agreed.
        RM: Here you can see that the neuron's rate of firing
(center panel) can be seen to be an analog of the angle of
the bar presented to the receptive field of the neuron (left
panel). So the receptive field can be consideredÂ equivalent
to the PCT perceptual function and the rate of firing can be
considered equivalent to the PCT perceptual signal.

Not agreed. The "PCT perceptual signal" is always taken to be a
neural current. However, as Bill pointed out early in B:CP, he
didn’t expect simulations based on neural currents to be as accurate
as they have turned out to be. We always must distinguish between
the “perceptual signal” that is the firing rate of a single neuron,
and the “PCT perceptual signal”, which exchanges spatial averaging
over many neurons for the time averaging that is necessary to
determine the recent firing rate of a neuron.
Martin

–
Richard S. MarkenÂ

"Perfection is achieved not when you have nothing more to add, but when you
have nothing left to take away.â€?
Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â --Antoine de Saint-Exupery