Anticipatory Postural Control

rsmarken · May 9, 2019, 1:00am

[Rick Marken 2019-05-08_17:57:30]

RM: Some time ago Heather Broccard-Bell asked me what the PCT explanation might be of a behavior that looks like a pretty robust example of anticipatory (or feedforward) control. Here’s how Heather described it to me:

HB-B: I was reading about this finding (and perhaps you are already familiar) that, if if you measure muscle activation during behavior, you often see compensatory activation of the axial musculature well before you get activation in the muscles that direct the part of the body that does the thing you’re focused on.Â For example, they trained cats to bat at a target with their forepaw.Â **Well before the limb/forepaw muscles are active, there is activation in the trunk musculature, seemingly anticipating the requisite compensatory stabilization that will become necessary when the forepaw is lifted.Â ** I am wondering what a PCT perspective might be on that?Â What is the relevant controlled variable, since the thing that actually changes sensory input that should activate control architecture hasn’t yet happened?Â Am I missing something?Â It appears to be a rather robust finding.[emphasis mine–RM]

RM: This finding certainly looks like it is inconsistent with the predictions of PCT since it looks like the cat is generating an output (activation of the trunk musculature) open loop in anticipation of a future disturbance (the change in orientation of the body when the forepaw is lifted, a change in orientation that would cause the cat to lose its balance if the trunk were not in the proper position).Â

RM: Apparently, quite a lot of research has been (and is still being) done on this “anticipatory postural control” phenomenon. Here is a paper by Aruin and Latash that Heather sent on the finding of "anticipatory postural control"Â when people drop a weight:

Â https://www.dropbox.com/s/hat2ua3ah7rqs79/aruin1995.pdf?dl=0

RM: Here there is again clear evidence that the person takes compensatory action (contracts muscles) in anticipation of a future disturbance that would cause a loss of balance if the anticipatory compensation had not occurred.Â

RM: Heather also described this related neurophysiological observation:

HB-B: **Related, it was discovered that if you knock out the indirect pathway from motor cortex to spinal cord (through the reticular formation), but leave the direct pathway intact, you abolish the “prestabilizing”, but still get the paw bat.Â ** I think you could explain that, in a super-oversimplified way, in terms of those pathways being about different control systems (these bastards have ascending, descending, and reciprocal connections at every level, not to mention all kinds of built-in sensory elements and sensory inputs connecting from elsewhere, as any good ultracomplex biological control system would).

RM: I am posting this to you experts on PCT (with Heather’s permission) because I think this is a very interesting discovery that seems to pose a challenge to PCT and I would like to see what you think about this. I have an idea about how PCT would explain this “anticipatory postural control” phenomenon and I think theÂ Â Aruin and Latash paper pointed to above suggests a way to test it. But before I give my PCT explanation I’d like to see what you think.Â

BestÂ

Rick

···

–
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 · May 9, 2019, 4:16am

[Martin Taylor 2019.05.09.00.09]

[Rick Marken 2019-05-08_17:57:30]

          RM: Some time ago Heather Broccard-Bell asked me what
the PCT explanation might be of a behavior that looks like
a pretty robust example of anticipatory (or feedforward)
control. Here’s how Heather described it to me:

            HB-B: I was reading
about this finding (and perhaps you are already
familiar) that, if if you measure muscle activation
during behavior, you often see compensatory activation
of the axial musculature well before you get activation
in the muscles that direct the part of the body that
does the thing you’re focused on. For example, they
trained cats to bat at a target with their forepaw. ** Well
before the limb/forepaw muscles are active, there is
activation in the trunk musculature, seemingly
anticipating the requisite compensatory stabilization
that will become necessary when the forepaw is
lifted. ** I am wondering what a PCT perspective
might be on that? What is the relevant controlled
variable, since the thing that actually changes sensory
input that should activate control architecture hasn’t
yet happened? Am I missing something? It appears to be
a rather robust finding.[emphasis mine–RM]

…

          RM: I am posting this to you experts on PCT
(with Heather’s permission) because I think this is a very
interesting discovery that seems to pose a challenge to
PCT and I would like to see what you think about this.

Here is  section of what I wrote to Heather on this question April

···

[Here is a] quick and dirty PCT approach. I start with the intention

to bat at the target, which would set a reference for a sequence
that could crudely be taken as “store energy, set trigger, wait,
release energy when required”, or in a more general sense “Prepare,
wait, execute”. You call the storage “prestabilizing”, but I think
Powers would have called that a “dormitive principle” that just
labels what happens without in any way explaining how or why. The
“stabilizing” is postural control of the normal kind, the same as
would happen if you pushed the cat. If you have seen a video of the
“Big Dog” robot, you have seen this kind of stabilization.

The magician Harry Houdini died because he hadn't stored the energy

to resist a blow to his abdomen when someone hit him before he was
prepared, which he did by tensing (supplying energy to) the stomach
muscles. He had offered a general challenge that nobody could hit
him there hard enough to hurt him, but the hitter was supposed to do
it in a formal kind of way, when he was prepared. When the fatal
blow was struck, his muscles had insufficient stored energy to
oppose the shock energy in the blow.

I think the actual stabilization is simply the use of some of the

pre-tensioned stored energy in normal postural control against a
force that would be destabilizing, but instead of the control being
entirely to oppose a sensed disturbance, it is part of the output
side support of the sequence controller. The “bat” is not simply
sending reference values to muscles that move the paw relative to
the torso, it is sending reference values to all the postural
muscles in a temporal pattern that has been reorganized into the
control hierarchy in the same way as a trained athlete such as a
ballerina coordinates the timings and strengths of her muscle
movements so that what an observer calls “the” action doesn’t
interfere with her balance.

Anyway, those are my first thoughts on the question, before

breakfast.

-------

 I haven't considered the problem since then, but I thought I might

offer the suggestion.

The saying "Reculer pour mieux sauter" seems to be about the same

phenomenon.

Martin

Heather_Broccard-Bel · May 9, 2019, 1:28pm

I suspect that the attached (Aruin, 2003) paper goes some way toward reconciling the “invariant, pre-programmed” feedforward conceptualization with the actual phenomenon: namely, it’s not really invariant (or, well, it’s not invariant in that there is a pre-programmed set of muscles activated in a pre-programmed way). So, I suspect that some of the problem in the “mainstream” is semantic.

Anyway, long (~500 ms) stimulation of motor cortex has already demonstrated that “motor output” seems to be about goals, and not specific, muscle-by-muscle motor programs (see Graziano, 2002; attached), and really the only way you could get that to work is hierarchical control with continuous feedback…

aruin2003.pdf (82 KB)

graziano2002.pdf (267 KB)

···

On Wed, May 8, 2019, 21:16 Martin Taylor csgnet@lists.illinois.edu wrote:

[Martin Taylor 2019.05.09.00.09]

[Rick Marken 2019-05-08_17:57:30]

          RM: Some time ago Heather Broccard-Bell asked me what
the PCT explanation might be of a behavior that looks like
a pretty robust example of anticipatory (or feedforward)
control. Here’s how Heather described it to me:

            HB-B: I was reading
about this finding (and perhaps you are already
familiar) that, if if you measure muscle activation
during behavior, you often see compensatory activation
of the axial musculature well before you get activation
in the muscles that direct the part of the body that
does the thing you’re focused on. For example, they
trained cats to bat at a target with their forepaw. ** Well
before the limb/forepaw muscles are active, there is
activation in the trunk musculature, seemingly
anticipating the requisite compensatory stabilization
that will become necessary when the forepaw is
lifted. ** I am wondering what a PCT perspective
might be on that? What is the relevant controlled
variable, since the thing that actually changes sensory
input that should activate control architecture hasn’t
yet happened? Am I missing something? It appears to be
a rather robust finding.[emphasis mine–RM]

…

          RM: I am posting this to you experts on PCT
(with Heather’s permission) because I think this is a very
interesting discovery that seems to pose a challenge to
PCT and I would like to see what you think about this.

Here is  section of what I wrote to Heather on this question April
--------



[Here is a] quick and dirty PCT approach. I start with the intention
to bat at the target, which would set a reference for a sequence
that could crudely be taken as “store energy, set trigger, wait,
release energy when required”, or in a more general sense “Prepare,
wait, execute”. You call the storage “prestabilizing”, but I think
Powers would have called that a “dormitive principle” that just
labels what happens without in any way explaining how or why. The
“stabilizing” is postural control of the normal kind, the same as
would happen if you pushed the cat. If you have seen a video of the
“Big Dog” robot, you have seen this kind of stabilization.
The magician Harry Houdini died because he hadn't stored the energy
to resist a blow to his abdomen when someone hit him before he was
prepared, which he did by tensing (supplying energy to) the stomach
muscles. He had offered a general challenge that nobody could hit
him there hard enough to hurt him, but the hitter was supposed to do
it in a formal kind of way, when he was prepared. When the fatal
blow was struck, his muscles had insufficient stored energy to
oppose the shock energy in the blow.
I think the actual stabilization is simply the use of some of the
pre-tensioned stored energy in normal postural control against a
force that would be destabilizing, but instead of the control being
entirely to oppose a sensed disturbance, it is part of the output
side support of the sequence controller. The “bat” is not simply
sending reference values to muscles that move the paw relative to
the torso, it is sending reference values to all the postural
muscles in a temporal pattern that has been reorganized into the
control hierarchy in the same way as a trained athlete such as a
ballerina coordinates the timings and strengths of her muscle
movements so that what an observer calls “the” action doesn’t
interfere with her balance.
Anyway, those are my first thoughts on the question, before
breakfast.
-------

 I haven't considered the problem since then, but I thought I might
offer the suggestion.
The saying "Reculer pour mieux sauter" seems to be about the same
phenomenon.
Martin

wmansell · May 9, 2019, 2:18pm

Hurray, PCT wins! But do any of the readers of these papers realise?

···

On 9 May 2019, at 14:28, Heather Broccard-Bell (hebell@ucsd.edu via csgnet Mailing List) csgnet@lists.illinois.edu wrote:

I suspect that the attached (Aruin, 2003) paper goes some way toward reconciling the “invariant, pre-programmed” feedforward conceptualization with the actual phenomenon: namely, it’s not really invariant (or, well, it’s not invariant in that there is a pre-programmed set of muscles activated in a pre-programmed way). So, I suspect that some of the problem in the “mainstream” is semantic.

Anyway, long (~500 ms) stimulation of motor cortex has already demonstrated that “motor output” seems to be about goals, and not specific, muscle-by-muscle motor programs (see Graziano, 2002; attached), and really the only way you could get that to work is hierarchical control with continuous feedback…

On Wed, May 8, 2019, 21:16 Martin Taylor csgnet@lists.illinois.edu wrote:

[Martin Taylor 2019.05.09.00.09]

[Rick Marken 2019-05-08_17:57:30]

          RM: Some time ago Heather Broccard-Bell asked me what
the PCT explanation might be of a behavior that looks like
a pretty robust example of anticipatory (or feedforward)
control. Here’s how Heather described it to me:

            HB-B: I was reading
about this finding (and perhaps you are already
familiar) that, if if you measure muscle activation
during behavior, you often see compensatory activation
of the axial musculature well before you get activation
in the muscles that direct the part of the body that
does the thing you’re focused on. For example, they
trained cats to bat at a target with their forepaw. ** Well
before the limb/forepaw muscles are active, there is
activation in the trunk musculature, seemingly
anticipating the requisite compensatory stabilization
that will become necessary when the forepaw is
lifted. ** I am wondering what a PCT perspective
might be on that? What is the relevant controlled
variable, since the thing that actually changes sensory
input that should activate control architecture hasn’t
yet happened? Am I missing something? It appears to be
a rather robust finding.[emphasis mine–RM]

…

          RM: I am posting this to you experts on PCT
(with Heather’s permission) because I think this is a very
interesting discovery that seems to pose a challenge to
PCT and I would like to see what you think about this.

Here is  section of what I wrote to Heather on this question April
--------



[Here is a] quick and dirty PCT approach. I start with the intention
to bat at the target, which would set a reference for a sequence
that could crudely be taken as “store energy, set trigger, wait,
release energy when required”, or in a more general sense “Prepare,
wait, execute”. You call the storage “prestabilizing”, but I think
Powers would have called that a “dormitive principle” that just
labels what happens without in any way explaining how or why. The
“stabilizing” is postural control of the normal kind, the same as
would happen if you pushed the cat. If you have seen a video of the
“Big Dog” robot, you have seen this kind of stabilization.
The magician Harry Houdini died because he hadn't stored the energy
to resist a blow to his abdomen when someone hit him before he was
prepared, which he did by tensing (supplying energy to) the stomach
muscles. He had offered a general challenge that nobody could hit
him there hard enough to hurt him, but the hitter was supposed to do
it in a formal kind of way, when he was prepared. When the fatal
blow was struck, his muscles had insufficient stored energy to
oppose the shock energy in the blow.
I think the actual stabilization is simply the use of some of the
pre-tensioned stored energy in normal postural control against a
force that would be destabilizing, but instead of the control being
entirely to oppose a sensed disturbance, it is part of the output
side support of the sequence controller. The “bat” is not simply
sending reference values to muscles that move the paw relative to
the torso, it is sending reference values to all the postural
muscles in a temporal pattern that has been reorganized into the
control hierarchy in the same way as a trained athlete such as a
ballerina coordinates the timings and strengths of her muscle
movements so that what an observer calls “the” action doesn’t
interfere with her balance.
Anyway, those are my first thoughts on the question, before
breakfast.
-------

 I haven't considered the problem since then, but I thought I might
offer the suggestion.
The saying "Reculer pour mieux sauter" seems to be about the same
phenomenon.
Martin

<aruin2003.pdf>

<graziano2002.pdf>

bleach · May 9, 2019, 5:50pm

May be my conception of PCT in this regard is flawed but I don’t
see where this is a challenge at all given:

The arrival of a ball (or other object) to bat is expected
Or it is planned to drop the weight.
Then based upon a world view developed through experience, the
mental system has a model that shows that preparation results in a
better overall outcome:
```
    There is a reference for maintaining a specific spatial
```
orientation (batting at something to avoid attack by another
purposeful being for example)
```
    There is a reference for not falling over (or having to
```
stumble to maintain balance).
Now if such pre-action tensioning occurs without any anticipation
for the need, then yes I have a problem.

-bill

···

On 5/8/19 7:00 PM, Richard Marken
( via csgnet Mailing List) wrote:

rsmarken@gmail.com

[Rick Marken 2019-05-08_17:57:30]

          RM: Some time ago Heather Broccard-Bell asked me what
the PCT explanation might be of a behavior that looks like
a pretty robust example of anticipatory (or feedforward)
control. Here’s how Heather described it to me:

            HB-B: I was reading
about this finding (and perhaps you are already
familiar) that, if if you measure muscle activation
during behavior, you often see compensatory activation
of the axial musculature well before you get activation
in the muscles that direct the part of the body that
does the thing you’re focused on.Â For example, they
trained cats to bat at a target with their forepaw.Â ** Well
before the limb/forepaw muscles are active, there is
activation in the trunk musculature, seemingly
anticipating the requisite compensatory stabilization
that will become necessary when the forepaw is
lifted.Â ** I am wondering what a PCT perspective
might be on that?Â What is the relevant controlled
variable, since the thing that actually changes sensory
input that should activate control architecture hasn’t
yet happened?Â Am I missing something?Â It appears to be
a rather robust finding.[emphasis mine–RM]

            RM: This finding certainly looks like it is
inconsistent with the predictions of PCT since it looks
like the cat is generating an output (activation of the
trunk musculature) open loop in anticipation of a future
disturbance (the change in orientation of the body when
the forepaw is lifted, a change in orientation that
would cause the cat to lose its balance if the trunk
were not in the proper position).Â

            RM: Apparently, quite a lot of research has been (and
is still being) done on this “anticipatory postural
control” phenomenon. Here is a paper by Aruin and
Latash that Heather sent on the finding of "anticipatory
postural control"Â when people drop a weight:

Â https://www.dropbox.com/s/hat2ua3ah7rqs79/aruin1995.pdf?dl=0

            RM: Here there is again clear evidence that the
person takes compensatory action (contracts muscles) in
anticipation of a future disturbance that would cause a
loss of balance if the anticipatory compensation had not
occurred.Â

            RM: Heather also described this related

neurophysiological observation:

HB-B: ** Related, it
was discovered that if you knock out the indirect
pathway from motor cortex to spinal cord (through the
reticular formation), but leave the direct pathway
intact, you abolish the “prestabilizing”, but still
get the paw bat.Â ** I think you could explain that,
in a super-oversimplified way, in terms of those
pathways being about different control systems (these
bastards have ascending, descending, and reciprocal
connections at every level, not to mention all kinds of
built-in sensory elements and sensory inputs connecting
from elsewhere, as any good ultracomplex biological
control system would).

          RM: I am posting this to you experts on PCT
(with Heather’s permission) because I think this is a very
interesting discovery that seems to pose a challenge to
PCT and I would like to see what you think about this. I
have an idea about how PCT would explain this
“anticipatory postural control” phenomenon and I think
theÂ Â Aruin and Latash paper pointed to above suggests a
way to test it. But before I give my PCT explanation I’d
like to see what you think.Â

BestÂ

Rick

–
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

Heather_Broccard-Bel · May 9, 2019, 5:53pm

I’m still not sure I satisfactorily answered my own original question, though. And also, if we can compensate for a predicted disturbance (even if we can update our compensation when the compensation is perturbed), the original reference signal is presumably coming from some kind of internal model, no? Is this cool? I thought we were anti-internal model (except for super sophisticated human cognition where we do clearly imagine things)? This suggests cats have one, no?

···

On Thu, May 9, 2019, 07:18 Warren Mansell wmansell@gmail.com wrote:

Hurray, PCT wins! But do any of the readers of these papers realise?

On 9 May 2019, at 14:28, Heather Broccard-Bell (hebell@ucsd.edu via csgnet Mailing List) csgnet@lists.illinois.edu wrote:

I suspect that the attached (Aruin, 2003) paper goes some way toward reconciling the “invariant, pre-programmed” feedforward conceptualization with the actual phenomenon: namely, it’s not really invariant (or, well, it’s not invariant in that there is a pre-programmed set of muscles activated in a pre-programmed way). So, I suspect that some of the problem in the “mainstream” is semantic.

Anyway, long (~500 ms) stimulation of motor cortex has already demonstrated that “motor output” seems to be about goals, and not specific, muscle-by-muscle motor programs (see Graziano, 2002; attached), and really the only way you could get that to work is hierarchical control with continuous feedback…

On Wed, May 8, 2019, 21:16 Martin Taylor csgnet@lists.illinois.edu wrote:

[Martin Taylor 2019.05.09.00.09]

[Rick Marken 2019-05-08_17:57:30]

          RM: Some time ago Heather Broccard-Bell asked me what
the PCT explanation might be of a behavior that looks like
a pretty robust example of anticipatory (or feedforward)
control. Here’s how Heather described it to me:

            HB-B: I was reading
about this finding (and perhaps you are already
familiar) that, if if you measure muscle activation
during behavior, you often see compensatory activation
of the axial musculature well before you get activation
in the muscles that direct the part of the body that
does the thing you’re focused on. For example, they
trained cats to bat at a target with their forepaw. ** Well
before the limb/forepaw muscles are active, there is
activation in the trunk musculature, seemingly
anticipating the requisite compensatory stabilization
that will become necessary when the forepaw is
lifted. ** I am wondering what a PCT perspective
might be on that? What is the relevant controlled
variable, since the thing that actually changes sensory
input that should activate control architecture hasn’t
yet happened? Am I missing something? It appears to be
a rather robust finding.[emphasis mine–RM]

…

          RM: I am posting this to you experts on PCT
(with Heather’s permission) because I think this is a very
interesting discovery that seems to pose a challenge to
PCT and I would like to see what you think about this.

Here is  section of what I wrote to Heather on this question April
--------



[Here is a] quick and dirty PCT approach. I start with the intention
to bat at the target, which would set a reference for a sequence
that could crudely be taken as “store energy, set trigger, wait,
release energy when required”, or in a more general sense “Prepare,
wait, execute”. You call the storage “prestabilizing”, but I think
Powers would have called that a “dormitive principle” that just
labels what happens without in any way explaining how or why. The
“stabilizing” is postural control of the normal kind, the same as
would happen if you pushed the cat. If you have seen a video of the
“Big Dog” robot, you have seen this kind of stabilization.
The magician Harry Houdini died because he hadn't stored the energy
to resist a blow to his abdomen when someone hit him before he was
prepared, which he did by tensing (supplying energy to) the stomach
muscles. He had offered a general challenge that nobody could hit
him there hard enough to hurt him, but the hitter was supposed to do
it in a formal kind of way, when he was prepared. When the fatal
blow was struck, his muscles had insufficient stored energy to
oppose the shock energy in the blow.
I think the actual stabilization is simply the use of some of the
pre-tensioned stored energy in normal postural control against a
force that would be destabilizing, but instead of the control being
entirely to oppose a sensed disturbance, it is part of the output
side support of the sequence controller. The “bat” is not simply
sending reference values to muscles that move the paw relative to
the torso, it is sending reference values to all the postural
muscles in a temporal pattern that has been reorganized into the
control hierarchy in the same way as a trained athlete such as a
ballerina coordinates the timings and strengths of her muscle
movements so that what an observer calls “the” action doesn’t
interfere with her balance.
Anyway, those are my first thoughts on the question, before
breakfast.
-------

 I haven't considered the problem since then, but I thought I might
offer the suggestion.
The saying "Reculer pour mieux sauter" seems to be about the same
phenomenon.
Martin

<aruin2003.pdf>

<graziano2002.pdf>

Heather_Broccard-Bel · May 9, 2019, 7:17pm

I see that according to Bill Leach, internal models are allowed. So then the question becomes, does the PCT interpretation generate any predictions that are meaningfully different from the “mainstream” approaches would predict?

···

Heather C. Broccard-Bell, Ph.D.

Adjunct Assistant Professor

Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems

Nieh Honey Bee Lab

Division of Biological Sciences; Section of Ecology, Behavior, and Evolution

University of California San Diego

619.757.4694

bleach · May 9, 2019, 7:53pm

Heather,

  I don't know if it was Rick or Martin or both that mentioned

sequence level. Preparing muscles for an action that is perceived
to be soon required is not beyond the realm of PCT.

  No, I don't think you can claim that models are outside of PCT

concepts. It is easily argued (in my opinion) that even some of
the physical neural structures themselves are indeed model based.
That is they are arranged and ‘rewired’ by the reorganizing system
(or inherited from an ancestral reorganization) to function
optimally.

  The postulated program level is implicitly model based (as are

all programs, computer or otherwise).

  I missed your original post so I am quite delighted that Rick

mentioned it and that has had the result of renewing discussion of
the idea.

bill

···

On 5/9/19 11:53 AM, Heather
Broccard-Bell ( via csgnet Mailing List) wrote:

hebell@ucsd.edu

    I'm still not sure I satisfactorily answered my
own original question, though. And also, if we can compensate
for a predicted disturbance (even if we can update our
compensation when the compensation is perturbed), the original
reference signal is presumably coming from some kind of internal
model, no? Is this cool? I thought we were anti-internal model
(except for super sophisticated human cognition where we do
clearly imagine things)? This suggests cats have one, no?

          On Thu, May 9, 2019, 07:18
Warren Mansell <wmansell@gmail.com >
wrote:

              Hurray, PCT wins! But do any of the

readers of these papers realise?

              On 9 May 2019, at 14:28, Heather Broccard-Bell (hebell@ucsd.edu
              via csgnet Mailing List) <csgnet@lists.illinois.edu                  >

wrote:

                        I suspect that the attached
(Aruin, 2003) paper goes some way toward
reconciling the “invariant, pre-programmed”
feedforward conceptualization with the
actual phenomenon: namely, it’s not really
invariant (or, well, it’s not invariant in
that there is a pre-programmed set of
muscles activated in a pre-programmed way).
So, I suspect that some of the problem in
the “mainstream” is semantic.

                          Anyway, long (~500 ms)
stimulation of motor cortex has already
demonstrated that “motor output” seems to
be about goals, and not specific,
muscle-by-muscle motor programs (see
Graziano, 2002; attached), and really the
only way you could get that to work is
hierarchical control with continuous
feedback…

                          On Wed,
May 8, 2019, 21:16 Martin Taylor <csgnet@lists.illinois.edu >
wrote:

[Martin Taylor 2019.05.09.00.09]

                                      [Rick Marken

2019-05-08_17:57:30]

                                      RM: Some time ago Heather
Broccard-Bell asked me what
the PCT explanation might be
of a behavior that looks like
a pretty robust example of
anticipatory (or feedforward)
control. Here’s how Heather
described it to me:

                                        HB-B: I was reading about this
finding (and perhaps you are
already familiar) that, if
if you measure muscle
activation during behavior,
you often see compensatory
activation of the axial
musculature well before you
get activation in the
muscles that direct the part
of the body that does the
thing you’re focused on.
For example, they trained
cats to bat at a target with
their forepaw. ** Well
before the limb/forepaw
muscles are active, there
is activation in the trunk
musculature, seemingly
anticipating the requisite
compensatory stabilization
that will become necessary
when the forepaw is
lifted. ** I am
wondering what a PCT
perspective might be on
that? What is the relevant
controlled variable, since
the thing that actually
changes sensory input that
should activate control
architecture hasn’t yet
happened? Am I missing
something? It appears to be
a rather robust
finding.[emphasis mine–RM]

…

                                      RM: I am posting
this to you experts on PCT
(with Heather’s permission)
because I think this is a very
interesting discovery that
seems to pose a challenge to
PCT and I would like to see
what you think about this.

                            Here is  section of what I wrote to
Heather on this question April 12.
                            --------



                            [Here is a] quick and dirty PCT
approach. I start with the intention to
bat at the target, which would set a
reference for a sequence that could
crudely be taken as “store energy, set
trigger, wait, release energy when
required”, or in a more general sense
“Prepare, wait, execute”. You call the
storage “prestabilizing”, but I think
Powers would have called that a
“dormitive principle” that just labels
what happens without in any way
explaining how or why. The “stabilizing”
is postural control of the normal kind,
the same as would happen if you pushed
the cat. If you have seen a video of the
“Big Dog” robot, you have seen this kind
of stabilization.
                            The magician Harry Houdini died because
he hadn’t stored the energy to resist a
blow to his abdomen when someone hit him
before he was prepared, which he did by
tensing (supplying energy to) the
stomach muscles. He had offered a
general challenge that nobody could hit
him there hard enough to hurt him, but
the hitter was supposed to do it in a
formal kind of way, when he was
prepared. When the fatal blow was
struck, his muscles had insufficient
stored energy to oppose the shock energy
in the blow.
                            I think the actual stabilization is
simply the use of some of the
pre-tensioned stored energy in normal
postural control against a force that
would be destabilizing, but instead of
the control being entirely to oppose a
sensed disturbance, it is part of the
output side support of the sequence
controller. The “bat” is not simply
sending reference values to muscles that
move the paw relative to the torso, it
is sending reference values to all the
postural muscles in a temporal pattern
that has been reorganized into the
control hierarchy in the same way as a
trained athlete such as a ballerina
coordinates the timings and strengths of
her muscle movements so that what an
observer calls “the” action doesn’t
interfere with her balance.
                            Anyway, those are my first thoughts on
the question, before breakfast.
                            -------

                             I haven't considered the problem since
then, but I thought I might offer the
suggestion.
                            The saying "Reculer pour mieux sauter"
seems to be about the same phenomenon.
                            Martin

<aruin2003.pdf>

<graziano2002.pdf>

_Erling_Jorgensen · May 9, 2019, 7:54pm

[From Erling Jorgensen (2019.05.09 1530 EDT)]

EJ: To say “internal models” still requires some notion of mechanism, so until that is specified a little better, in my mind the term remains ‘hand waving’. It seems like “pre-action tensioning” is a valid shorthand description of the phenomenon, for which “anticipatory stabilization” had been proposed as an explanation.

EJ: I rather like Martin’s proposal for controlling a prior stage in a sequence. I think of the similarity to how Segway machines are operated, as an analogue for human walking. Before an actual step is taken, there is leaning forward and the center of gravity shifts. I wonder whether that is akin to the “activation in the trunk musculature” that is measured before a cat’s forepaw is actually lifted. But “tensioning a muscle” is still a present moment control process, via equal increase of force to both sides of an opponent-process muscle. It requires neither anticipatory effects nor discernible movement.

EJ: I appreciate the way Heather is fashioning the question, in terms of “meaningfully different predictions”. The standard PCT model says output is generated by error already occurring, which means either a disturbance moved an existing perception away from its preferred value, or a change in reference happened toward which the perception now moves. Feedforward effects are a different model, and it is not yet clear whether PCT has yet had to make much use of such mechanisms.

Erling

Confidentiality: * This message is intended only for the addressee, and may contain information that is privileged and confidential under HIPAA, 42CFR Part 2, and/or other applicable State and Federal laws. If you are not the addressee, or the employer or agent responsible for delivering the message to the addressee, any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this in error, please notify the sender immediately and delete the material from your computer. Thank you for your cooperation.*

Please also note: * Under 42 CFR part 2 you are prohibited from making any further disclosure of information that identifies an individual as having or having had a substance use disorder unless it is expressly permitted by the written consent of the individual whose information is being disclosed or as otherwise permitted by 42 CFR Part 2.*

“Heather Broccard-Bell” (hebell@ucsd.edu via csgnet Mailing List) csgnet@lists.illinois.edu 5/9/2019 3:17 PM >>>

CAUTION - This is an EXTERNAL email - DO NOT open attachments or links in unexpected emails or from unknown senders *

I see that according to Bill Leach, internal models are allowed. So then the question becomes, does the PCT interpretation generate any predictions that are meaningfully different from the “mainstream” approaches would predict?

···

Heather C. Broccard-Bell, Ph.D.

Adjunct Assistant Professor

Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems

Nieh Honey Bee Lab

Division of Biological Sciences; Section of Ecology, Behavior, and Evolution

University of California San Diego

619.757.4694

Heather_Broccard-Bel · May 9, 2019, 7:59pm

Sorry, that was meant to go to everyone, not just Martin

···

Heather C. Broccard-Bell, Ph.D.

Adjunct Assistant Professor

Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems

Nieh Honey Bee Lab

Division of Biological Sciences; Section of Ecology, Behavior, and Evolution

University of California San Diego

619.757.4694

Heather_Broccard-Bel · May 9, 2019, 8:26pm

Also, I totally agree that things like connection changes are some kind of internal model… but was under the (I guess false?) impression that internal representation was not allowed? Just like how signal transduction from sensory receptors is some kind of internal representation…

···

Heather C. Broccard-Bell, Ph.D.

Adjunct Assistant Professor

Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems

Nieh Honey Bee Lab

Division of Biological Sciences; Section of Ecology, Behavior, and Evolution

University of California San Diego

619.757.4694

MartinT · May 9, 2019, 8:40pm

[Martin Taylor 2019.05.09.16.13]

If you call the intended result of an action (bringing a perception

toward its reference value) a prediction, I would say that’s a
strange use of “prediction”. better understanding of what is going
on. If I decide to go out and do go out, have I predicted that I
will go out? The Friston “Free Energy” people say exactly that,
probably for the reasons you suggest below. I would call it at best
a self-fulfilling prediction, one that comes to pass because of the
actions of the one who predicts and wouldn’t come to pass in the
absence of those actions.
I could be happy with saying that there is a form of prediction
here, which is a prediction that the Real Reality world will
continue to act much as it has been acting for a reasonably long
time. It’s not a prediction that when the cat wants to bat at
something it will do what is necessary to bat at the thing without
falling over. But who is predicting? Nobody. Both evolution and
reorganization work with past data, not predicted future data. The
“prediction” is simply the working of Natural Law, whether we know
the law or not.
Martin

···

On 2019/05/9 3:59 PM, Heather
Broccard-Bell ( via csgnet Mailing List) wrote:

hebell@ucsd.edu

      Sorry, that was meant to go to everyone, not just

Martin

        On Thu, May 9, 2019 at 12:58
PM Heather Broccard-Bell <hebell@ucsd.edu >
wrote:

          All right, so what then is the difference
between operating on a reference signal from an internal
model, and making a prediction?

          Does that not amount to the same thing? 
After all, the parameters of the internal model are set
through experience (or evolution – which is some kind of
sort of larger “experience”). So then the system can be
biased in a way that allows it to function better than
ones that aren’t biased that way in a given environment.

mmt@mmtaylor.net

[Martin Taylor 2019.05.09.14.35]
                On
2019/05/9 1:53 PM, Heather Broccard-Bell wrote:

                  I'm still not sure I satisfactorily
answered my own original question, though. And
also, if we can compensate for a predicted
disturbance (even if we can update our
compensation when the compensation is perturbed),
the original reference signal is presumably coming
from some kind of internal model, no? Is this
cool? I thought we were anti-internal model
(except for super sophisticated human cognition
where we do clearly imagine things)? This
suggests cats have one, no?

              I guess your question is about what the pre-stored
energy in the cats muscles is for, and whether it
implies prediction of an external disturbance. I don’t
think it does. Remember that reference values change,
and that reorganization changes parameters and
level-to-level interconnection parameters. Maybe the
first time the cat ever tried to bat at something it
didn’t hit it properly, and maybe it stumbled or fell
after the attempt. Nothing of that happened before the
cat tried to compensate for a disturbance from
outside, so there’s no prediction involved. No
subsequent unpredicted disturbance is involved in your
question.
              Everything else is internally generated as changing
reference values feed down the levels of control all
the way to the individual muscle tension control
loops. The feedback effects of the actions generated
by the control units whose reference levels were
changed are fairly consistent over many attempts at
batting at locally moving things (as an observer sees
it), so reorganization has the opportunity to build
the proper coordinations and sequences of muscle
tensions.
              Yes, there's an internal model. It's in the
reorganized connections and parameter settings for the
control loops used to do the batting, and part of that
is pre-tensioning the muscles both to perform the bat
and to reduce the sudden effect of the action until
fast normal low-level control loops correct any
remaining perceptual errors. I bet you do the same if
you pick up a pile of dishes to put them on a shelf.
If you started out with slack muscles, you might fall
when you picked them up, and when you released them
onto the shelf you might fall in the other direction
if all the tension was devoted to holding up the
plates, rather than to compensate for the change in
balance, which you have encountered every time you
picked up or released a weight since you were old
enough to try.
              It might be evolution, not primarily reorganization in
the individual cat, though, because cats are
notoriously good at keeping their attitude with
respect to gravity, landing with feet toward the
ground if falling from a height, and performing
remarkable rapid body contortions while remaining
stable.
              Martin

                        On Thu, May
9, 2019, 07:18 Warren Mansell <wmansell@gmail.com >
wrote:

                            Hurray, PCT wins! But do

any of the readers of these papers
realise?

                            On 9 May 2019, at 14:28, Heather
Broccard-Bell (hebell@ucsd.edu
via csgnet Mailing List) <csgnet@lists.illinois.edu >
wrote:

                                      I suspect that
the attached (Aruin, 2003)
paper goes some way toward
reconciling the “invariant,
pre-programmed” feedforward
conceptualization with the
actual phenomenon: namely,
it’s not really invariant (or,
well, it’s not invariant in
that there is a pre-programmed
set of muscles activated in a
pre-programmed way). So, I
suspect that some of the
problem in the “mainstream” is
semantic.

                                        Anyway, long
(~500 ms) stimulation of
motor cortex has already
demonstrated that “motor
output” seems to be about
goals, and not specific,
muscle-by-muscle motor
programs (see Graziano,
2002; attached), and really
the only way you could get
that to work is hierarchical
control with continuous
feedback…

                                        On Wed,
May 8, 2019, 21:16 Martin
Taylor <csgnet@lists.illinois.edu >
wrote:

                                          [Martin Taylor

2019.05.09.00.09]

                                                    [Rick

Marken
2019-05-08_17:57:30]

                                                    RM: Some time
ago Heather
Broccard-Bell
asked me what
the PCT
explanation
might be of a
behavior that
looks like a
pretty robust
example of
anticipatory (or
feedforward)
control. Here’s
how Heather
described it to
me:

                                                      HB-B:
I was reading
about this
finding (and
perhaps you
are already
familiar)
that, if if
you measure
muscle
activation
during
behavior, you
often see
compensatory
activation of
the axial
musculature
well before
you get
activation in
the muscles
that direct
the part of
the body that
does the thing
you’re focused
on. For
example, they
trained cats
to bat at a
target with
their
forepaw. ** Well
before the
limb/forepaw
muscles are
active, there
is activation
in the trunk
musculature,
seemingly
anticipating
the requisite
compensatory
stabilization
that will
become
necessary when
the forepaw is
lifted. ** I
am wondering
what a PCT
perspective
might be on
that? What is
the relevant
controlled
variable,
since the
thing that
actually
changes
sensory input
that should
activate
control
architecture
hasn’t yet
happened? Am
I missing
something? It
appears to be
a rather
robust
finding.[emphasis
mine–RM]

…

                                                    RM:
I am posting
this to you
experts on PCT
(with Heather’s
permission)
because I think
this is a very
interesting
discovery that
seems to pose a
challenge to PCT
and I would like
to see what you
think about
this.

                                          Here is  section of what I
wrote to Heather on this
question April 12.
                                          --------



                                          [Here is a] quick and
dirty PCT approach. I
start with the intention
to bat at the target,
which would set a
reference for a sequence
that could crudely be
taken as “store energy,
set trigger, wait, release
energy when required”, or
in a more general sense
“Prepare, wait, execute”.
You call the storage
“prestabilizing”, but I
think Powers would have
called that a “dormitive
principle” that just
labels what happens
without in any way
explaining how or why. The
“stabilizing” is postural
control of the normal
kind, the same as would
happen if you pushed the
cat. If you have seen a
video of the “Big Dog”
robot, you have seen this
kind of stabilization.
                                          The magician Harry Houdini
died because he hadn’t
stored the energy to
resist a blow to his
abdomen when someone hit
him before he was
prepared, which he did by
tensing (supplying energy
to) the stomach muscles.
He had offered a general
challenge that nobody
could hit him there hard
enough to hurt him, but
the hitter was supposed to
do it in a formal kind of
way, when he was prepared.
When the fatal blow was
struck, his muscles had
insufficient stored energy
to oppose the shock energy
in the blow.
                                          I think the actual
stabilization is simply
the use of some of the
pre-tensioned stored
energy in normal postural
control against a force
that would be
destabilizing, but instead
of the control being
entirely to oppose a
sensed disturbance, it is
part of the output side
support of the sequence
controller. The “bat” is
not simply sending
reference values to
muscles that move the paw
relative to the torso, it
is sending reference
values to all the postural
muscles in a temporal
pattern that has been
reorganized into the
control hierarchy in the
same way as a trained
athlete such as a
ballerina coordinates the
timings and strengths of
her muscle movements so
that what an observer
calls “the” action doesn’t
interfere with her
balance.
                                          Anyway, those are my first
thoughts on the question,
before breakfast.
                                          -------

                                           I haven't considered the
problem since then, but I
thought I might offer the
suggestion.
                                          The saying "Reculer pour
mieux sauter" seems to be
about the same phenomenon.
                                          Martin

<aruin2003.pdf>

<graziano2002.pdf>

–
Heather C.
Broccard-Bell, Ph.D.

                                              Adjunct

Assistant Professor

                                              Department of

Psychological Sciences

                                              University of

San Diego

                                            Visiting

Scholar

                                            The role

of noise / variability
in complex, adaptive
systems

                                            Nieh

Honey Bee Lab

                                            Division

of Biological Sciences;
Section of Ecology,
Behavior, and Evolution

                                            University

of California San Diego

                                            619.757.4694

–
Heather C.
Broccard-Bell, Ph.D.

                                          Adjunct

Assistant Professor

                                          Department

of Psychological Sciences

                                          University

of San Diego

                                        Visiting

Scholar

                                        The role of

noise / variability in
complex, adaptive systems

                                        Nieh Honey

Bee Lab

                                        Division of

Biological Sciences; Section
of Ecology, Behavior, and
Evolution

                                        University

of California San Diego

                                        619.757.4694

Heather_Broccard-Bel · May 9, 2019, 8:49pm

I agree, nobody is actively making a prediction, but the system is tuned to perform better in the world as it as previously experienced it (as stated), because mostly this works out – because often the world is kind of consistent. What makes a prediction different, and what then does it mean to make a prediction? Does a prediction have to be a fully-formed set of anticipated data? I think I am not understanding what makes “prediction” different from “bias based on past experience”.

···

Heather C. Broccard-Bell, Ph.D.

Adjunct Assistant Professor

Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems

Nieh Honey Bee Lab

Division of Biological Sciences; Section of Ecology, Behavior, and Evolution

University of California San Diego

619.757.4694

bleach · May 9, 2019, 9:33pm

Heather, don’t make the mistake of presuming that I am somehow an
authority on PCT here!

  There is not much that applied PCT can predict in a formal manner

yet. Bill Powers made quite a few predictions that turned out to
be quite good when he developed the theory. The hierarchy is
pretty strongly supported by the actual measured time delays for
various kinds of behavior changes is an example. Some PCT
stimulators have predicted a subject’s performance on a new
“tracking task” with remarkable accuracy.

  One difficulty that I perceive is that many the behavioral

science arena do often make pretty accurate predictions but their
‘school of behavioral science’ actually does not provide any
support for (or for how) they reached their conclusions. Another
way of talking about what I’m trying to say is that these
successful practitioners (usually Psychiatrists) are using some
sort of a model in their own mind to generate their conclusions
but that model has no support in their school of the science.

  An additional difference is that in PCT there is no such thing as

an ‘outlier.’ If an outlier shows up then there is something
wrong with the experiment or there is something wrong with the
theory (or both).

  There is a subtle but vitally important difference between seeing

that an environmental disturbance causes a behavior change and
seeing that a control system changes behavior (output) to control
for disturbances to a perception.

  I don't know if this helps or hinders.  The ability to use PCT in

applied science is in its infancy.

bill

···

On 5/9/19 1:17 PM, Heather
Broccard-Bell ( via csgnet Mailing List) wrote:

hebell@ucsd.edu

    I see that according to Bill Leach, internal models
are allowed. So then the question becomes, does the PCT
interpretation generate any predictions that are meaningfully
different from the “mainstream” approaches would predict?

      On Thu, May 9, 2019 at 10:53
AM Heather Broccard-Bell hebell@ucsd.edu wrote:

        I'm still not sure I satisfactorily answered
my own original question, though. And also, if we can
compensate for a predicted disturbance (even if we can
update our compensation when the compensation is perturbed),
the original reference signal is presumably coming from some
kind of internal model, no? Is this cool? I thought we
were anti-internal model (except for super sophisticated
human cognition where we do clearly imagine things)? This
suggests cats have one, no?

              On Thu, May 9, 2019,
07:18 Warren Mansell <wmansell@gmail.com >
wrote:

                  Hurray, PCT wins! But do any of the

readers of these papers realise?

                  On 9 May 2019, at 14:28, Heather Broccard-Bell (hebell@ucsd.edu via
csgnet Mailing List) <csgnet@lists.illinois.edu >
wrote:

                            I suspect that the
attached (Aruin, 2003) paper goes some
way toward reconciling the “invariant,
pre-programmed” feedforward
conceptualization with the actual
phenomenon: namely, it’s not really
invariant (or, well, it’s not invariant
in that there is a pre-programmed set of
muscles activated in a pre-programmed
way). So, I suspect that some of the
problem in the “mainstream” is semantic.

                              Anyway, long (~500 ms)
stimulation of motor cortex has
already demonstrated that “motor
output” seems to be about goals, and
not specific, muscle-by-muscle motor
programs (see Graziano, 2002;
attached), and really the only way you
could get that to work is hierarchical
control with continuous feedback…

                              On
Wed, May 8, 2019, 21:16 Martin Taylor
<csgnet@lists.illinois.edu >
wrote:

                                [Martin Taylor

2019.05.09.00.09]

                                          [Rick Marken

2019-05-08_17:57:30]

                                          RM: Some time ago
Heather Broccard-Bell
asked me what the PCT
explanation might be of a
behavior that looks like a
pretty robust example of
anticipatory (or
feedforward) control.
Here’s how Heather
described it to me:

                                            HB-B:
I was reading about this
finding (and perhaps you
are already familiar)
that, if if you measure
muscle activation during
behavior, you often see
compensatory activation
of the axial musculature
well before you get
activation in the
muscles that direct the
part of the body that
does the thing you’re
focused on. For
example, they trained
cats to bat at a target
with their forepaw. ** Well
before the
limb/forepaw muscles
are active, there is
activation in the
trunk musculature,
seemingly anticipating
the requisite
compensatory
stabilization that
will become necessary
when the forepaw is
lifted. ** I am
wondering what a PCT
perspective might be on
that? What is the
relevant controlled
variable, since the
thing that actually
changes sensory input
that should activate
control architecture
hasn’t yet happened? Am
I missing something? It
appears to be a rather
robust finding.[emphasis
mine–RM]

…

                                          RM: I am
posting this to you
experts on PCT (with
Heather’s permission)
because I think this is a
very interesting discovery
that seems to pose a
challenge to PCT and I
would like to see what you
think about this.

                                Here is  section of what I wrote to
Heather on this question April 12.
                                --------



                                [Here is a] quick and dirty PCT
approach. I start with the intention
to bat at the target, which would
set a reference for a sequence that
could crudely be taken as “store
energy, set trigger, wait, release
energy when required”, or in a more
general sense “Prepare, wait,
execute”. You call the storage
“prestabilizing”, but I think Powers
would have called that a “dormitive
principle” that just labels what
happens without in any way
explaining how or why. The
“stabilizing” is postural control of
the normal kind, the same as would
happen if you pushed the cat. If you
have seen a video of the “Big Dog”
robot, you have seen this kind of
stabilization.
                                The magician Harry Houdini died
because he hadn’t stored the energy
to resist a blow to his abdomen when
someone hit him before he was
prepared, which he did by tensing
(supplying energy to) the stomach
muscles. He had offered a general
challenge that nobody could hit him
there hard enough to hurt him, but
the hitter was supposed to do it in
a formal kind of way, when he was
prepared. When the fatal blow was
struck, his muscles had insufficient
stored energy to oppose the shock
energy in the blow.
                                I think the actual stabilization is
simply the use of some of the
pre-tensioned stored energy in
normal postural control against a
force that would be destabilizing,
but instead of the control being
entirely to oppose a sensed
disturbance, it is part of the
output side support of the sequence
controller. The “bat” is not simply
sending reference values to muscles
that move the paw relative to the
torso, it is sending reference
values to all the postural muscles
in a temporal pattern that has been
reorganized into the control
hierarchy in the same way as a
trained athlete such as a ballerina
coordinates the timings and
strengths of her muscle movements so
that what an observer calls “the”
action doesn’t interfere with her
balance.
                                Anyway, those are my first thoughts
on the question, before breakfast.
                                -------

                                 I haven't considered the problem
since then, but I thought I might
offer the suggestion.
                                The saying "Reculer pour mieux
sauter" seems to be about the same
phenomenon.
                                Martin

<aruin2003.pdf>

<graziano2002.pdf>

–
Heather C.
Broccard-Bell, Ph.D.

                                        Adjunct

Assistant Professor

                                        Department

of Psychological Sciences

                                        University

of San Diego

                                      Visiting

Scholar

                                      The role of

noise / variability in
complex, adaptive systems

                                      Nieh Honey Bee

Lab

                                      Division of
Biological Sciences; Section
of Ecology, Behavior, and
Evolution

                                      University of

California San Diego

                                      619.757.4694

Heather_Broccard-Bel · May 9, 2019, 9:56pm

Indeed, it is in its infancy, and I wish more people would work on it

···

Heather C. Broccard-Bell, Ph.D.

Adjunct Assistant Professor

Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems

Nieh Honey Bee Lab

Division of Biological Sciences; Section of Ecology, Behavior, and Evolution

University of California San Diego

619.757.4694

bleach · May 9, 2019, 10:02pm

Feedforward in engineering control, is model based. The model is
often implicit, as in experience where the control system retains
information concerning how well it controlled a particular sensed
parameter and other related parameters. The easiest example that
I can think of, is the room heating system. In such a system, if
you add sensors to monitor outside temperature, status of doors
and windows, and the number of occupants; and record the amount of
error in maintaining the room temperature, then you can compute
anticipatory actions to minimize control error. The system might
start the heater before the “set temperature” by raising the
controller reference if the door to the outside was opened and the
outside temperature is very low.

  There is, in my mind, no reason why experience should not allow

for similar action in living systems. Indeed, even Martin’s
example of sequencing involves a model in that the sequence is
a model. In a specific subject that model was either constructed
or inherited but at some point, maybe eons ago it was
constructed. Some form or reorganization was involved in the
construction to correct for uncontrolled or poorly controlled
error.

  "Pre-action tensioning" is a fine label for the sequence step but

keep in mind that the whole sequence is based upon anticipating
the need to run the sequence to ensure satisfactory control of
perception including an expected future perception.

bill

···

On 5/9/19 1:54 PM, “Erling Jorgensen”
( via csgnet Mailing List) wrote:

EJorgensen@riverbendcmhc.org

[From Erling Jorgensen (2019.05.09 1530 EDT)]

      EJ:  To say "internal models" still requires some notion of
mechanism, so until that is specified a little better, in my
mind the term remains ‘hand waving’. It seems like
“pre-action tensioning” is a valid shorthand description of
the phenomenon, for which “anticipatory stabilization” had
been proposed as an explanation.

      EJ:  I rather like Martin's proposal for controlling a
prior stage in a sequence. I think of the similarity to how
Segway machines are operated, as an analogue for human
walking. Before an actual step is taken, there is leaning
forward and the center of gravity shifts. I wonder whether
that is akin to the “activation in the trunk musculature” that
is measured before a cat’s forepaw is actually lifted. But
“tensioning a muscle” is still a present moment control
process, via equal increase of force to both sides of
an opponent-process muscle. It requires neither anticipatory
effects nor discernible movement.

      EJ:  I appreciate the way Heather is fashioning the
question, in terms of “meaningfully different predictions”.
The standard PCT model says output is generated by error
already occurring, which means either a disturbance moved an
existing perception away from its preferred value, or a change
in reference happened toward which the perception now moves.
Feedforward effects are a different model, and it is not yet
clear whether PCT has yet had to make much use of such
mechanisms.

Erling

Confidentiality:*
This message is intended only for the addressee, and
may contain information that is privileged and
confidential under HIPAA, 42CFR Part 2, and/or other
applicable State and Federal laws. If you are not the
addressee, or the employer or agent responsible for
delivering the message to the addressee, any
dissemination, distribution or copying of this
communication is strictly prohibited. ** If you have
received this in error, please notify the sender
immediately and delete the material from your
computer. Thank you for your cooperation.***

Please also note:*
Under 42 CFR part 2 you are prohibited from making
any further disclosure of information that
identifies an individual as having or having had a
substance use disorder unless it is expressly
permitted by the written consent of the individual
whose information is being disclosed or as otherwise
permitted by 42 CFR Part 2.*

        >>> "Heather Broccard-Bell" ( via
csgnet Mailing List) 5/9/2019 3:17 PM >>>
            * **CAUTION** - This is an
EXTERNAL email - DO NOT open attachments or links in
unexpected emails or from unknown senders *

          I see that according to Bill Leach, internal
models are allowed. So then the question becomes, does the
PCT interpretation generate any predictions that are
meaningfully different from the “mainstream” approaches
would predict?

            On Thu, May 9, 2019 at
10:53 AM Heather Broccard-Bell <hebell@ucsd.edu >
wrote:

              I'm still not sure I satisfactorily
answered my own original question, though. And also,
if we can compensate for a predicted disturbance (even
if we can update our compensation when the
compensation is perturbed), the original reference
signal is presumably coming from some kind of internal
model, no? Is this cool? I thought we were
anti-internal model (except for super sophisticated
human cognition where we do clearly imagine things)?
This suggests cats have one, no?

                    On Thu, May 9,
2019, 07:18 Warren Mansell <wmansell@gmail.com >
wrote:

                        Hurray, PCT wins! But do any of

the readers of these papers realise?

                        On 9 May 2019, at 14:28, Heather
Broccard-Bell (hebell@ucsd.edu
via csgnet Mailing List) <csgnet@lists.illinois.edu >
wrote:

                                  I suspect that the
attached (Aruin, 2003) paper goes
some way toward reconciling the
“invariant, pre-programmed”
feedforward conceptualization with
the actual phenomenon: namely,
it’s not really invariant (or,
well, it’s not invariant in that
there is a pre-programmed set of
muscles activated in a
pre-programmed way). So, I suspect
that some of the problem in the
“mainstream” is semantic.

                                    Anyway, long (~500
ms) stimulation of motor cortex
has already demonstrated that
“motor output” seems to be about
goals, and not specific,
muscle-by-muscle motor programs
(see Graziano, 2002; attached),
and really the only way you
could get that to work is
hierarchical control with
continuous feedback…

                                    On
Wed, May 8, 2019, 21:16 Martin
Taylor <csgnet@lists.illinois.edu >
wrote:

                                      [Martin

Taylor 2019.05.09.00.09]

                                                [Rick

Marken
2019-05-08_17:57:30]

                                                RM: Some time ago
Heather
Broccard-Bell asked
me what the PCT
explanation might be
of a behavior that
looks like a pretty
robust example of
anticipatory (or
feedforward)
control. Here’s how
Heather described it
to me:

                                                  HB-B: I
was reading about
this finding (and
perhaps you are
already familiar)
that, if if you
measure muscle
activation during
behavior, you
often see
compensatory
activation of the
axial musculature
well before you
get activation in
the muscles that
direct the part of
the body that does
the thing you’re
focused on. For
example, they
trained cats to
bat at a target
with their
forepaw. ** Well
before the
limb/forepaw
muscles are
active, there is
activation in
the trunk
musculature,
seemingly
anticipating the
requisite
compensatory
stabilization
that will become
necessary when
the forepaw is
lifted.** I
am wondering what
a PCT perspective
might be on that?
What is the
relevant
controlled
variable, since
the thing that
actually changes
sensory input that
should activate
control
architecture
hasn’t yet
happened? Am I
missing something?
It appears to be a
rather robust
finding.[emphasis
mine–RM]

…

                                                RM: I
am posting this to
you experts on PCT
(with Heather’s
permission) because
I think this is a
very interesting
discovery that seems
to pose a challenge
to PCT and I would
like to see what you
think about this.

                                      Here is section of what I
wrote to Heather on this
question April 12.
                                      --------



                                      [Here is a] quick and dirty
PCT approach. I start with the
intention to bat at the
target, which would set a
reference for a sequence that
could crudely be taken as
“store energy, set trigger,
wait, release energy when
required”, or in a more
general sense “Prepare, wait,
execute”. You call the storage
“prestabilizing”, but I think
Powers would have called that
a “dormitive principle” that
just labels what happens
without in any way explaining
how or why. The “stabilizing”
is postural control of the
normal kind, the same as would
happen if you pushed the cat.
If you have seen a video of
the “Big Dog” robot, you have
seen this kind of
stabilization.
                                      The magician Harry Houdini
died because he hadn’t stored
the energy to resist a blow to
his abdomen when someone hit
him before he was prepared,
which he did by tensing
(supplying energy to) the
stomach muscles. He had
offered a general challenge
that nobody could hit him
there hard enough to hurt him,
but the hitter was supposed to
do it in a formal kind of way,
when he was prepared. When the
fatal blow was struck, his
muscles had insufficient
stored energy to oppose the
shock energy in the blow.
                                      I think the actual
stabilization is simply the
use of some of the
pre-tensioned stored energy in
normal postural control
against a force that would be
destabilizing, but instead of
the control being entirely to
oppose a sensed disturbance,
it is part of the output side
support of the sequence
controller. The “bat” is not
simply sending reference
values to muscles that move
the paw relative to the torso,
it is sending reference values
to all the postural muscles in
a temporal pattern that has
been reorganized into the
control hierarchy in the same
way as a trained athlete such
as a ballerina coordinates the
timings and strengths of her
muscle movements so that what
an observer calls “the” action
doesn’t interfere with her
balance.
                                      Anyway, those are my first
thoughts on the question,
before breakfast.
                                      -------

                                      I haven't considered the
problem since then, but I
thought I might offer the
suggestion.
                                      The saying "Reculer pour mieux
sauter" seems to be about the
same phenomenon.
                                      Martin

<aruin2003.pdf>

<graziano2002.pdf>

–

                                          Heather C.

Broccard-Bell, Ph.D.

                                              Adjunct

Assistant Professor

                                              Department of

Psychological Sciences

                                              University of

San Diego

                                            Visiting

Scholar

                                            The role

of noise / variability
in complex, adaptive
systems

                                            Nieh

Honey Bee Lab

                                            Division

of Biological Sciences;
Section of Ecology,
Behavior, and Evolution

                                            University

of California San Diego

                                            619.757.4694

hebell@ucsd.educsgnet@lists.illinois.edu

bleach · May 9, 2019, 10:39pm

I agree, nobody is actively making a prediction, but the system is tuned to perform better in the world as it as previously experienced it (as stated), because mostly this works out -- because often the world is kind of consistent. What makes a prediction different, and what then does it mean to make a prediction? Does a prediction have to be a fully-formed set of anticipated data? I think I am not understanding what makes "prediction" different from "bias based on past experience".

I suggest that "bias based on past experience" is the basis for all prediction with respect to control system operation. The term prediction also encompasses things like "the world is going to end in 12 years" which I think it is reasonable to assume is not experience based.

bill

[Martin Taylor 2019.05.09.16.13]

Sorry, that was meant to go to everyone, not just Martin

All right, so what then is the difference between operating on a reference signal from an internal model, and making a prediction?

If you call the intended result of an action (bringing a perception toward its reference value) a prediction, I would say that's a strange use of "prediction". better understanding of what is going on. If I decide to go out and do go out, have I predicted that I will go out? The Friston "Free Energy" people say exactly that, probably for the reasons you suggest below. I would call it at best a self-fulfilling prediction, one that comes to pass because of the actions of the one who predicts and wouldn't come to pass in the absence of those actions.

Does that not amount to the same thing? After all, the parameters of the internal model are set through experience (or evolution -- which is some kind of sort of larger "experience"). So then the system can be biased in a way that allows it to function better than ones that aren't biased that way in a given environment.

I could be happy with saying that there is a form of prediction here, which is a prediction that the Real Reality world will continue to act much as it has been acting for a reasonably long time. It's not a prediction that when the cat wants to bat at something it will do what is necessary to bat at the thing without falling over. But who is predicting? Nobody. Both evolution and reorganization work with past data, not predicted future data. The "prediction" is simply the working of Natural Law, whether we know the law or not.

Martin

[Martin Taylor 2019.05.09.14.35]

I'm still not sure I satisfactorily answered my own original question, though. And also, if we can compensate for a predicted disturbance (even if we can update our compensation when the compensation is perturbed), the original reference signal is presumably coming from some kind of internal model, no? Is this cool? I thought we were anti-internal model (except for super sophisticated human cognition where we do clearly imagine things)? This suggests cats have one, no?

I guess your question is about what the pre-stored energy in the cats muscles is for, and whether it implies prediction of an external disturbance. I don't think it does. Remember that reference values change, and that reorganization changes parameters and level-to-level interconnection parameters. Maybe the first time the cat ever tried to bat at something it didn't hit it properly, and maybe it stumbled or fell after the attempt. Nothing of that happened before the cat tried to compensate for a disturbance from outside, so there's no prediction involved. No subsequent unpredicted disturbance is involved in your question.

Everything else is internally generated as changing reference values feed down the levels of control all the way to the individual muscle tension control loops. The feedback effects of the actions generated by the control units whose reference levels were changed are fairly consistent over many attempts at batting at locally moving things (as an observer sees it), so reorganization has the opportunity to build the proper coordinations and sequences of muscle tensions.

Yes, there's an internal model. It's in the reorganized connections and parameter settings for the control loops used to do the batting, and part of that is pre-tensioning the muscles both to perform the bat and to reduce the sudden effect of the action until fast normal low-level control loops correct any remaining perceptual errors. I bet you do the same if you pick up a pile of dishes to put them on a shelf. If you started out with slack muscles, you might fall when you picked them up, and when you released them onto the shelf you might fall in the other direction if all the tension was devoted to holding up the plates, rather than to compensate for the change in balance, which you have encountered every time you picked up or released a weight since you were old enough to try.

It might be evolution, not primarily reorganization in the individual cat, though, because cats are notoriously good at keeping their attitude with respect to gravity, landing with feet toward the ground if falling from a height, and performing remarkable rapid body contortions while remaining stable.

Martin

Hurray, PCT wins! But do any of the readers of these papers realise?

I suspect that the attached (Aruin, 2003) paper goes some way toward reconciling the "invariant, pre-programmed" feedforward conceptualization with the actual phenomenon: namely, it's not really invariant (or, well, it's not invariant in that there is a pre-programmed set of muscles activated in a pre-programmed way). So, I suspect that some of the problem in the "mainstream" is semantic.
Anyway, long (~500 ms) stimulation of motor cortex has already demonstrated that "motor output" seems to be about goals, and not specific, muscle-by-muscle motor programs (see Graziano, 2002; attached), and really the only way you could get that to work is hierarchical control with continuous feedback...

[Martin Taylor 2019.05.09.00.09]

[Rick Marken 2019-05-08_17:57:30]
RM: Some time ago Heather Broccard-Bell asked me what the PCT explanation might be of a behavior that looks like a pretty robust example of anticipatory (or feedforward) control. Here's how Heather described it to me:

HB-B: I was reading about this finding (and perhaps you are already familiar) that, if if you measure muscle activation during behavior, you often see compensatory activation of the axial musculature well before you get activation in the muscles that direct the part of the body that does the thing you're focused on. For example, they trained cats to bat at a target with their forepaw. Well before the limb/forepaw muscles are active, there is activation in the trunk musculature, seemingly anticipating the requisite compensatory stabilization that will become necessary when the forepaw is lifted. I am wondering what a PCT perspective might be on that? What is the relevant controlled variable, since the thing that actually changes sensory input that should activate control architecture hasn't yet happened? Am I missing something? It appears to be a rather robust finding.[emphasis mine--RM]

...

RM: I am posting this to you experts on PCT (with Heather's permission) because I think this is a very interesting discovery that seems to pose a challenge to PCT and I would like to see what you think about this.

Here is section of what I wrote to Heather on this question April 12.

--------

[Here is a] quick and dirty PCT approach. I start with the intention to bat at the target, which would set a reference for a sequence that could crudely be taken as "store energy, set trigger, wait, release energy when required", or in a more general sense "Prepare, wait, execute". You call the storage "prestabilizing", but I think Powers would have called that a "dormitive principle" that just labels what happens without in any way explaining how or why. The "stabilizing" is postural control of the normal kind, the same as would happen if you pushed the cat. If you have seen a video of the "Big Dog" robot, you have seen this kind of stabilization.

The magician Harry Houdini died because he hadn't stored the energy to resist a blow to his abdomen when someone hit him before he was prepared, which he did by tensing (supplying energy to) the stomach muscles. He had offered a general challenge that nobody could hit him there hard enough to hurt him, but the hitter was supposed to do it in a formal kind of way, when he was prepared. When the fatal blow was struck, his muscles had insufficient stored energy to oppose the shock energy in the blow.

I think the actual stabilization is simply the use of some of the pre-tensioned stored energy in normal postural control against a force that would be destabilizing, but instead of the control being entirely to oppose a sensed disturbance, it is part of the output side support of the sequence controller. The "bat" is not simply sending reference values to muscles that move the paw relative to the torso, it is sending reference values to all the postural muscles in a temporal pattern that has been reorganized into the control hierarchy in the same way as a trained athlete such as a ballerina coordinates the timings and strengths of her muscle movements so that what an observer calls "the" action doesn't interfere with her balance.

Anyway, those are my first thoughts on the question, before breakfast.

-------
I haven't considered the problem since then, but I thought I might offer the suggestion.

The saying "Reculer pour mieux sauter" seems to be about the same phenomenon.

Martin

<aruin2003.pdf>
<graziano2002.pdf>

--
Heather C. Broccard-Bell, Ph.D.
Adjunct Assistant Professor
Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems
Nieh Honey Bee Lab
Division of Biological Sciences; Section of Ecology, Behavior, and Evolution
University of California San Diego

619.757.4694

--
Heather C. Broccard-Bell, Ph.D.
Adjunct Assistant Professor
Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems
Nieh Honey Bee Lab
Division of Biological Sciences; Section of Ecology, Behavior, and Evolution
University of California San Diego

619.757.4694

--
Heather C. Broccard-Bell, Ph.D.
Adjunct Assistant Professor
Department of Psychological Sciences

University of San Diego

Visiting Scholar

The role of noise / variability in complex, adaptive systems
Nieh Honey Bee Lab
Division of Biological Sciences; Section of Ecology, Behavior, and Evolution
University of California San Diego

···

On 5/9/19 2:49 PM, Heather Broccard-Bell (<mailto:hebell@ucsd.edu>hebell@ucsd.edu via csgnet Mailing List) wrote:

On Thu, May 9, 2019 at 1:40 PM Martin Taylor <<mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu> wrote:

On 2019/05/9 3:59 PM, Heather Broccard-Bell (<mailto:hebell@ucsd.edu>hebell@ucsd.edu via csgnet Mailing List) wrote:

On Thu, May 9, 2019 at 12:58 PM Heather Broccard-Bell <<mailto:hebell@ucsd.edu>hebell@ucsd.edu> wrote:

On Thu, May 9, 2019 at 12:14 PM Martin Taylor <<mailto:mmt@mmtaylor.net>mmt@mmtaylor.net> wrote:

On 2019/05/9 1:53 PM, Heather Broccard-Bell wrote:

On Thu, May 9, 2019, 07:18 Warren Mansell <<mailto:wmansell@gmail.com>wmansell@gmail.com> wrote:

On 9 May 2019, at 14:28, Heather Broccard-Bell (<mailto:hebell@ucsd.edu>hebell@ucsd.edu via csgnet Mailing List) <<mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu> wrote:

On Wed, May 8, 2019, 21:16 Martin Taylor <<mailto:csgnet@lists.illinois.edu>csgnet@lists.illinois.edu> wrote:

619.757.4694

wmansell · May 10, 2019, 7:52am

Hi folks,

See what you make of this: http://www.pctweb.org/FeedforwardinPCT.pdf

I need to get a full version published one day!

Essentially the rest of psychology continue to use the word ‘prediction’ because it’s the done thing, and can’t let go of the fact that the term doesn’t best describe what they see in their data, theories and mathematical models, which often begin to converge on PCT but ultimately, don’t.

The ‘internal model’ in PCT is purely the functions and parameters that extract and control perceptual variables. Rarely should these approximate to any physical variables in the world. For example, simultaneously controlling the lateral angle, and velocity of the image of a flyball on one’s retina seems to be the best way to get to the right place to catch it. This can be computed from the physical parameters when modelling it on a computer but the brain doesn’t need to do any of this - because it’s world is perceptual, and these two neat perceptual variables dynamically update relative x and y coordinates without the brain having to do ANY computation based on the physical variables that the computer uses.

Other internal models in psychology don’t do this - they assume that the brain works like a VR machine, trying to rebuild all the physical parameters in its model, to have an simulated physics engine in the head.

The other advantage of perceptual control is that you don’t have to perceive the world accurately - in fact perceiving it in a biased way can be more efficient. Max, ccd has found, following Martin, that if in a PCT model you bias the perception of target position by its current velocity so it looks further along its path than it really is, then the model is a more accurate match with human tracking. This looks like anticipation from the outside, but it’s actually controlling for a biased perception that doesn’t currently exist in the world, but will do in a fraction of a second, but the brain doesn’t need to know this.

Please reply as I do want to check Rick, Martin, Bill, Heather, etc, agree?

Cheers,

Warren

···

On 5/9/19 2:49 PM, Heather
Broccard-Bell ( via csgnet Mailing List) wrote:

hebell@ucsd.edu

    I agree, nobody is actively making a prediction,
but the system is tuned to perform better in the world as it as
previously experienced it (as stated), because mostly this works
out – because often the world is kind of consistent. What
makes a prediction different, and what then does it mean to make
a prediction? Does a prediction have to be a fully-formed set
of anticipated data? I think I am not understanding what makes
“prediction” different from “bias based on past experience”.

    I suggest that "bias based on past

experience" is the basis for all prediction with respect to
control system operation. The term prediction also encompasses
things like “the world is going to end in 12 years” which I
think it is reasonable to assume is not experience based.

bill

      On Thu, May 9, 2019 at 1:40 PM
Martin Taylor <csgnet@lists.illinois.edu >
wrote:

[Martin Taylor 2019.05.09.16.13]
          On
2019/05/9 3:59 PM, Heather Broccard-Bell (hebell@ucsd.edu via csgnet
Mailing List) wrote:

              Sorry, that was meant to go to everyone,

not just Martin

                On Thu, May 9, 2019
at 12:58 PM Heather Broccard-Bell <hebell@ucsd.edu >
wrote:

                  All right, so what then is the
difference between operating on a reference signal
from an internal model, and making a prediction?

        If you call the intended result of an action (bringing a
perception toward its reference value) a prediction, I would
say that’s a strange use of “prediction”. better
understanding of what is going on. If I decide to go out and
do go out, have I predicted that I will go out? The Friston
“Free Energy” people say exactly that, probably for the
reasons you suggest below. I would call it at best a
self-fulfilling prediction, one that comes to pass because
of the actions of the one who predicts and wouldn’t come to
pass in the absence of those actions.
                  Does that not amount to the same
thing? After all, the parameters of the internal
model are set through experience (or evolution –
which is some kind of sort of larger
“experience”). So then the system can be biased
in a way that allows it to function better than
ones that aren’t biased that way in a given
environment.

        I could be happy with saying that there is a form of
prediction here, which is a prediction that the Real Reality
world will continue to act much as it has been acting for a
reasonably long time. It’s not a prediction that when the
cat wants to bat at something it will do what is necessary
to bat at the thing without falling over. But who is
predicting? Nobody. Both evolution and reorganization work
with past data, not predicted future data. The “prediction”
is simply the working of Natural Law, whether we know the
law or not.
        Martin





        On Thu, May 9, 2019 at 12:14 PM Martin Taylor <mmt@mmtaylor.net> wrote:
                      [Martin Taylor
2019.05.09.14.35]
                        On
2019/05/9 1:53 PM, Heather Broccard-Bell
wrote:

                          I'm still not sure I
satisfactorily answered my own original
question, though. And also, if we can
compensate for a predicted disturbance
(even if we can update our compensation
when the compensation is perturbed), the
original reference signal is presumably
coming from some kind of internal model,
no? Is this cool? I thought we were
anti-internal model (except for super
sophisticated human cognition where we do
clearly imagine things)? This suggests
cats have one, no?

                      I guess your question is about what the
pre-stored energy in the cats muscles is for,
and whether it implies prediction of an
external disturbance. I don’t think it does.
Remember that reference values change, and
that reorganization changes parameters and
level-to-level interconnection parameters.
Maybe the first time the cat ever tried to bat
at something it didn’t hit it properly, and
maybe it stumbled or fell after the attempt.
Nothing of that happened before the cat tried
to compensate for a disturbance from outside,
so there’s no prediction involved. No
subsequent unpredicted disturbance is involved
in your question.
                      Everything else is internally generated as
changing reference values feed down the levels
of control all the way to the individual
muscle tension control loops. The feedback
effects of the actions generated by the
control units whose reference levels were
changed are fairly consistent over many
attempts at batting at locally moving things
(as an observer sees it), so reorganization
has the opportunity to build the proper
coordinations and sequences of muscle
tensions.
                      Yes, there's an internal model. It's in the
reorganized connections and parameter settings
for the control loops used to do the batting,
and part of that is pre-tensioning the muscles
both to perform the bat and to reduce the
sudden effect of the action until fast normal
low-level control loops correct any remaining
perceptual errors. I bet you do the same if
you pick up a pile of dishes to put them on a
shelf. If you started out with slack muscles,
you might fall when you picked them up, and
when you released them onto the shelf you
might fall in the other direction if all the
tension was devoted to holding up the plates,
rather than to compensate for the change in
balance, which you have encountered every time
you picked up or released a weight since you
were old enough to try.
                      It might be evolution, not primarily
reorganization in the individual cat, though,
because cats are notoriously good at keeping
their attitude with respect to gravity,
landing with feet toward the ground if falling
from a height, and performing remarkable rapid
body contortions while remaining stable.
                      Martin

                                On
Thu, May 9, 2019, 07:18 Warren
Mansell <wmansell@gmail.com >
wrote:

                                    Hurray, PCT wins!

But do any of the readers of
these papers realise?

                                    On 9 May 2019, at 14:28, Heather
Broccard-Bell (hebell@ucsd.edu
via csgnet Mailing List) <csgnet@lists.illinois.edu >
wrote:

                                              I
suspect that the
attached (Aruin, 2003)
paper goes some way
toward reconciling the
“invariant,
pre-programmed”
feedforward
conceptualization with
the actual phenomenon:
namely, it’s not
really invariant (or,
well, it’s not
invariant in that
there is a
pre-programmed set of
muscles activated in a
pre-programmed way).
So, I suspect that
some of the problem in
the “mainstream” is
semantic.

                                                Anyway,
long (~500 ms)
stimulation of motor
cortex has already
demonstrated that
“motor output” seems
to be about goals,
and not specific,
muscle-by-muscle
motor programs (see
Graziano, 2002;
attached), and
really the only way
you could get that
to work is
hierarchical control
with continuous
feedback…

                                                On
Wed, May 8, 2019,
21:16 Martin Taylor
<csgnet@lists.illinois.edu >
wrote:

                                                  [Martin Taylor

2019.05.09.00.09]

                                                      [Rick

Marken
2019-05-08_17:57:30]

                                                      RM: Some
time ago
Heather
Broccard-Bell
asked me what
the PCT
explanation
might be of a
behavior that
looks like a
pretty robust
example of
anticipatory
(or
feedforward)
control.
Here’s how
Heather
described it
to me:

                                                      HB-B:
I was reading
about this
finding (and
perhaps you
are already
familiar)
that, if if
you measure
muscle
activation
during
behavior, you
often see
compensatory
activation of
the axial
musculature
well before
you get
activation in
the muscles
that direct
the part of
the body that
does the thing
you’re focused
on. For
example, they
trained cats
to bat at a
target with
their
forepaw. ** Well
before the
limb/forepaw
muscles are
active, there
is activation
in the trunk
musculature,
seemingly
anticipating
the requisite
compensatory
stabilization
that will
become
necessary when
the forepaw is
lifted. ** I
am wondering
what a PCT
perspective
might be on
that? What is
the relevant
controlled
variable,
since the
thing that
actually
changes
sensory input
that should
activate
control
architecture
hasn’t yet
happened? Am
I missing
something? It
appears to be
a rather
robust
finding.[emphasis
mine–RM]

…

                                                      RM:
I am posting
this to you
experts on PCT
(with
Heather’s
permission)
because I
think this is
a very
interesting
discovery that
seems to pose
a challenge to
PCT and I
would like to
see what you
think about
this.

                                                  Here is  section
of what I wrote to
Heather on this
question April 12.
                                                  --------



                                                  [Here is a] quick
and dirty PCT
approach. I start
with the intention
to bat at the
target, which
would set a
reference for a
sequence that
could crudely be
taken as “store
energy, set
trigger, wait,
release energy
when required”, or
in a more general
sense “Prepare,
wait, execute”.
You call the
storage
“prestabilizing”,
but I think Powers
would have called
that a “dormitive
principle” that
just labels what
happens without in
any way explaining
how or why. The
“stabilizing” is
postural control
of the normal
kind, the same as
would happen if
you pushed the
cat. If you have
seen a video of
the “Big Dog”
robot, you have
seen this kind of
stabilization.
                                                  The magician Harry
Houdini died
because he hadn’t
stored the energy
to resist a blow
to his abdomen
when someone hit
him before he was
prepared, which he
did by tensing
(supplying energy
to) the stomach
muscles. He had
offered a general
challenge that
nobody could hit
him there hard
enough to hurt
him, but the
hitter was
supposed to do it
in a formal kind
of way, when he
was prepared. When
the fatal blow was
struck, his
muscles had
insufficient
stored energy to
oppose the shock
energy in the
blow.
                                                  I think the actual
stabilization is
simply the use of
some of the
pre-tensioned
stored energy in
normal postural
control against a
force that would
be destabilizing,
but instead of the
control being
entirely to oppose
a sensed
disturbance, it is
part of the output
side support of
the sequence
controller. The
“bat” is not
simply sending
reference values
to muscles that
move the paw
relative to the
torso, it is
sending reference
values to all the
postural muscles
in a temporal
pattern that has
been reorganized
into the control
hierarchy in the
same way as a
trained athlete
such as a
ballerina
coordinates the
timings and
strengths of her
muscle movements
so that what an
observer calls
“the” action
doesn’t interfere
with her balance.
                                                  Anyway, those are
my first thoughts
on the question,
before breakfast.
                                                  -------

                                                   I haven't
considered the
problem since
then, but I
thought I might
offer the
suggestion.
                                                  The saying
“Reculer pour
mieux sauter”
seems to be about
the same
phenomenon.
                                                  Martin

<aruin2003.pdf>

<graziano2002.pdf>

–
Heather
C. Broccard-Bell,
Ph.D.

Adjunct Assistant Professor

                                                      Department

of
Psychological
Sciences

                                                      University

of San Diego

                                                    Visiting

Scholar

The

role of noise /
variability in
complex,
adaptive systems

                                                    Nieh

Honey Bee Lab

                                                    Division

of Biological
Sciences;
Section of
Ecology,
Behavior, and
Evolution

                                                    University

of California
San Diego

                                                    619.757.4694

–
Heather
C. Broccard-Bell,
Ph.D.

                                                  Adjunct

Assistant
Professor

                                                  Department

of Psychological
Sciences

                                                  University

of San Diego

                                                Visiting

Scholar

The

role of noise /
variability in
complex, adaptive
systems

                                                Nieh

Honey Bee Lab

                                                Division

of Biological
Sciences; Section of
Ecology, Behavior,
and Evolution

                                                University

of California San
Diego

                                                619.757.4694

–
Heather C.
Broccard-Bell, Ph.D.

                                        Adjunct

Assistant Professor

                                        Department

of Psychological Sciences

                                        University

of San Diego

                                      Visiting

Scholar

                                      The role of

noise / variability in
complex, adaptive systems

                                      Nieh Honey Bee

Lab

                                      Division of
Biological Sciences; Section
of Ecology, Behavior, and
Evolution

                                      University of

California San Diego

                                      619.757.4694

MartinT · May 10, 2019, 2:32pm

[Martin Taylor 2019.05.10.10.08]

Well put. The first paragraph more or less offers the basis for what

I have been trying to get across, less skilfully, in the RREV-CEV
discussion. The RREV is the fact, the CEV is the model. The second paragraph follows the Friston “Free Energy” way of
thinking, which is – I claim – largely embodied in PCT, but which
is somewhat misleading in its wording and in the thinking that leads
to the labels given to its elements. In that context, every action
is based upon a future need for any part of the
hierarchy that has been reorganized. The PCT wording would say that
whatever in the hierarchy is used to reduce the difference between
the perception and its reference value is an available means to
control the perceptual variable. You might say that the need was by evolution or
reorganization if you wanted to be teleological and say that some
agent in the past knew that the cat would at some future time want
to bat at a moving object. I would prefer to say that the useful
part of the hierarchy was in place to be used for “batting” because
of the survival and incremental modification of elements that had
been useful in the past for that and other purposes.
It certainly is possible to construct things because of an
anticipated future need. Teachers of mathematics to children are
trying to do just that. The teacher or whoever prescribed the
curriculum presumably is controlling a perception that at some
future time most children will be able to control some perceptions
better if they have available the tools of mathematics. I would call
that a prediction, but even that might be open to argument.
Martin

···

On 2019/05/9 6:02 PM, Bill Leach
( via csgnet Mailing List) wrote:

wrleach@cableone.net

  ...Indeed, even Martin's example of sequencing involves a model in
that the sequence is a model. In a specific subject that
model was either constructed or inherited but at some point, maybe
eons ago it was constructed. Some form or reorganization was
involved in the construction to correct for uncontrolled or poorly
controlled error.
“Pre-action tensioning” is a fine label for the sequence step
but keep in mind that the whole sequence is based upon anticipating
the need to run the sequence to ensure satisfactory control of
perception including an expected future perception.

anticipating
anticipated

_Erling_Jorgensen · May 10, 2019, 5:38pm

[From Erling Jorgensen (2019.05.10 1243 EDT)]

Rick Marken 2019-05-08_17:57:30

2019/05/9 6:02 PM, Bill Leach…

Confidentiality: * This message is intended only for the addressee, and may contain information that is privileged and confidential under HIPAA, 42CFR Part 2, and/or other applicable State and Federal laws. If you are not the addressee, or the employer or agent responsible for delivering the message to the addressee, any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this in error, please notify the sender immediately and delete the material from your computer. Thank you for your cooperation.*

Please also note: Under 42 CFR part 2 you are prohibited from making any further disclosure of information that identifies an individual as having or having had a substance use disorder unless it is expressly permitted by the written consent of the individual whose information is being disclosed or as otherwise permitted by 42 CFR Part 2.

Martin Taylor 2019.05.10.10.08

BL: “Pre-action tensioning” is a fine label for the sequence step but keep in mind that the whole sequence is based upon anticipating the need to run the sequence to ensure satisfactory control of perception including an expected future perception.

EJ: I believe this is one of the questions at issue. Does a PCT control system, using negative feedback in present time, controlling for a sequence perception or some other type of complicated body configuration, specifically need “anticipation” to carry out its control?

EJ: I also have a concern about the last three words of the above quote. While technically a PCT goal (i.e., reference standard) could be called “an expected future perception,” I believe that pre-decides the question, by leaving open the implication that PCT control has to operate by means of … [fill in the blank]: some call it “anticipation,” some call it “prediction,” some call if “feedforward mechanisms.” There are ways for a PCT hierarchy to handle such processes, but I don’t think we say they are necessary for PCT control.

MT: You might say that the need was anticipated by evolution or reorganization if you wanted to be teleological and say that some agent in the past knew that the cat would at some future time want to bat at a moving object. I would prefer to say that the useful part of the hierarchy was in place to be used for “batting” because of the survival and incremental modification of elements that had been useful in the past for that and other purposes.

EJ: I thought Heather’s original raising of anticipation was in a different context. Here’s the relevant passage, quoted by:

Rick Marken 2019-05-08_17:57:30 –

HB-B: … **Well before the limb/forepaw muscles are active, there is activation in the trunk musculature, seemingly anticipating the requisite compensatory stabilization that will become necessary when the forepaw is lifted. **

EJ: This is the “behavior that looks like a pretty robust example of anticipatory (or feedforward) control.” To my mind, “anticipating” here is not about the processes of either evolution or reorganization in constructing the elementary control systems that may be involved. It seems rather to be about producing specific values within those control systems during a specific situation, values that show themselves as changes of perception, reference, or output.

EJ: While I have only skimmed the research papers that both Rick & Heather attached, the ‘pre-action tensioning’ (as I’ll continue to call the phenomenon) seems at odds with a prediction of the PCT model. Rick expressed the issue well:

RM: This finding certainly looks like it is inconsistent with the predictions of PCT since it looks like the cat is generating an output (activation of the trunk musculature) open loop in anticipation of a future disturbance (the change in orientation of the body when the forepaw is lifted, a change in orientation that would cause the cat to lose its balance if the trunk were not in the proper position).

EJ: I don’t want PCT to prematurely surrender the field, by saying, yes of course this has to be feedforward. Does it?? Henry Yin has published some pretty strong arguments that “posture control” is a robust negative feedback control process, readily compatible with PCT as it stands. How much of this phenomenon can the current model of PCT handle?

All the best,

Erling