Levels of perception (Re: PCT researcher who doesn't talk)

[From Rick Marken (2009.02.22.1400)]

Bill Powers (2009.02.21.1617 mst)]

I don't think so. Remember that the quantity actually plotted is not the
perceptual signal, but a very compressed function of its magnitude relative
to the noise level. With d'^2 = 9, you say the probability of an error is 3
in 1000, which is pretty close to asymptote on a linear scale and is just
above the highest recorded point at d^2 = 8.

Your comments about this experiment being similar to my studies of
hierarchical perception led me to think about what might be going on
here in comparison to what subjects are doing in my experiments
(which, for those interested, are up on the net
at:http://www.mindreadings.com/ControlDemo/HP.html). To the extent
that I correctly understand Martin's experiment, it sounds like on
each trial there are three beeps and, at some time before the third
one of two buttons is lit and the subject is to press that button when
the third beep occurs. The x axis on the data chart shows (I think)
the lead time (in msec) of the button light relative to the last tone.
Based on the graph, it looks like subjects are doing this task nearly
perfectly (at least in terms of pushing the correct button is not also
timing their push to match the last beep) when the button light leads
the last tone by 300 msec. Is that right? If so, it looks like
subjects are probably controlling whatever variable they are
controlling when the light comes on at least 300 msec before the last
tone.

I better stop there for now and make sure my concept of what's going
on in this experiment is correct.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.02.22.20.57]


[From Rick Marken (2009.02.22.1400)]
To
Bill Powers (2009.02.21.1617 mst)]

To the extent
that I correctly understand Martin's experiment, it sounds like on
each trial there are three beeps and, at some time before the third
one of two buttons is lit and the subject is to press that button when
the third beep occurs. The x axis on the data chart shows (I think)
the lead time (in msec) of the button light relative to the last tone.

Correct.

Based on the graph, it looks like subjects are doing this task nearly
perfectly (at least in terms of pushing the correct button is not also
timing their push to match the last beep) when the button light leads
the last tone by 300 msec. Is that right?

Depends what you mean by “nearly perfectly”. Is 99% correct “nearly
perfectly”, if so, then you are correct.

If so, it looks like
subjects are probably controlling whatever variable they are
controlling when the light comes on at least 300 msec before the last
tone.

I think they are controlling it no matter when the light comes on. The
only difference the timing of the light makes is the amount of
information available for selecting which button to push.

I better stop there for now and make sure my concept of what's going
on in this experiment is correct.

Your description is correct, so far as it goes. Whether your
interpretation is, that’s yet to be determined by experiment, theory,
or consistency with accepted interpretations of related control
behaviours. If you have an interpretation different from the one
already proposed, then let’s see if we can devise an experiment to
distinguish them.

Martin

[From Rick Marken (2009.02.22.2140)]

Martin Taylor (2009.02.22.20.57)

Rick Marken (2009.02.22.1400)--

To the extent
that I correctly understand Martin's experiment, it sounds like on
each trial there are three beeps and, at some time before the third
one of two buttons is lit and the subject is to press that button when
the third beep occurs. The x axis on the data chart shows (I think)
the lead time (in msec) of the button light relative to the last tone.

Correct.

OK, thanks. So on every trial when the subject makes a correct
response (pushes the lighted button) the sequence of events before
that is like this:

bip ...bip...light....bip/push correct button

Is that right? Or can the light come on at any point before the final
bip? Also, is the time between bips also variable?

Based on the graph, it looks like subjects are doing this task nearly

>perfectly (at least in terms of pushing the correct button is not also

timing their push to match the last beep) when the button light leads
the last tone by 300 msec. Is that right?

Depends what you mean by "nearly perfectly". Is 99% correct "nearly
perfectly", if so, then you are correct.

Yes, 99% is actually perfect rom my permissive perspective.

If so, it looks like
subjects are probably controlling whatever variable they are
controlling when the light comes on at least 300 msec before the last
tone.

I think they are controlling it no matter when the light comes on.

I don't understand this. If they are making lots of errors, as is the
case when the light comes on only shortly before the 3rd bip, then I
don't see how they could be considered to be in control of anything,
unless making errors is what they intend. But that's not important
right now. We can talk about that once I am pretty sure I know what
actually was happening in the experiment.

Your description is correct, so far as it goes. Whether your interpretation
is, that's yet to be determined by experiment, theory, or consistency with
accepted interpretations of related control behaviours.

I didn't think I had given an interpretation yet, though perhaps you
see my mention of a controlled variable as an interpretation (which it
is). But I will give a more detailed interpretation once I know what
was actually going on in the experiment and I will (I hope) propose
further experiments to test it.

If you have an
interpretation different from the one already proposed, then let's see if we
can devise an experiment to distinguish them.

I guess I'm not quite sure what interpretation was already proposed.
But I will eventually propose one, along with some proposed tests. If
you tell me what the current interpretation is I will try to develop a
test that can distinguish mine from that one, if they are different.

Best

Rick

···

---
Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.02.23.15.07

[From Rick Marken (2009.02.22.2140)]
Martin Taylor (2009.02.22.20.57)
Rick Marken (2009.02.22.1400)--
...

So on every trial when the subject makes a correct
response (pushes the lighted button) the sequence of events before
that is like this:
bip ...bip...light....bip/push correct button
Is that right? Or can the light come on at any point before the final
bip? Also, is the time between bips also variable?

In principle, the light could come on at any time at all, but the data
wouldn’t be very useful if it frequently came on so early that subjects
aways got the right answer or always purely guessed, so in practice
thre wouldn’t be much point in making the delay between “light-on” and
“third bip” less than 150 msec or more than 350 msec or so.

As for your second question, I can only guess, but my guess is that the
bip sequence would have come on at regular intervals, such as every 2
seconds. Doing that would obviate the need for the kind of warning
signal I mentioned in some earlier messages.


Based on the graph, it looks like subjects are doing this task nearly
perfectly (at least in terms of pushing the correct button is not also
timing their push to match the last beep) when the button light leads
the last tone by 300 msec. Is that right?
Depends what you mean by "nearly perfectly". Is 99% correct "nearly
perfectly", if so, then you are correct.
Yes, 99% is actually perfect rom my permissive perspective.

OK, then yes, nearly perfectly at about 350 msec. (If my calculations
are correct, d’^2 = 8 means about 98.4% correct).


If so, it looks like
subjects are probably controlling whatever variable they are
controlling when the light comes on at least 300 msec before the last
tone.
I think they are controlling it no matter when the light comes on.
I don't understand this. If they are making lots of errors, as is the
case when the light comes on only shortly before the 3rd bip, then I
don't see how they could be considered to be in control of anything,
unless making errors is what they intend.

“Controlling” a perception is not the same as “being in control of” the
perception. Lots of times we do the best we can, without it being as
good as we would like it to be. We are “controlling”, without “being in
control” of the peception in question.

If you have an
interpretation different from the one already proposed, then let's see if we
can devise an experiment to distinguish them.
I guess I'm not quite sure what interpretation was already proposed.
But I will eventually propose one, along with some proposed tests. If
you tell me what the current interpretation is I will try to develop a
test that can distinguish mine from that one, if they are different.

The consensus interpretation that Bill and I arrived at is that two key
control systems are involved in each trial (The upper one is Bill’s
“V”). A high-level (I say category) system controls for a match between
the presentation category (left light or right light) and an answer
(left or right). In time for the third bip (which is why there are
three bips rather than one), the currently selected answer is fed as a
reference value to a loop that controls a perception of which button is
pushed. There’s not a lot more to it than that, though of course in a
full description one would have to include control systems for matching
bip-time to button-push time, as well as the layered protocol dialogues
between Experimenter and Subject by which the Subject discovers what
the Experimenter wants done.

For a single trial, the diagram is something like this (this version
has been augmented to cover the more general case, in which there are
several possible answering modalities):

ButtonPresentationMatch_v22.jpg

Maybe you don’t call this an “interpretation”, since it’s only a sketch
of a circuit diagram, but it’s the best we’ve got at the moment. The
red arrows mark the component being investigated.

Martin

[From Rick Marken (2009.02.23.2320)]

Martin Taylor (2009.02.23.15.07)--

The consensus interpretation that Bill and I arrived at is that two key
control systems are involved in each trial (The upper one is Bill's "V"). A
high-level (I say category) system controls for a match between the
presentation category (left light or right light) and an answer (left or
right). In time for the third bip (which is why there are three bips rather
than one), the currently selected answer is fed as a reference value to a
loop that controls a perception of which button is pushed. There's not a lot
more to it than that, though of course in a full description one would have
to include control systems for matching bip-time to button-push time, as
well as the layered protocol dialogues between Experimenter and Subject by
which the Subject discovers what the Experimenter wants done...

Maybe you don't call this an "interpretation", since it's only a sketch of a
circuit diagram, but it's the best we've got at the moment. The red arrows
mark the component being investigated.

Yes, this "interpretation" is exactly the same as mine. I've attached
your graph. This is exactly my interpretation of what is happening in
the experiment. The only changes I would make are basically cosmetic.
For example, I would put the E (which I assume means "Environment"
between the outputs (like the button) and the presentation (which I
take to be the entire presentation: bips, light and press). I would
also change the label of the "Sense" box to "Controlled quantity",
which is the environmental variable that corresponds to the controlled
perception (which is the output of the box labeled "Interpret and
compare"). I see the "Sense" box as the controlled quantity because it
is in the role of an environmental variable that is made up of the
bips and lights that are the presentation (P) and the presses and
results of those presses that are the outputs of the control system. I
would also re-label the "Interpret and compare" box as "Perception of
state of the relationship between presentation and answer" because the
output of this box is a perception that is being compared to a
reference for a "match between presentation and answer". I take
"match" to be the intended state of the relationship between
presentation and answer.

But other than these little nits, this is precisely the model of the
subjects' situation in this experiment that I would use. In fact, your
nice model has encouraged me to make a similar model of the situation
that exists in my Hierarchical Control demo. My model of that
situation is attached. The "Presentation" in this case is the computer
presentation of different sized squares and circles moving clockwise
or counter clockwise in a sequence-- large, medium small (L, M, S) or
small, medium large (S,M,L) -- at different rates (specified when the
experiment is started). Only one output affects the state of this
presentation: a mouse click. Then the mouse is clicked, the
presentation changes from square to circle or vice versa depending on
the shape currently displayed; it also changes from clockwise to
counter clockwise movement or vice versa depending on the direction of
motion currently displayed; and it also changes the sequence to S, M,
L or vice versa depending on the sequence currently displayed. The
diagram shows that the subject can control any one of these
perceptions (configuration, transition or sequence) keeping it in an
intended state (shown by the reference for that perception; I chose to
show a reference of "square" for configuration; "clockwise" for
transition and "L,M,S" for sequence.

The point of the demo is that when the presentation rate is fast, all
that can be perceived (and, hence, controlled, is the lowest level
perception (shape). Slow the presentation down a but and you can
perceive (and control) transition; slow it down even more and you can
perceive (and control) sequence. All these variables were present all
the time (as controlled quantities) but they become controllable
perceptions only when they become perceptible. Under the assumption
that higher level perceptions have slower temporal characteristics,
the slower the rate of presentation needed to perceive and control a
variable, the high level the variable. I measure how well a perception
can be controlled by measuring the proportion of a trial that the
perception is kept in the reference state. In order to keep a
perception in the reference state the subject must press the mouse
when the computer changes the state of the controlled variable. For
example, at random time the computer changes the shape presented from
square to circle; the subject can change the shape back to square by
pressing the mouse. Of course, in order to know when to do this the
subject has to be able to perceive the state of the shape variable.
And the subject perceive the state of the shape variable when the
stimuli are presented at the fastest rate. But at the fastest rate
it's hard do perceive (and, thus control) the direction of movement
(transition) and its impossible to perceive and control the sequence
at the highest rate; but when you slow it down, you can control it
pretty well.

My program keeps track not only of the proportion of a trial that a
variable is kept in the reference state but it also measures how long
it takes between the time the controlled quantity changes state (from
square to circle) and the subject presses the mouse. This is a measure
of reaction time. It turns out that the reaction time when controlling
for configuration is about 200 msec. For transition it's about 400 and
for sequence, 600 msec. These are only approximate and I have a little
concern about the timing in the java programs but I think the times
are certainly in the ballpark. These reactions times can be considered
the transport lag from input to output in each of the loops shown in
my model. Based on these times and the fact that in your experiment
some of the reaction time (which was about 300msec I believe) involved
the time to move a finger from one point to another (tie that is not
part of the transport lag in the loop) I would say that the perception
being controlled in your experiment is a configuration, which makes
sense (I think) because what the person has to control is
audio/visual/tactile "shape". But that would have to be tested, of
course. I'll get back to you with some suggested tests if you're
interested.

Best

Rick

MartinModel.jpg

···

--
Richard S. Marken PhD
rsmarken@gmail.com

Content-Type: image/jpeg; name="MartinModel.jpg"
Content-Disposition: attachment; filename="MartinModel.jpg"
X-Attachment-Id: f_frk6s1xh0

Opps, forgot to attach my diagram. Here it is.

Best

Rick

ThreeLevels.jpg

···

On Mon, Feb 23, 2009 at 11:18 PM, Richard Marken <rsmarken@gmail.com> wrote:

[From Rick Marken (2009.02.23.2320)]

Martin Taylor (2009.02.23.15.07)--

The consensus interpretation that Bill and I arrived at is that two key
control systems are involved in each trial (The upper one is Bill's "V"). A
high-level (I say category) system controls for a match between the
presentation category (left light or right light) and an answer (left or
right). In time for the third bip (which is why there are three bips rather
than one), the currently selected answer is fed as a reference value to a
loop that controls a perception of which button is pushed. There's not a lot
more to it than that, though of course in a full description one would have
to include control systems for matching bip-time to button-push time, as
well as the layered protocol dialogues between Experimenter and Subject by
which the Subject discovers what the Experimenter wants done...

Maybe you don't call this an "interpretation", since it's only a sketch of a
circuit diagram, but it's the best we've got at the moment. The red arrows
mark the component being investigated.

Yes, this "interpretation" is exactly the same as mine. I've attached
your graph. This is exactly my interpretation of what is happening in
the experiment. The only changes I would make are basically cosmetic.
For example, I would put the E (which I assume means "Environment"
between the outputs (like the button) and the presentation (which I
take to be the entire presentation: bips, light and press). I would
also change the label of the "Sense" box to "Controlled quantity",
which is the environmental variable that corresponds to the controlled
perception (which is the output of the box labeled "Interpret and
compare"). I see the "Sense" box as the controlled quantity because it
is in the role of an environmental variable that is made up of the
bips and lights that are the presentation (P) and the presses and
results of those presses that are the outputs of the control system. I
would also re-label the "Interpret and compare" box as "Perception of
state of the relationship between presentation and answer" because the
output of this box is a perception that is being compared to a
reference for a "match between presentation and answer". I take
"match" to be the intended state of the relationship between
presentation and answer.

But other than these little nits, this is precisely the model of the
subjects' situation in this experiment that I would use. In fact, your
nice model has encouraged me to make a similar model of the situation
that exists in my Hierarchical Control demo. My model of that
situation is attached. The "Presentation" in this case is the computer
presentation of different sized squares and circles moving clockwise
or counter clockwise in a sequence-- large, medium small (L, M, S) or
small, medium large (S,M,L) -- at different rates (specified when the
experiment is started). Only one output affects the state of this
presentation: a mouse click. Then the mouse is clicked, the
presentation changes from square to circle or vice versa depending on
the shape currently displayed; it also changes from clockwise to
counter clockwise movement or vice versa depending on the direction of
motion currently displayed; and it also changes the sequence to S, M,
L or vice versa depending on the sequence currently displayed. The
diagram shows that the subject can control any one of these
perceptions (configuration, transition or sequence) keeping it in an
intended state (shown by the reference for that perception; I chose to
show a reference of "square" for configuration; "clockwise" for
transition and "L,M,S" for sequence.

The point of the demo is that when the presentation rate is fast, all
that can be perceived (and, hence, controlled, is the lowest level
perception (shape). Slow the presentation down a but and you can
perceive (and control) transition; slow it down even more and you can
perceive (and control) sequence. All these variables were present all
the time (as controlled quantities) but they become controllable
perceptions only when they become perceptible. Under the assumption
that higher level perceptions have slower temporal characteristics,
the slower the rate of presentation needed to perceive and control a
variable, the high level the variable. I measure how well a perception
can be controlled by measuring the proportion of a trial that the
perception is kept in the reference state. In order to keep a
perception in the reference state the subject must press the mouse
when the computer changes the state of the controlled variable. For
example, at random time the computer changes the shape presented from
square to circle; the subject can change the shape back to square by
pressing the mouse. Of course, in order to know when to do this the
subject has to be able to perceive the state of the shape variable.
And the subject perceive the state of the shape variable when the
stimuli are presented at the fastest rate. But at the fastest rate
it's hard do perceive (and, thus control) the direction of movement
(transition) and its impossible to perceive and control the sequence
at the highest rate; but when you slow it down, you can control it
pretty well.

My program keeps track not only of the proportion of a trial that a
variable is kept in the reference state but it also measures how long
it takes between the time the controlled quantity changes state (from
square to circle) and the subject presses the mouse. This is a measure
of reaction time. It turns out that the reaction time when controlling
for configuration is about 200 msec. For transition it's about 400 and
for sequence, 600 msec. These are only approximate and I have a little
concern about the timing in the java programs but I think the times
are certainly in the ballpark. These reactions times can be considered
the transport lag from input to output in each of the loops shown in
my model. Based on these times and the fact that in your experiment
some of the reaction time (which was about 300msec I believe) involved
the time to move a finger from one point to another (tie that is not
part of the transport lag in the loop) I would say that the perception
being controlled in your experiment is a configuration, which makes
sense (I think) because what the person has to control is
audio/visual/tactile "shape". But that would have to be tested, of
course. I'll get back to you with some suggested tests if you're
interested.

Best

Rick

--
Richard S. Marken PhD
rsmarken@gmail.com

--
Richard S. Marken PhD
rsmarken@gmail.com

Content-Type: image/jpeg; name="ThreeLevels.jpg"
Content-Disposition: attachment; filename="ThreeLevels.jpg"
X-Attachment-Id: f_frk8zas91

[From Rick Marken (2009.02.24.1010)]

I made a mistake (that I noticed; there may be others) in my last
post. I stated that reaction time was measured in the Schouten
experiment that Martin presented:

Rick Marken (2009.02.23.2320)--

Based on these times and the fact that in your experiment
some of the reaction time (which was about 300msec I believe)

Actually no reaction time was measured in that experiment. The 300
msec figure is actually the amount of time by which the button light
onset precedes the last bip that allows for a d' score that I consier
large enough to count as "good control" (about 90% correct). So this
300 msec figure actually corresponds to the frame rate (slow, medium
or fast) in my Hierarchical Control demo that allows good control of a
particular perception (90% of the time the perception is in the
reference state). A frame rate of 300 msec is near the fastest rate
that allows a subject to have good control and it turns out to be a
bit faster than the rate needed to control a sequence (500 msec/frame
in my demo) and a bit slower than that needed to control a
configuration (150 msec/frame). If we take both these time
measurements and Powers' definitions of the types of perceptions
controlled at each level of control seriously, then the type of
perceptual variable controlled in the Schouten experiment is between
the configuration and sequence levels, which, as I recall, is an event
perception. So the perception of

bip...bip..light....bip/press

is an event that is at the reference level when the press occurs
nearly simultaneously with the third bip and it is a press of the
lighted button. In order to get that event to happen consistently
(keep it under control) the components of the event must unfold at the
rate of about 300 msec/component. At least, that's my hypothesis based
on my data and the data from the Schouten study.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[From Bill Powers (2009.02.24.1116 MST)]

Rick Marken (2009.02.23.2320) --

Yes, this "interpretation" is exactly the same as mine. I've attached
your graph. This is exactly my interpretation of what is happening in
the experiment. The only changes I would make are basically cosmetic.
For example, I would put the E (which I assume means "Environment"

Actually it means "experimenter". In the original diagrams there were
two panels, the left one being the Experimenter and the right one
being the Subject. Rather than redraw the Experimenter panel, Martin
just drew some arrows going to E and back.

But other than these little nits, this is precisely the model of the
subjects' situation in this experiment that I would use. In fact, your
nice model has encouraged me to make a similar model of the situation
that exists in my Hierarchical Control demo.

I suggest drawing this so the higher-order perceptions are built on
signals coming fromt the lower-order perceptions rather than directly
from the environment. It's still a lovely demo.

Best,

Bill P.

No virus found in this outgoing message.
Checked by AVG - www.avg.com
Version: 8.0.237 / Virus Database: 270.11.3/1969 - Release Date: 02/24/09 06:43:00

[From Rick Marken (2009.02.24.1040)]

Bill Powers (2009.02.24.1116 MST)--

I suggest drawing this so the higher-order perceptions are built on signals
coming fromt the lower-order perceptions rather than directly from the
environment.

Yes, I didn't want to get too theoretical (or complex, since each
higher order perceptual function is probably fed by more than just the
one lower order signal shown in the diagram). But I think it's fair to
add paths showing the perceptual signals coming out of the lower order
functions also serving as inputs to the higher order perceptual
functions.

It's still a lovely demo.

Thanks. I was going to re-write it so that it would be possible for
the participant to vary the rate continuously using the mouse. There
are all kinds of other improvements that I would like to make; now
that I revisit it more carefully after all these years I see that it
might be worth investing some time in it.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[From Rick Marken (2009.02.25.1930)]

Bill Powers (2009.02.24.1116 MST)--

Rick Marken (2009.02.23.2320) --

For example, I would put the E (which I assume means "Environment"

Actually it means "experimenter". In the original diagrams there were two
panels, the left one being the Experimenter and the right one being the
Subject. Rather than redraw the Experimenter panel, Martin just drew some
arrows going to E and back.

By the way, where did Martin go? Did I shock him into silence with my
agreement about his (and your) interpretation of the Schouten
experiment? Or is it all settled now, whatever it was that was
unsettled?

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[From Bill Powers (2009;.02.25.2048 MST)]

Rick Marken (2009.02.25.1930) --

By the way, where did Martin go? Did I shock him into silence with my
agreement about his (and your) interpretation of the Schouten
experiment? Or is it all settled now, whatever it was that was
unsettled?

I just now replied to a direct post from Martin last night about
details of the TrackAnalyze program, which he's learning to modify
using Turbo Delphi 7 Personal Edition, which is free. Most of the
discussion is on CSGnet where you see what I see. I'm glad we're all
in reasonable agreement about the Schouten experiment -- no reason
not to be. By the way, the beeps are indeed an event by my
definition, and so is the button press, and there's a relationship of
those two events, and as Martin says probably a Category perception
-- though that one is a little fuzzy. "Left" and "Right" are surely categories.

I have spent a large part of the last two days going slowly mad while
my desktop computer exhibits symptoms that only Bill Gates could
love. What a load of trash Windows XP is, when you burrow down into
it a little way. Did you know that there is a program called CHKDSK
that can run from "Start|Run|chkdsk /f" to check the integrity of the
hard disk in three rather long steps? And did you know that by going
through My Computer, right clicking on drive C, then Properties,
Tools, and Check Now, you can run the very same program, only with
five long steps that now include checking the file integrity, which
the other way doesn't check?

You can also run a program called SFC from the Start|Run window,
which checks the integrity of the Windows programs themselves against
a CD Rom if you have it (I have it for one computer but it didn't
come with the other). Oh yeah -- you have to type in "SFC /scannow"
to get that program actually to do anything. If you just type SFC,
the computer doesn't know you want it to do something, and it doesn't
ask what you want it to do.

That's only a little bit of it.Right now my desktop is running fine
again, except that it can't connect to the internet because it can't,
and I quote, "start the Windows Firewall/Internet Connection Sharing
(ICS) service," which apparently it can't do because "WSA Startup
failed." That could very well be because of, or responsible for, the
fact that "Windows socket initialization failed."

If my productivity drops for a while, you'll know what it's being used for.

Best,

Bill P.

No virus found in this outgoing message.
Checked by AVG - http://www.avg.com
Version: 8.0.173 / Virus Database: 270.11.3/1971 - Release Date: 2/25/2009 6:40 AM

[Martin Taylor 2009.02.25.23.00]

[From Rick Marken (2009.02.25.1930)]
Bill Powers (2009.02.24.1116 MST)--
Rick Marken (2009.02.23.2320) --

For example, I would put the E (which I assume means "Environment"
Actually it means "experimenter". In the original diagrams there were two
panels, the left one being the Experimenter and the right one being the
Subject. Rather than redraw the Experimenter panel, Martin just drew some
arrows going to E and back.
By the way, where did Martin go? Did I shock him into silence with my
agreement about his (and your) interpretation of the Schouten
experiment? Or is it all settled now, whatever it was that was
unsettled?

I do have another life, outside of CSGnet – several, in fact :slight_smile: One
of them is trying to understand Turbo Delphi so as to work with Bill on
modelling tracking with different kinds of distubance. It’s slow work,
for me.

I hadn’t planned to comment on your message, because it seemed
basically just fine, despite one or two – as you call them – “nits”
that I might well have picked in the appropriate season. However, if
you want comment, I’ll do what I can.

Since I’m copy-pasting from a message [From Rick Marken
(2009.02.23.2320)] other than the one to which I’m replying, I have to
use [Rick][Martin] quoting, rather than the automatic form, so bear
with me on that. The figure being referenced is this, from [Martin
Taylor 2009.02.23.15.07]:

ButtonPresentationMatch_v22.jpg

[Rick]


I would also change the label of the "Sense" box to "Controlled quantity", which is the environmental variable that corresponds to the controlled perception (which is the output of the box labeled "Interpret and compare"). I see the "Sense" box as the controlled quantity because it is in the role of an environmental variable that is made up of the bips and lights that are the presentation (P) and the presses and results of those presses that are the outputs of the control system.
[Martin]
The "Presentation" intended in the diagram is not the whole complex, but the lights alone. "Interpret and Compare" compares the perception of which light is lit with the imagined/remembered answers initially provided by the experimneter's instructions. It is called "Interpret" because at that level it's not a light that is being perceived, but a category of location. The controlled variable is the match between that perceived location and the answer. The perceptual output of "Sense" in this diagram is not a controlled quantity. To be more complete, I guess the diagram should have two "Sense" boxes, one for each light, and a "Difference" box that output the difference between the outputs of the two "Sense" boxes. But the perceptual signal that would be the output of the "Difference" box is not a controlled quantity, either. It would provide one of the inputs to "Interpret and Compare", a two-input perceptual function whose perceptual signal is a controlled quantity.
The control that relates to the bips is not represented in the diagram. It's interesting in a quite different context that we haven't discussed, but it's not interesting in the control of the display of the selected answer to the experimenter.
[Rick]
I would also re-label the "Interpret and compare" box as "Perception of state of the relationship between presentation and answer" because the output of this box is a perception that is being compared to a reference for a "match between presentation and answer". I take "match" to be the intended state of the relationship between presentation and answer.

[Martin]

I’ll buy that. It’s a better label, if rather long for normal
conversation or for use in a diagram.

[Rick]

In fact, your nice model has encouraged me to make a similar model of
the situation that exists in my Hierarchical Control demo…My program
keeps track not only of the proportion of a trial that a variable is
kept in the reference state but it also measures how long it takes
between the time the controlled quantity changes state (from square to
circle) and the subject presses the mouse. This is a measure of
reaction time. It turns out that the reaction time when controlling for
configuration is about 200 msec. For transition it’s about 400 and for
sequence, 600 msec.

[Martin]

It would be interesting if you could instantiate your sketch diagram in
software. Also, if you have your old paper and my comments of September
1991, it might still be worth trying to rewrite it so that it doesn’t
read as though it was aimed at PCT cognoscenti. It’s a good demo, and
worth developing.

[Rick]


Based on these times and the fact that in your experiment some of the reaction time (which was about 300msec I believe) involved
the time to move a finger from one point to another ...
[Martin]
No, I think there was a finger on each button, but that wouldn't matter for the analysis.
[Rick]
... (tie that is not part of the transport lag in the loop) I would say that the perception being controlled in your experiment is a configuration, which makes sense (I think) because what the person has to control is audio/visual/tactile "shape".

[Martin]

You have a problem trying to associate the values in your experiment
with those in Schouten’s. Yours is ongoing, and the question is to be
able to perceive changes in an ongoing pattern of change at several
levels. Transport lag shouldn’t affect the ability of a person to
perceive changes at any of the levels, though it would affect the
bandwidth available for good control. In the Schouten experiment, the
presentation is a one-shot event, and transport lag effects the timing
of the button push relative to the perception. There also might be a
transport lag between presentation and interpretation (which would be
the case in your experiment as well), and we can’t separate the two.
I’m not sure how one could cleanly make a comparison between his
timings and yours. Perhaps it can be done, but it’s not obvious to me.

I’m not clear why you say that the controlled variable is a
configuration. It looks to me like a difference between categories (the
provided answers and the perception of which light). One of the
categories is in imagination/memory, the other is computed from sensory
input.

Now back to work.

Martin

[From Rick Marken (2009.02.26.0920)]

Martin Taylor (2009.02.25.23.00)--

I do have another life, outside of CSGnet -- several, in fact :slight_smile:

You're lucky. This is all I've got;-)

[Martin]
The "Presentation" intended in the diagram is not the whole complex, but the
lights alone.

But it is combined (via E) with the subject's output and enters the
perceptual function that turns it into the perception that is
controlled, which must include the tones and the button press, because
that is what the subject does seem to control (when the light precedes
the final tone by enough time).

"Interpret and Compare" compares the perception of which light
is lit with the imagined/remembered answers initially provided by the
experimneter's instructions.

So it's a comparator and not a perceptual function? If so, then there
should be a reference arrow entering that box, shouldn't there?

It is called "Interpret" because at that level
it's not a light that is being perceived, but a category of location.

So you think of it as both a comparator and a perceptual function?
Given it's place in the diagram I think it must be just a perceptual
function. I think it's perceiving the state of the whole event: bip...
bip ...light...bip/press and comparing the state of this event to the
reference for it. The reference specifies that that this perception
should be: bip bip (left light) bip/(left press) or bip bip (right
light) bip/(right press); the perception of bip bip (left light)
bip/(right press) or bip bip (right light) bip/(left press) would be
an error so the output of the comparison has to be an action (right or
left press) that produces a perception that matches the reference. At
least, that's how I would model it.

The
controlled variable is the match between that perceived location and the
answer.

As per what I said above, II would say that the controlled variables
is the state of the event bip... bip ...light...bip/press. I would say
that the _match_ between light (left or right button) and answer
(right or left press) is a _state_ of that variable -- the reference
state in this case. By the way, the reference state could just has
easily been a non-match between light location and press location; the
subject could have been asked to press the left button when the right
light came on, and vise versa; same controlled variable, different
reference state.

The perceptual output of "Sense" in this diagram is not a controlled
quantity.

I was actually suggesting that the box labeled "sense" would make more
"sense" to me if it represented an environmental variable; the event
variable -- bip...bip...light..bip/press -- that the subject is asked
to control. The "output" of this box would then be the state of an
environmental, not a perceptual variable.

[Rick]
I would also re-label the "Interpret and compare" box as "Perception of
state of the relationship between presentation and answe

[Martin]
I'll buy that. It's a better label, if rather long for normal conversation
or for use in a diagram.

Good, that makes a lot more sense to me.

You have a problem trying to associate the values in your experiment with
those in Schouten's. Yours is ongoing, and the question is to be able to
perceive changes in an ongoing pattern of change at several levels.

I did have a problem -- because I was thinking that reaction times
were being measured in the Schouten study. I corrected myself in a
subsequent post. What makes the Schouten study similar to my demo is
the fact that the delay between light and third bip is equivalent to
the frame rate in my study. When the light/third bip delay is long it
is equivalent to the slower frame rates in my study. When light/third
bip delay time is long enough in the Schouten study (about 350msec)
the subjects can do what the subjects in my study can do when the
frame rate is slow enough; they can control the perception of
bip..bip...light...bip/press, keeping it in the reference state.

I could implement the Schouten study in the form of my demo by
presenting the event to be controlled at different rates and changing
the state of the event (disturbing it) occasionally and seeing if the
subject can control the event (by pressing the mouse to restore the
event to the reference state) at that rate. For example, I could
present the event bip bip (left light) bip/(left press) and bip bip
(right light) bip/(right press) repeatedly in sequence at varying
rates, say 50, 100, 200, 400 msec/frame (I would substitute a visual
analog of left and right pressing, such as a box appearing next to the
"light"). At some point I would shift the display to bip bip (left
light) bip/(right press) and bip bip (right light) bip/(left press).
Whenever the subject presses the mouse button the even toggles to the
other state so if the mouse is pressed after the change the event will
return to the "correct" state (as in my demo). At too fast a rate (I
suspect any rate below 200msec/frame) the subject will not be able to
perceive the state of this event and will not know when it has
changed. Thus my measure of control (time event is in reference state
out of total time) will be low for fast frame rates; when the frame
rate is slow enough the subject should be able to control the event
nearly perfectly.

I'm not sure how one could cleanly make a comparison
between his timings and yours. Perhaps it can be done, but it's not obvious
to me.

I hope my explanation above helps. If you do my demo and try to
control the sequence (L,M,S) at the slow rate I think you'll see what
I'm talking about. Think of the sequence of events in the Schouten
study as analogous to the sequence of objects of different sizes in my
demo. The main difference between my study and Schouten's is this:
inthe Schouten study the subject had to complete the event (by pushing
the "correct" button); in my study the computer produces the "correct"
or "incorrect" event and it's up to the subject to "correct" the
computer by pushing the mouse button when necessary.

I'm not clear why you say that the controlled variable is a configuration.

Actually, I'm guessing that it's more like an event than a
configuration and I am saying this based only on the relative timing
in my demo; it takes about a 250 msec/frame rate to control what seems
like a transition (movement clockwise or counterclockwise) and about a
500msec/frame rate to control a sequence. So it seems like the
variable controlled in the Schouten study would be intermediate
between these perceptual types, which, I believe, puts it at the event
level. But to find out what is actually being controlled in the
Schouten study we would, of course, have to test for the controlled
variable in some way.

Now back to work.

Now back to relaxing;-)

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.02.26.17.51]

[From Rick Marken (2009.02.26.0920)]

I’m sorry, but I don’t have time for detailed comments at the moment.
Maybe later, if it seems warranted. Much of what I would do in detail
can be encapsulated in two comments:

 (1) Most of the time, I think of perceptual functions as having a

single scalar output, and

 (2) I assume most perceptual functions have two or more inputs,

even though quite often we draw control diagrams in which only one
input is shown to a perceptual function.


Martin Taylor (2009.02.25.23.00)--

[Martin]
The "Presentation" intended in the diagram is not the whole complex, but the
lights alone.
But it is combined (via E) with the subject's output and enters the
perceptual function that turns it into the perception that is
controlled, which must include the tones and the button press, because
that is what the subject does seem to control (when the light precedes
the final tone by enough time).

Comment (1) applies.


"Interpret and Compare" compares the perception of which light
is lit with the imagined/remembered answers initially provided by the
experimneter's instructions.
So it's a comparator and not a perceptual function? If so, then there
should be a reference arrow entering that box, shouldn't there?

Comment (2) applies.

Must go. Sorry.

Martin

[From Rick Marken (2009.02.26.1800)]

Martin Taylor (2009.02.26.17.51) --

Much of what I would do in detail can be
encapsulated in two comments:
���� (1) Most of the time, I think of perceptual functions as having a
single scalar output, and
���� (2) I assume most perceptual functions have two or more inputs, even
though quite often we draw control diagrams in which only one input is shown
to a perceptual function.

Me:

But it {the "presentation] is combined (via E) with the subject's output and
enters the perceptual function that turns it into the perception that is
controlled, which must include the tones and the button press, because
that is what the subject does seem to control (when the light precedes

>the final tone by enough time).

Comment (1) applies.

Yes, that's the way I think of it too; the output of the perceptual
function (which takes the "presentation" as the input) is a scalar
variable, the magnitude of which indicates the state of an event
perceived in the bips, lights, and presses that are the inputs to the
perceptual function. I see it kind of like this:

bip
bip
bip ---> | P| ---> p (state of bip...bip...light...bip/press event)
light
press

The sensed values of the environmental events on the left are the
inputs to the perceptual function, P, which is a neural network that
computes the value of a perceptual variable, p. The value of p
indicates whether the event is bip...bip...right light...bip/right
press or bip...bip...left light...bip/left press (if it's either one
of those assume that P computes the value 20 for p) or whether the
event is bip...bip...right light...bip/left press or bip...bip...left
light...bip/right press (if it's either one of those assume that P
computes the value 0 for p).

>So it's [the "Interpret and Compare" box] a comparator and not a

perceptual function? If so, then there should be a reference arrow
entering that box, shouldn't there?

Comment (2) applies.

Yes, exactly. The "Interpret and Compare" function has multiple inputs
(sensory representations of all those bips, lights and presses). So
we're in agreement?

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[Martin Taylor 2009.02.27.09.57]

[From Rick Marken (2009.02.26.1800)]

Martin Taylor (2009.02.26.17.51) --
Much of what I would do in detail can be
encapsulated in two comments:
     (1) Most of the time, I think of perceptual functions as having a
single scalar output, and
     (2) I assume most perceptual functions have two or more inputs, even
though quite often we draw control diagrams in which only one input is shown
to a perceptual function.

Me:

But it {the "presentation] is combined (via E) with the subject's output and
enters the perceptual function that turns it into the perception that is
controlled, which must include the tones and the button press, because
that is what the subject does seem to control (when the light precedes
the final tone by enough time).

Comment (1) applies.

Yes, that's the way I think of it too; the output of the perceptual
function (which takes the "presentation" as the input) is a scalar
variable, the magnitude of which indicates the state of an event
perceived in the bips, lights, and presses that are the inputs to the
perceptual function. I see it kind of like this:
bip
bip
bip ---> | P| ---> p (state of bip...bip...light...bip/press event)
light
press
The sensed values of the environmental events on the left are the
inputs to the perceptual function, P, which is a neural network that
computes the value of a perceptual variable, p. The value of p
indicates whether the event is bip...bip...right light...bip/right
press or bip...bip...left light...bip/left press (if it's either one
of those assume that P computes the value 20 for p) or whether the
event is bip...bip...right light...bip/left press or bip...bip...left
light...bip/right press (if it's either one of those assume that P
computes the value 0 for p).

I don’t see control of p in this. I see control of a button press. I
grant that this is one of the components of your p. But at the time of
the button press, there is no more bip, so p has a different value, and
depending on when the third bip occurs relative to the light onset, p
will have yet another value – all of which must be countered by the
choice of pushing left or pushing right. It just doesn’t wash.


So it's [the "Interpret and Compare" box] a comparator and not a
perceptual function? If so, then there should be a reference arrow
entering that box, shouldn't there?

Comment (2) applies.
Yes, exactly. The "Interpret and Compare" function has multiple inputs
(sensory representations of all those bips, lights and presses). So
we're in agreement?

Very far from it.

I see neither value in, nor justification for, combining into one
scalar variable three completely different perceptions that have
different functions in control: a perception of time that has as input
an auditory sensation, a perception of location that has a visual
sensation as input, and a perception of moving a chosen finger to a
button, which has haptic or kinaesthetic sensation as input. It’s much
the same as putting in one scalar variable the meaning of what one is
reading, the feeling of the book cover in one’s hands, the music in the
background, and the smell of dinner cooking…

What the “Interpret and Compare” has to interpret is the location
category, which has no input from the auditory system, and is
(presumably) based on the difference between two visual sensation
values. What it has to compare is the interpreted location category and
the categories of answers made available by the experimenter. The
interpreted location category is not a controlled variable. Control of
the match is achieved by variation of the answer category.

Here’s a slightly more elaborated diagram, which splits “Interpret and
Compare” into component levels of perception. In it, I’ve split the
original diagram’s “Sense” into two “See” perceptions, one for each
light, and have split “Interpret and Compare” into three levels to
indicate its separate functions. This diagram still omits some obvious
control loops, but I hope it is enough to show wherein I disagree with
your suggestion that everything the subject senses is combined at some
stage into a single scalar variable.

SelectionTiming.jpg

I have shown a generic arrow as output from the timing control loop to
the Output Gain function of the button-press control loop. A signal on
this path either switches the output of the button-press control loop
from imagination into muscular action or changes the gain of that loop
from zero to some appropriate value. I don’t have any commitment to
particular mechanisms here, but something of the kind must exist, or
the button-press loop would be continuously active. In a more standard
experiment in which the instruction was “press as quickly as possible
after the light comes on”, some similar switching control would be
necessary, but it would be activated by some perception of the
likelihood that the light had turned on rather than by the occurrence
of the third auditory “bip”.

I don’t know about anyone else, but to me, this kind of diagram, in
which the different sensed inputs are kept separate according to their
functions in distinct but interacting control loops, seems very much
more likely than one in which all the sensory inputs are combined into
one single “p” that is influenced only by a one-time push of one of two
buttons.

Martin

[From Rick Marken (2009.02.27.1710)]

Martin Taylor (2009.02.27.09.57) --

I don't see control of p in this. I see control of a button press.

OK, but the button press is a perception, too, right.

I grant that this is one of the components of your p. But at the time of the
button press, there is no more bip, so p has a different value, and depending on
when the third bip occurs relative to the light onset, p will have yet
another value -- all of which must be countered by the choice of pushing
left or pushing right. It just doesn't wash.

The perceptual function, as I conceive it, captures all of this and
computes a p whose value represents the state of the event, which I
call bip...bip...light...bip/light. Many perceptions unfold over time
and we perceive them as a unified entity: an event, sequence,
principle, etc. For example, a golf swing, a melody, honesty. Bill
discusses the perception of things that occur over time in Chapter 11
of B:CP. See Figure 11.3 for a model of a perceptual function that
perceives the word "juice". Just as in the
"bip...bip...light...bip/light" case, where, as you say, there is no
more bip at the time of the press, in the case of "juice" there is no
more "j" sound at the time of the "oo" or the "s". But we perceive
(and control) the word as a whole just as (I imagine) the subject in
the Schouten study controls the "bip...bip...light...bip/light" event
as a whole.

Me:

Yes, exactly. The "Interpret and Compare" function has multiple inputs
(sensory representations of all those bips, lights and presses). So
we're in agreement?

Martin:

Very far from it.

I can see from your new graph that this is apparently true.

I see neither value in, nor justification for, combining into one scalar
variable three completely different perceptions that have different
functions in control: a perception of time that has as input an auditory
sensation, a perception of location that has a visual sensation as input,
and a perception of moving a chosen finger to a button, which has haptic or
kinaesthetic sensation as input. It's much the same as putting in one scalar
variable the meaning of what one is reading, the feeling of the book cover
in one's hands, the music in the background, and the smell of dinner
cooking.

Why not? That could be a perception, p, of what might be called
"level of comfort". Change one of the components, such as the music,
and p might change from "comfortable" to "uncomfortable". The
hierarchical model of PCT is based on the idea that higher order
perceptions are constructed from lower order ones. Consider the
perception of "sequence" in my hierarchical control demo. The perceive
sequence that unfolds is constructed from the lines, shapes, changes
in position of the shapes over time, the size of the shapes, etc.
That's actually why I asked you to try to control the sequence (at the
slow rate) in the demo; you could see that you have to perceive the
state of the sequence (which is, according to PCT, the state of a
perceptual variable, p) in order to control the sequence (keep it at
S, M, L, S, M... rather than S, L, M, S,L...).

What the "Interpret and Compare" has to interpret is the location category,
which has no input from the auditory system, and is (presumably) based on
the difference between two visual sensation values. What it has to compare
is the interpreted location category and the categories of answers made
available by the experimenter. The interpreted location category is not a
controlled variable. Control of the match is achieved by variation of the
answer category.

I thought control of the match (or press location to light location)
is accomplished by pressing on the lighted button?

Here's a slightly more elaborated diagram, which splits "Interpret and
Compare" into component levels of perception. In it, I've split the original
diagram's "Sense" into two "See" perceptions, one for each light, and have
split "Interpret and Compare" into three levels to indicate its separate
functions. This diagram still omits some obvious control loops, but I hope
it is enough to show wherein I disagree with your suggestion that everything
the subject senses is combined at some stage into a single scalar variable.

I think this model has some problems. The "timing" loop would not know
whether the press occurred at the same time as the third beep -- or at
all, for that matter -- and the "match" (of press to lighted button)
loop would not know if the lighted button was pressed because both of
these loops are open. The "match" control loop is not really
controlling for a match because it is not perceiving it (in order to
perceive it the press would have to be an input to the "match" box as
well as the light). The "timing" control loop is not really
controlling for simultaneous 3rd bip/press because it doesn't perceive
it. In order to control for a match or simultaneity the systems
controlling these variables must be able to perceive them. So neither
system in your diagram is a control system; this is an S-R model of
the participant in the experiment. It might seem like an OK model
given that there were no serious disturbances in the Schouten study.
But if you replicated the study with disturbances -- say by
occasionally pushing the participant's hand away from the button as it
was being pressed -- you would see corrective action. I think you will
find that you are dealing with a closed loop control system.

I don't know about anyone else, but to me, this kind of diagram, in which
the different sensed inputs are kept separate according to their functions
in distinct but interacting control loops, seems very much more likely than
one in which all the sensory inputs are combined into one single "p" that is
influenced only by a one-time push of one of two buttons.

It's not the distinctness of the systems that is the problem; the
problem is that they are not control systems.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

[From Bill Powers (2009.02.27.1105 MST)]

[Martin Taylor
2009.02.27.09.57]

[From Rick Marken (2009.02.26.1800)]
> > But it {the "presentation] is combined (via E) with the subject's
> > output and
> > enters the perceptual function that turns it into the perception that is
> > controlled, which must include the tones and the button press, because
> > that is what the subject does seem to control (when the light precedes
> > the final tone by enough time).
[Martin] Comment (1) applies.

[Rick] Yes, that's the way I think of it too; the output of the
perceptual
function (which takes the "presentation" as the input) is a
scalar
variable, the magnitude of which indicates the state of an event
perceived in the bips, lights, and presses that are the inputs to the
perceptual function. I see it kind of like this:
bip
bip
bip         ---> | P| --->
p (state of bip...bip...light...bip/press event)
light
press
The sensed values of the environmental events on the left are the
inputs to the perceptual function, P, which is a neural network that
computes the value of a perceptual variable, p. The value of p
indicates whether the event is bip...bip...right light...bip/right
press or bip...bip...left light...bip/left press (if it's either one
of those assume that P computes the value 20 for p) or whether the
event is bip...bip...right light...bip/left press or bip...bip...left
light...bip/right press (if it's either one of those assume that P
computes the value 0 for p).

[Martin] I don’t see control of p in this. I see control of a button
press. I grant that this is one of the components of your p. But at the
time of the button press, there is no more bip, so p has a different
value, and depending on when the third bip occurs relative to the light
onset, p will have yet another value – all of which must be countered by
the choice of pushing left or pushing right. It just doesn’t
wash.

[WTP] I’m very pleased to see you, Martin and Rick, engaging in this
exploration, which is what it is very much in the MOL sense. You are
examining your perceptions of a complex situation and seeing its elements
all mixed together at many levels, and trying to sort them out much as
Bob Clark and Kirk Sattley and I were doing 50-odd years ago. What you
are doing in the excerpt above is noticing different aspects of the
situation – meaning variables at different levels, as well as the same
level – and for the moment disagreeing because of focusing on different
but coexisting levels. Martin sees no control of p in the relationship
between bips and button presses because he is looking at the
configuration level instant by instant, while Rick is looking at that
relationship as a different perception being controlled over a long
period of time, and not seeing what he could see, which is that the
button-press is also being controlled at a lower level. You are both
right, I believe, but as you continue to work this out you will either
come to the same conclusions I reached with Bob Clark, or you will see
something different, and either outcome will be worth discovering. The
important thing is to be discussing and pointing out experiences, rather
than arguing only in terms of what fits the theory and what doesn’t. If
you stick to descriptions of what you are in fact experiencing (not what
you ought to be experiencing) you will find out what is actually the
case. I do believe it is there to be seen.

I see the button as a configuration of sensations. I see a bip as a very
minimal configuration of sensations. At the configuration level, control
systems work strictly in present time. There is no sequence of bips,
there is no “pressing” of a button, there is no relationship
between a bip and a press, nor is there ever any change in a bip or a
press. There is only the state of a bip and the state of a press, and
whatever error there is if there’s a reference condition. The bip happens
far too fast (we can say from a different level) to be controlled, but
the button-finger combination may be in a state different from the
reference state, which (theoretically, not experientially) is accompanied
by an error that accounts for the current level of muscle force (and its
derivatives) in the eternal present. At the configuration level there is
a perception, and an error, and an effort, and that is all. At that level
Martin is right: there is a button and a finger and pressing and perhaps
a bip and a light or perhaps not.

At the transition level we begin to notice the ongoing changes. The
pressure in the button/fingertip is steady or increasing, a bip may or
may not suddenly be appearing, a light may or may not be suddenly
increasing in intensity.

At the next level, the string of bips may be under way with the time
between the first two yielding a sense of an interval beginning and
ending; an Event. That interval signal may be stored in memory and be
used as a reference for the next interval. A finger may or may not be
feeling the increase in pressure that we label, at a higher level, a
“press.” The reference signal may or may not be calling for
such a perception of increased pressing.

At the relationship level, the increase in fingertip pressure is supposed
to be in the correct temporal relationship with the occurrance of the
next bip. On the first two bips this relationship requires the finger to
be held away from the button to create a “bip and no press”
relationship, and an instant before the third bip occurs the reference
level for that relationship may change so now the fingertip pressure must
increase at the same instant the next bip occurs. At a higher level, this
change in reference is determined as a logical condition: if and only if
the second bip has occurred and a light is on, the finger near the button
associated with the light is to increase its pressure at the time of the
third bip. Over a long series of trials, the time of the change in the
pressing event reference level in relation to the second bip event, when
required, is reorganized – adjusted slowly until synchronism is achieved
(a reaction time of zero every time this is called for).

At the category level, the events become logical variables: they have
occurred or they have not occurred. Each different event is represented
by a different logical variable. The events are first bip, second bip,
third bip, light onset, relationship of bip to finger-press onset (the
only controllable relationship), and relationship of light onset to bip.
There may be duplications of some of these variables for the two
light-button sets.

At the sequence level, we have bip, bip-bip, bip-bip-bip, bip-bip-light
(or possibly bip-light or even light-bip), and bip-bip-(press/bip). Only
the last sequence is controllable.

At the logic level we have the condition already described:

<------ reference condition is TRUE
------> <reference

(bip1-bip2-light OR bip1-light OR light-bip1) —>
bip-bip-(press/bip) (ref)

The only variable that can be altered by the subject to control the above
sequence is the state of the press, which is accomplished by varying the
time when the reference level for the press relationship to the next bip
is changed. Then the reference level for the transition of the finger
pressure from low to high is set to a nonzero value, and so on down to
the act itself.

I know that this leaves out some things I haven’t noticed or didn’t know
what to do with, but that’s what I find when I imagine doing this
experiment, and do the imagining from the level where theoreticians
think, wherever that is.

[Rick]

Yes, exactly. The "Interpret and Compare" function has
multiple inputs
(sensory representations of all those bips, lights and presses). So
we're in agreement?

[Martin] Very far from it.

I see neither value in, nor justification for, combining into one scalar
variable three completely different perceptions that have different
functions in control: a perception of time that has as input an auditory
sensation, a perception of location that has a visual sensation as input,
and a perception of moving a chosen finger to a button, which has haptic
or kinaesthetic sensation as input. It’s much the same as putting in one
scalar variable the meaning of what one is reading, the feeling of the
book cover in one’s hands, the music in the background, and the smell of
dinner cooking…

I think this disagreement is primarily a difference is the level of
control one is thinking about. All the levels are operating at the same
time, and they are physically distinct from each other; each level knows
only what it perceives and controls, and knows nothing of what is
happening at lower or higher levels. Unfortunately for understanding, we
consciously experience most of these processes as if they were all
projected into the same plane so we lose the dimension we refer to as
“level”. We have to attend to the details carefully so we can
sort out what has to happen before what else happens, and then how the
result is controlled, if it is.

I am half-reminded of many little illuminations that happened back when I
was first going through this sort of thing, one being the idea that one
level knows nothing of other levels. Another is a principle to the effect
that one has to find ALL the levels in order for each step to appear
necessary rather than arbitrary. There seems to be nothing left out when
we see that change arises from different states of configurations, with
no intermediate step being necessary: the change in the value of X simply
IS (X2 - X1). Ideally, we would see the progression from each
level to the one above it as equally compelling and minimal.

But perhaps the biggest illumination was simply concluding that nothing
important is hidden: if you look, you can see what is there.

I hope both of you will continue with this. My word on it is not the last
word, and being objectively or logically right is not the point. The
point is to observe what is actually there and tease it apart into levels
as well as one can. When you succeed in seeing what is really there it’s
possible that you will find that others can agree that they see the same
thing, as near as can be determined. If there is agreement, then maybe we
can claim to have discovered something that is a common property of homo
sapiens.

Best,

Bill P.

[Martin Taylor 2009.02.28.10.46]

[From Rick Marken (2009.02.27.1710)]

When I got up this morning, I had a different view of what you were
talking about. It is about the control loops I meant when I said
[Martin Taylor 2009.02.27.09.57]: “This diagram still omits some
obvious
control loops,” but I was wrong to have followed with: " I hope it is
enough to show wherein I disagree with
your suggestion that everything the subject senses is combined at some
stage into a single scalar variable." I should have said “a single
scalar variable for which the control has some influence on the
probability that the subject’s button push corresponds to the light
that turns on.” During the experiment, there are many such variables,
one of which I represented in my first diagram on the subject [Martin
Taylor 2009.02.14.14.32]. If you want to go all the way up the
hierarchy, you can talk about the subject’s control of perception of
self, and of the subject’s control of perception of others’ perception
of oneself, which provide reference values to such control systems as
get the subject into the experiment in the first place, allow the
subject to control for the experimenter’s satisfaction, and such-like.

My earlier diagram of the control systems relevant to the correctness
of the subject’s button push [Martin Taylor 2009.02.23.15.07] had
nothing related to the timing of the subject’s answer – how the
subject is able to push the button on the third bip. You wanted to
include the bips into a single controlled variable p along with the
buttons and the lights – and, I presume, the experimenter, though you
didn’t mention it. I couldn’t see how the choice of either left or
right button could control such a “p”, and used the analogy of a
perception consisting of all the visual, auditory and tactile
sensations that did NOT relate to the perception of the meaning of the
content that was being read. You mentioned “comfort”, and that was what
allowed me to see that you were not at all talking about the control
systems that affect the correctness of the subject’s button press, but
about control systems related to the subject’s being in, and staying
in, the experiment, and performing the task as the experimenter asks
(which, in the subject’s view, could be one requirement for staying in
the experiment).

When I came to write this message, I found Bill
[From Bill Powers (2009.02.27.1105 MST)] had also noted that we were
talking about different levels of control. He mentioned a couple of
other control systems possibly involved in the process of making a
response, but they don’t influence the correctness of the subject’s
button press except insofar as error in the sequence-control system
would affect the moment of the button press, and Schouten’s experimnet
showed that to affect the probability of a correct response.


Martin Taylor (2009.02.27.09.57) --

I don't see control of p in this. I see control of a button press.
OK, but the button press is a perception, too, right.

Yes. My diagrams have explicitly shown that control loop.

I think this model has some problems. The "timing" loop would not know
whether the press occurred at the same time as the third beep -- or at
all, for that matter -- and the "match" (of press to lighted button)
loop would not know if the lighted button was pressed because both of
these loops are open.

In general, in hierarchic control system, higher-level control systems
output signals that are reference levels for lower systems. They don’t
usually monitor to see whether those lower control systems achieve the
desired perceptual values. What they control is their own perceptions,
to which the lower-level controlled perceptions contribute.

You question two of these control systems because they are “open”. In
past discussions, such systems have been called “fire and forget”.
There’s nothing a gunner can do to make a shell more or less likely to
hit its target after it has been launched. But the gunner can observe
the fall of shot and correct the aim for the next shot. You could call
this a kind of “learning”. A different control loop that controls the
perception of fall of shot affects the parameters of the “shooting”
control loop, not the signal values during the execution of a single
shot.

We have the same situation here. The “Match” control loop must “fire
the shot” (provide a reference value to the button-press control loop)
at the appropriate time. In my diagram, I suggested that it does this
continuously, and placed the timing switch at the execution of the
button press, but it could equally be at the output of the “Match”
loop. In fact that’s where I put it in an early version of this
diagram. If there is any learning to be done, it cannot affect this
particular “shot” but must affect something related to the
interpretation of subsequent presentations. Likewise for the timing
loop. I drew it as a match to a count of bips, but Bill suggests it is
sequence control, which sounds right. However, it is another
fire-and-forget system. The accuracy of the match of button-press-time
to third-bip-time can be used by another control system to change the
parameters of the sequence control system in the same way that a
fall-of-shot control system can be used by the gunner to change the
aiming parameters of the gun, but that cannot influence when the button
was pushed this last time.

The "match" control loop is not really
controlling for a match because it is not perceiving it (in order to
perceive it the press would have to be an input to the "match" box as
well as the light). The "timing" control loop is not really
controlling for simultaneous 3rd bip/press because it doesn't perceive
it. In order to control for a match or simultaneity the systems
controlling these variables must be able to perceive them. So neither
system in your diagram is a control system;

True.

this is an S-R model of
the participant in the experiment.

False. Many (most) perceptions are uncontrolled, and many are
controlled in imagination even though the output is not manifest in
overt action. Neither requires that a person be an S-R system.

Perhaps you might refer back to [From Bill Powers (2009.02.20.1946
MST)], and specifically to his comment: “Martin emphasized that the
action does not affect the presentation. That loop is not closed,
right? It could be closed if the action did affect the presentation,
which isn’t impossible, but it doesn’t, so that potential loop is not
closed – which means it’s open, as we use the term. There IS an
open-loop relationship between S and R, isn’t there? Or are you saying
that you have to have a closed-loop relationship and open it before you
can call it open-loop?”

It might seem like an OK model
given that there were no serious disturbances in the Schouten study.
But if you replicated the study with disturbances -- say by
occasionally pushing the participant's hand away from the button as it
was being pressed -- you would see corrective action. I think you will
find that you are dealing with a closed loop control system.

Very many control loops, yes. The particular one you mention is
actually in my diagram.


I don't know about anyone else, but to me, this kind of diagram, in which
the different sensed inputs are kept separate according to their functions
in distinct but interacting control loops, seems very much more likely than
one in which all the sensory inputs are combined into one single "p" that is
influenced only by a one-time push of one of two buttons.
It's not the distinctness of the systems that is the problem; the
problem is that they are not control systems.

I think that’s a theological comment, not a scientific one. It comes
from a doctrine that says all perceptions are controlled, and its
equivalent, that “fire-and-forget” components cannot exist within a
control hierarchy.

Martin

Martin

[Martin Taylor 2009.02.28.11.52]

[Martin Taylor 2009.02.28.10.46] To

[From Rick Marken (2009.02.27.1710)]
The "match" control loop is not really
controlling for a match because it is not perceiving it (in order to
perceive it the press would have to be an input to the "match" box as
well as the light). The "timing" control loop is not really
controlling for simultaneous 3rd bip/press because it doesn't perceive
it. In order to control for a match or simultaneity the systems
controlling these variables must be able to perceive them. So neither
system in your diagram is a control system;

True.

I should not have said “True” here. I was thinking of loops closed
through the external environment, and these loops are not completed
through the external environment. They are controlling in imagination.
The “Match” control loop controls a perception of the relation between
the category perception of which light was lit and a proposed “answer”
(also a category perception). The control output is to vary the
imagined answer. Assuming the reference is for the answer to correspond
to the lit light (which it would be if the subject intends to be
cooperative), the error in that control system is a failure of the two
to correspond.

I’m less clear about the working of sequence control, but that’s
because I’ve always been unclear about sequence control when dealing
with a single ongoing sequence. I re-read Chapter 11 of B:CP following
your message in which you suggested I should, but I didn’t come away
any clearer on the subject. In your multi-level demo, the sequences
repeat, and the situation is no longer fire-and-forget. When we are
dealing with one occurrence of a sequence, it seems to me that control
– involving comparison of reference with actual sequence – must
involve an imagined representation of the sequence. In this case, the
overt fire-and-forget action (activating the button press) does affect
which sequence will have been perceived (bip bip bip press, bip bip
press bip, or bip bip bip+press), but that perception will be too late
to affect what sequence actually occurred. Control must be in
imagination, I think. But I’m open to other proposals.

Either way, it is false that the match control loop and the timing
control loop are not control systems. I should not have said “True”.

Martin