Prediction

[Bill_Powers1]

[From Bill Powers (2005.02.13.1150 MST)]

Rick Marken (2005.02.13.2035)--

The "predictive" model was only implemented when the target was a sine wave. If the same predictive model had been implemented for the random target (and it would have been very simple to do that since the "model" only involved setting the reference for target position slightly ahead of the present target position in the target disturbance (dt) table) you would have seen a large stability value in the random/pursuit condition. But the non-predictive model behaves exactly like the subject in this condition. This suggests to me that there is nothing for prediction to explain in the random/pursuit condition.

This is interesting because the random target is not completely unpredictable. Because it is low pass filtered noise, the target position is actually fairly predictable. Yet there is no evidence at all that there is any prediction involved in the random target condition. "Prediction" only seems to kick in when the target movement is a repeated temporal pattern, as in the case of the sine wave. I take this as evidence that what is actually going on in the sine wave target case is that the subject is controlling (at the pursuit level) for producing a sinusoidal temporal pattern of cursor movement that is in phase with the sinusoidal target movement. That is, in the sine wave case the target (at the pursuit level) is a temporal pattern (the back and forth sinusoidal movement of the target line); in the random case the target is just the position of the target line since there is no regular temporal configuration to track.

This is, I think, a workable solution for this case. I think we have to keep saying that the point is not to deny that prediction occurs, but to avoid attributing performances to prediction that are more properly explained in a different way (as in the case of catching flyballs or sailing on collision courses).

Your proposal brings up a subject long put off, the idea of the "central pattern generator." It seems pretty clear to me that such things exist; it's pretty hard to explain walking and running gaits with out them, for one thing. A sine-wave generator is one example of a central pattern generator. Control would be exterted by using error signals to adjust frequency and amplitude of the pattern being generated by the output function. In a more generalized form, this might turn out to be what we need to replace the "event" level of control, with which I have never been very happy.

This should not be too difficult to model. I hope.

Best,

Bill P.

···

Best

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

[Rohan_Lulham]

[From Bill Powers (2005.02.13.1150
MST)]

Rick Marken (2005.02.13.2035)–

The “predictive” model
was only implemented when the target was a sine wave. If the same
predictive model had been implemented for the random target (and it would
have been very simple to do that since the “model” only
involved setting the reference for target position slightly ahead of the
present target position in the target disturbance (dt) table) you would
have seen a large stability value in the random/pursuit condition. But
the non-predictive model behaves exactly like the subject in this
condition. This suggests to me that there is nothing for prediction to
explain in the random/pursuit condition.

This is interesting because the random target is not completely
unpredictable. Because it is low pass filtered noise, the target
position is actually fairly predictable. Yet there is no evidence at all
that there is any prediction involved in the random target condition.
“Prediction” only seems to kick in when the target movement is
a repeated temporal pattern, as in the case of the sine wave. I take this
as evidence that what is actually going on in the sine wave target case
is that the subject is controlling (at the pursuit level) for producing a
sinusoidal temporal pattern of cursor movement that is in phase with the
sinusoidal target movement. That is, in the sine wave case the target (at
the pursuit level) is a temporal pattern (the back and forth sinusoidal
movement of the target line); in the random case the target is just the
position of the target line since there is no regular temporal
configuration to track.

This is, I think, a workable solution for this case. I think we have to
keep saying that the point is not to deny that prediction occurs, but to
avoid attributing performances to prediction that are more properly
explained in a different way (as in the case of catching flyballs or
sailing on collision courses).

Your proposal brings up a subject long put off, the idea of the
“central pattern generator.” It seems pretty clear to me that
such things exist; it’s pretty hard to explain walking and running gaits
with out them, for one thing. A sine-wave generator is one example of a
central pattern generator. Control would be exterted by using error
signals to adjust frequency and amplitude of the pattern being generated
by the output function. In a more generalized form, this might turn out
to be what we need to replace the “event” level of control,
with which I have never been very happy.

This should not be too difficult to model. I hope.

Best,

Bill P.

Best

Rick

---

Richard S. Marken

marken@mindreadings.com

Home 310 474-0313

Cell 310
729-1400

looking forward to the next installment_ finding this
interesting
with the time difference will likely be there in the morning

···

At 05:57 PM 14/02/2005, you wrote:
Rohan Lulham
Ph.D. Student
Environment, Behaviour and Society Research Group
Faculty of Architecture, University of Sydney
Australia

[Martin_Taylor2]

[From Rick Marken (2005.02.13.2035)]

Martin Taylor (2005.02.13.18:00) --

Anyway, that's all beside the point, isn't it? The point is that whichever way you model it, both Rick and I found that incorporating the predictive component improved the fit of the model to real data. Rick may be right that the same result can be interpreted differently, but he'd have to prove that the different interpretation was at least as plausible withint the context of the general theory. I think that would be hard

Actually, I think it's quite possible to see that there is no evidence of prediction just based on the data I posted. Here's the subject data for the random and sine targets:

Type of Target Movement (dt)
          Random Sine Wave

Pursuit 6.88 15.8

Comp 3.7 2.1

and here are the results for the "predictive" model:

     Type of Target Movement (dt)
          Random Sine Wave

Pursuit 6.67 15.85

Comp 3.78 2.4

The "predictive" model was only implemented when the target was a sine wave. If the same predictive model had been implemented for the random target (and it would have been very simple to do that since the "model" only involved setting the reference for target position slightly ahead of the present target position in the target disturbance (dt) table) you would have seen a large stability value in the random/pursuit condition. But the non-predictive model behaves exactly like the subject in this condition. This suggests to me that there is nothing for prediction to explain in the random/pursuit condition.

You could be right, but you provided a different explanation in your original message, in a passage quoted by both of us. At the risk of excessive redundancy, I quote it again:

There is actually a third level of control evident in the
behavior in this experiment. It is the level that notices
that the target has become predictable. The subject is clearly
not using predictive control with the random target -- there is
nothing to predict. The third level system "switches" in the second
level system that let's the purpuit control system control (t'-c)
rather than (t-c). I became personally aware of this third
level while I was blithly testing my program. At one point, though
inattention, I failed to notice the change from random to predictable
target (marked only by the posting of data from the end of the run).
So I just kept tracking the target position as though I did not
know here it was going to be at the next instant; in other words,
I kept tracking the sine wave target as though it was still the
random target. When the program printed out the results I was
alarmed becuase it shows that my ability to control with the
predictable target was EXACTLY the same as my ability to control
with the random target. At first, I though that perhaps results
like those shown in the first table above did not occur reliably. In
fact, the results only occur reliably if the subject is actually
controlling a higher level variable (t'-c) rather than (t-c) in the
predictable target condition. I loved it; my little lapse showed (once
again) that it is the subject, not the "stimulus situation", who controls
what happens in this experiment. And with PCT you can tell exactly what the
subject is controlling.

You were pretty clear about the _subjective_ experience.

When you expected NOT to be able to predict the sine wave, and behaved accordingly, your results did not match the model with prediction, but they were the same as you achieved with the random waveform. You did not expect to be able to use prediction when you were tracking the random waveform, and the model without prediction fitted your data.

You interpreted this at the time as suggesting a level of control that switched the use of prediction in or out, depending on whether you implicitly assumed it to be potentially useful. Now you propose that there is no prediction in either situation but the subject is simply matching a remembered sine wave pattern.

To see which of your two interpretations is more correct, you would have to do two things: write the model so it used true prediction rather than the pseudo-prediction of a phase advance of the target signal, and do the human tracking with the subjective intent of predicting in the random condition. As you say, with a random bandwidth of 0.6 Hz, some useful prediction 180ms ahead should be possible (the limit of prediction with that bandwidth is about 1.6 sec, but the quality of possible prediction declines down to zero, rather than dropping abruptly at the 1.6 sec mark. It should still be pretty good at 180 msec.)

So far, in your lovely experiment, we have both you and the model apparently using prediction with the sine wave, you not using prediction with the sine wave, and you and the model not using prediction with the random wave. We lack you and the model using prediction with the random wave, and the model not using prediction for the sine wave (though one might well assume that the model results for this last case would match its results for the random wave reasonably well, as did your human results).

For my part, I need to go back and figure out why my analysis of the prediction situation gave the wrong answer. It's not much good saying (or showing) that prediction is useful unless one can analyze the way it's being used. This one experiment is really neat for what it does, and its principles could be extended to more complex predictable situations, but it needs more subjects and a better analysis before it can be considered definitive.

Martin

[Bill_Powers1]

[From Bill Powers (2005.02.14.0648 MST)]

Rick Marken (2005.02.13.2035)]

Martin Taylor (2005.02.13.18:00)

[Martin] Now you propose that
there is no prediction in either situation but the subject is simply
matching a remembered sine wave pattern.

I think Rick will say this too, but I’ll chime in. The proposal is
(as I’m presently thinking about it) that there is another mode of
control that involves generating a sine-wave pattern of reference signals
for cursor position, with the controlled variable at the higher level
being the separation of cursor and target as usual. Two dimensions of
control are now involved: phase and amplitude. The phase error is
corrected by adjusting the phase of the sine-wave generator in the output
function; the amplitude error is corrected by adjusting amplitude of the
sine-wave. So there is no memory of a sine-wave involved.
In my “portable demonstrator” (see 1960 paper) there is one
demonstration that is pertinent to this. The experimenter moves a finger
up and down in a fairly rapid sine-wave pattern while the other person
tracks it with one finger. This should be rapid enough so that simple
positional tracking can’t be done very accurately. What normally happens
is that the subject gets his finger going in the same pattern, matching
the whole pattern to the experimenter’s movements. Then, when tracking is
pretty good, the experimenter simply stops his finger at some point in
the cycle. The subject’s finger goes through a large part of a sine-wave
before stopping, showing that the pattern is being independently
generated. The delay before the controller stops is much longer than it
is when the experimenter’s movements are random.
I’m sure this can be instrumented easily.
This pattern-generator idea can be investigated further. Consider a
pattern of movement of the target that is regular and repeated but not a
sine-wave. The repetition is an essential element of this sort of
situation. In the random movement case there is no pattern to be
repeated; the regularity in the sine-wave case is specifically a
repeated regularity. The fact that it is a sine-wave is
unimportant. This would seem to be a missing dimension of the level I
have been calling the “event” level. And it may have an
important bearing on one main form of prediction.

I certainly agree with you about needing more experiments with more
subjects.

Best,

Bill P.

Best,

Bill P.

[Martin_Taylor2]

Re: Prediction
[Martin Taylor 2005.02.14.12.00]

[From Bill Powers (2005.02.14.0648
MST)]

Rick Marken (2005.02.13.2035)]

Martin Taylor (2005.02.13.18:00) –

[Martin] Now you propose that there is no
prediction in either situation but the subject is simply matching a
remembered sine wave pattern.

I think Rick will say this too, but I’ll
chime in. The proposal is (as I’m presently thinking about it)
that there is another mode of control that involves generating a
sine-wave pattern of reference signals for cursor position, with the
controlled variable at the higher level being the separation of cursor
and target as usual. Two dimensions of control are now involved: phase
and amplitude. The phase error is corrected by adjusting the phase of
the sine-wave generator in the output function; the amplitude error is
corrected by adjusting amplitude of the sine-wave. So there is no
memory of a sine-wave involved.

I think that’s plausible, but I like the following better, as it
doesn’t assert anything special about the form of the wave. At the
moment, I can’t think of an evolutionary reason for us to be provided
with a specialized sine-wave generator.

This pattern-generator idea can be
investigated further. Consider a pattern of movement of the target
that is regular and repeated but not a sine-wave. The repetition is an
essential element of this sort of situation. In the random movement
case there is no pattern to be repeated; the regularity in the
sine-wave case is specifically a repeated regularity. The fact
that it is a sine-wave is unimportant. This would seem to be a missing
dimension of the level I have been calling the “event”
level. And it may have an important bearing on one main form of
prediction.

Worth thinking further about, for sure.

I certainly agree with you about needing
more experiments with more subjects.

Yes. Filling in the missing cells of Rick’s test matrix would be
a good start. What probably should be added beyond filling those empty
cells is a third kind of waveform, one that repeats with a fundamental
frequency of, say, 0.3 Hz but isn’t a sine wave.

As an aside…Perhaps related, there’s an auditory phenomenon
that my friend Roy Patterson demonstrated to me many years ago in
Cambridge. If you have a repetitive noise waveform with a fundamental
frequency above some lower bound, say 20 Hz (I forget the number), you
hear a continuous tone. If the fundamental frequency is a bit lower,
it’s kind of rumbly and variable in quality (I guess the
inter-relations of the harmonics must be like those at the onset and
offset of natural noises), and if it’s yet lower, in the 1 Hz range,
you hear a repetitive pattern of auditory events such as clicks,
bumps, tone bursts … The clicks, bumps, tone bursts, etc, don’t see
to be heard right away, so far as I remember, but emerge out of the
noisy background. But if it is too slow a repetition, you don’t hear
the repetitive nature at all. It’s just an ongoing noisy
environment.

Getting back to the pattern generator idea, it seems to me that
it would be implemented very much as the Artificial Cerebellum was
(and would perform the function that I had thought the A-C performed).
Intuitively, I think that both raw prediction and pattern generation
of that kind are likely to occur, and perhaps to be used
together.

Again, that’s an idea that needs to be worked out, but it
“feels right” to me.

Martin

[Richard_Marken2]

[From Rick Marken (2005.02.13.0930)]

Bill Powers (2005.02.13.1150 MST)--

>Rick Marken (2005.02.13.2035)--

I take this as evidence that
what is actually going on in the sine wave target case is that the subject
is controlling (at the pursuit level) for producing a sinusoidal temporal
pattern of cursor movement that is in phase with the sinusoidal target
movement...

This is, I think, a workable solution for this case. I think we have to
keep saying that the point is not to deny that prediction occurs, but to
avoid attributing performances to prediction that are more properly
explained in a different way (as in the case of catching flyballs or
sailing on collision courses).

Yes. Very important. This morning, I predicted that there would be heavy
traffic on Santa Monica Blvd. so I took Ohio Ave instead. I have no idea
what the traffic actually was like on Santa Monica but my prediction
influenced my controlling (how I got to work).

But prediction of this sort (imagining the future state of a variable) is
not necessarily involved in all controlling. It is clearly not involved in
our standard tracking task (as evidenced by the fact that we can account for
99% of the variance using a model that has no prediction in it) and I
suspect that it is not involved in a tracking task in which the controlled
variable is a regular temporal pattern (like a sine wave).

I think it just _seems like_ prediction is involved in control when people
are controlling variables that are defined over time, like the sine wave
pattern. David Goldstein's (2005.02.13.2250 EST) examples of apparent
prediction in tennis are another nice example. You can describe the
situation as prediction of where the ball will be. But what could be
involved is control of a temporal pattern, where in this case the pattern
involved the temporal sequence involving what David describes in his post:
where the ball bounces in the box of the receiver and how the receiver is
positioned and moving relative to the ball just before and after it leaves
the receiver's racket.

Your proposal brings up a subject long put off, the idea of the "central
pattern generator." It seems pretty clear to me that such things exist;
...
This should not be too difficult to model. I hope.

Not too difficult for you maybe. I think I gave up on trying to build such a
model because it was too tough for me. Indeed, I think this may be why I
abandoned the research Martin posted. I think I figured out back then that
what was going on was control of a temporal pattern but I didn't know how to
build the temporal pattern controller. Perhaps now is the time to try to
figure it out. I think it might really be worth it to design a system that
controls a temporal pattern. This would help people understand that the
perceptions we control include complex patterns that are defined over time.
I think there is a tendency to think of perceptions as configurations like
the cursor and target in our tracking tasks. People forget that controlled
perceptions can be defined over time. Temporal events like a sine pattern,
tennis "poach" or piano sonata are just as controllable as the position of a
cursor, but when people control these events it can look like they are
predicting the future -- since the future is part of these perceptual
events.

Best regards

Rick

···

--
Richard S. Marken
MindReadings.com
Home: 310 474 0313
Cell: 310 729 1400

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

This email message is for the sole use of the intended recipient(s) and
may contain privileged information. Any unauthorized review, use,
disclosure or distribution is prohibited. If you are not the intended
recipient, please contact the sender by reply email and destroy all copies
of the original message.

[Richard_Marken2]

[From Rick Marken (2005.02.14.0945)]

Bill Powers (2005.02.13.1100 MST) to Martin Taylor (2005.02.13.18:00) --

Rick says that he adjusted the sine-wave frequency so as to be near the
middle of the assumed frequency range of the smoothed random disturbance.
If that is so, there should be more error in the random case than the
sine-wave case simply because of the higher-frequency components in the
random case, which would increase the tracking error. It is not a simple
matter to show that any improvement in the sine-wave case is due to
prediction. What needs to be done is to fit a model to the random case, and
then change the target movement to a sine-wave and adjust the frequency
until the tracking error is the same as in the random case with the same
model parameters in effect... This will eliminate improvements in tracking due
only to the frequency spectra involved. If there is then an improvement in
human tracking with the sine-wave, we can conclude that the regularity is
allowing better control, and modify the model to do the same thing. If
there is no improvement, we can explain the apparent difference in
performance in a simple way that does not use prediction.

Actually, something like this was done. I ran a same simple control model in
the random and sine disturbance case. The level of control exhibited by the
model in both cases (measured as stability -- expected over observed
variance so a big number means better control) is shown below:

     Type of Target Movement (dt)
          Random Sine Wave

Pursuit 6.67 6.26

So a simple control model controlled about the same with the random and sine
target movement (actually, slightly better with the random disturbance). But
the results for the human subject were quite different:

     Type of Target Movement (dt)
          Random Sine Wave

Pursuit 6.67 15.85

So the human did _much_ better with the sine wave target movement.

Best

Rick

···

--
Richard S. Marken
MindReadings.com
Home: 310 474 0313
Cell: 310 729 1400

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

This email message is for the sole use of the intended recipient(s) and
may contain privileged information. Any unauthorized review, use,
disclosure or distribution is prohibited. If you are not the intended
recipient, please contact the sender by reply email and destroy all copies
of the original message.

[Bill_Powers1]

[From Bill Powers (2005.02.15.0805 MST)]

Martin Taylor 2005.02.14.12.00--

At the moment, I can't think of an evolutionary reason for us to be provided with a specialized sine-wave generator.

Here is a sine-wave generator:

y = 100;
x = 0;

repeat
y := y + k*x*dt
x := x - k*y*dt
indefinitely

The frequency of oscillation is k.

So a "specialized" sine-wave generator is two integrators connected in a closed loop, with one sign inversion. Almost any other kind of waveform generator would be more complex (not counting variants on the above form produced by nonlinearities and limits). Control of amplitude requires minor additions (phase is controlled by varying k).

I do think there are more complex pattern generators. It will be interesting to try some experiments with people imitating trajectories of various kinds.

I'm watching the Abrams thread from the sidelines. Not much happening there.

Best,

Bill P.

[Bill_Powers1]

[From Bill Powers (2005.02.15.0816 MST)]

Rick Marken (2005.02.13.0930)--

I think it just _seems like_ prediction is involved in control when people
are controlling variables that are defined over time, like the sine wave
pattern.

It may be more fruitful to turn this on its head. What do we mean by "predicting?" One meaning is surely to make a statement or form an expectation about something that hasn't happened yet. But another is simply to act on the basis of the extrapolated future state of some variable -- i.e., its first derivative times some adjustable constant, added to its present state. That's how "anticipation" works in heating and cooling systems, and elsewhere. The effect of prediction or anticipation in this latter sense is, as we often find in control systems, the opposite of what intuition suggests. The action is slowed rather than accelerated. Engineers just use words like "first derivative" and get away from all the unwanted connotations and incorrect implications of informal language. We waste a lot of time arguing about what informal language "really means." It doesn't usually have very clear meanings. The real question is, "When I use words like predict and anticipate and expect, what am I really trying to describe?" There is a good reason why scientists develop specialized languages (mathematics being one). It's the only way to be clear about what you mean.

Cognition is described by saying it's any thought. However, "cogito" means "think", so this explanation really says that thought is described by saying it's any thought. To that we can add that intelligence means being smart, and the reason that sleeping powder makes you drowsy is that it contains a dormitive principle.

Best,

Bill P.

[Martin_Taylor2]

[Martin Taylor 2005.02.15.10.43]

[From Bill Powers (2005.02.15.0805 MST)]

Martin Taylor 2005.02.14.12.00--

At the moment, I can't think of an evolutionary reason for us to be provided with a specialized sine-wave generator.

Here is a sine-wave generator: ....

I do think there are more complex pattern generators. It will be interesting to try some experiments with people imitating trajectories of various kinds.

We are thinking of different things, I guess. When you say "there are more complex pattern generators" it sounds as though you are thinking of some kind of inborn structure that exists for the purpose of producing a certain pattern. In thinking about Rick's experiment and its interpretation, I had not considered that as a possibility. Whether it is or not, it would take experiment to determine whether such a pattern generator would better model the human than would simple prediction.

I accept the possibility that evolution has provided some body-forms with useful pattern generators (I think of leg motions in a millipede as one such possibility). But on the whole, when I think of pattern generators -- if I do -- I think of them as being developed through the individual organism's interaction with the environment, either, as you suggested, by the observation of repetitive patterns that arrive as disturbances from the environment, or as repeatable consequences of the organism's own output (as in the Artificial Cerebellum, if I now understand it correctly).

This thread is branching out in interesting ways. But we shouldn't lose sight of the main thrust, which was whether prediction could be (or had been) shown to be a possible means of improving control. Whether prediction or some other mechanism is used in any particular circumstance is interesting, but only after it has been accepted that prediction could actually improve control. I think Rick's modelling does argue that it could, even if he did approximate a linear predictor by a small phase advance.

Martin

[Fred_Nickols3]

I wonder how the "prediction" aspect of the fire control problem fits into
all this. Consider these four values (as best I can recall them):

R1 = current range to target
R2 = range to the target at the time of firing the gun (now plus dead time)
R3 = some intermediate value I never paid any attention to
R4 = range to the target at the time the projectile gets there

Only R1 is measured; all the rest are calculated based on the relative
motion between the target and the firing ship as well as the rate of change
in the various angles and distances (trigonometry, plane geometry and both
forms of calculus enter into those calculations)

R2 and R4 were both known as "predicted range" but both were simply
calculations, extrapolations of current patterns and changes in those
patterns.

I assume that if we can engineer gunnery systems, including radars and
computers, that we do similar kinds of things ourselves. I can "predict"
that the stock market will go up or down or remain flat and I can do so on
the basis of hunch, insight, inside information or, like the fire control
problem, on the basis of calculations. I can "predict" that Marc Abrams
will post a snotty email. I can "predict" .....

We also "predict" contingently, that is, we say "if condition Y persists, it
will lead to condition X". For example, I recently moved back to Ohio from
Florida. As the truck was unloading here in Ohio, I noticed that the marble
top to an antique table had been set on the edge of the door through which
the movers were unloading. I thought to myself that I'd better point that
out to the movers but just as I started to do so, they knocked it off the
truck and it broke in half. I predicted just fine but I acted too late.

So, to go back to Bill's comments below, I think we use "predict" in lots of
ways; chief among them is to visualize or describe the future based on a
projection or extrapolation of current conditions. That visualization or
description can be based on lots of things; calculations, other forms of
reasoning and logic, or just plain gut feel, and it might or might not
involve contingencies or other factors.

But is any special accommodation required in PCT to allow or account for
"prediction"?

I don't think so. It seems to me that "prediction" is handled just fine by
the way a control system works.

Regards,

Fred Nickols

···

-----Original Message-----
From: Control Systems Group Network (CSGnet)
[mailto:CSGNET@listserv.uiuc.edu] On Behalf Of Bill Powers
Sent: Tuesday, February 15, 2005 10:33 AM
To: CSGNET@listserv.uiuc.edu
Subject: Re: Prediction

From Bill Powers (2005.02.15.0816 MST)

Rick Marken (2005.02.13.0930)--

>I think it just _seems like_ prediction is involved in control when
people
>are controlling variables that are defined over time, like the sine wave
>pattern.

It may be more fruitful to turn this on its head. What do we mean by
"predicting?" One meaning is surely to make a statement or form an
expectation about something that hasn't happened yet. But another is
simply
to act on the basis of the extrapolated future state of some variable --
i.e., its first derivative times some adjustable constant, added to its
present state. That's how "anticipation" works in heating and cooling
systems, and elsewhere. The effect of prediction or anticipation in this
latter sense is, as we often find in control systems, the opposite of what
intuition suggests. The action is slowed rather than accelerated.
Engineers
just use words like "first derivative" and get away from all the unwanted
connotations and incorrect implications of informal language. We waste a
lot of time arguing about what informal language "really means." It
doesn't
usually have very clear meanings. The real question is, "When I use words
like predict and anticipate and expect, what am I really trying to
describe?" There is a good reason why scientists develop specialized
languages (mathematics being one). It's the only way to be clear about
what
you mean.

Cognition is described by saying it's any thought. However, "cogito" means
"think", so this explanation really says that thought is described by
saying it's any thought. To that we can add that intelligence means being
smart, and the reason that sleeping powder makes you drowsy is that it
contains a dormitive principle.

Best,

Bill P.

[Bill_Powers1]

[From Bill Powers (2005.02.15.1009 MST)]

Fred Nickols: At 09:07 AM 2/15/2005, you wrote:

I assume that if we can engineer
gunnery systems, including radars and

computers, that we do similar kinds of things ourselves.

… We also “predict”
contingently, that is, we say “if condition Y persists, it will lead
to condition X”.

We can learn a lot from what is required for radar-directed gunnery and
other similar things. For these uses of prediction to work, we must have
very good physical knowlege, good models of the environment, good
knowledge of mathematics. We need very powerful computers with high
accuracy, and we need high accuracy and consistency in both the required
sensory information and the required means of action. In fact, what you
witnessed of the radar-directed gunnery was, at the time, the very best
that the most advanced hierarchical control systems of all – human
beings – were able to accomplish using that mode of control. Human
beings need those scientific and engineering tools to be able to use
prediction for controlling in situations like this. Those systems are
what you refer to when you say human beings can do those things
themselves. But they can’t do it without those aids.

When people try to exert predictive control without artificial aids, they
have to settle for relatively poor control. Naked-eye estimates are
nowhere near as accurate as measurements made with scientific devices.
The human ability to make mental calculations is very limited, and for
anything at all complicated almost nonexistent. And the ability to
produce precise actions once the required action has been calculated is
also limited. We do best at predictive control, at calculating actions
and executing them or predicting effects and altering them, when we’re in
a circumscribed environment with simple reproducible properties, as in
throwing a ball in still air through a hoop that’s not too far away, or
rolling a ball down a smooth level well-lit bowling lane, or hitting a
standardized ball with a standard piece of equipment into a fairly large
area that is reasonably close. Any disturbances that happen are normally
fully effective if they occur in an interval of a few minutes. We can
compensate for gradual changes in the environment, like a change in wind
direction, a setting sun, or the slow deterioration of the oil coating on
the bowling lane. But anything that alters the situation in a shorter
interval simply overtaxes the bandwidth of the predictive system

We can count those artificial systems as part of the human control
hierarchy, for it is the ability to devise, construct, and operate those
systems that constitutes the control systems at whatever levels are
involved – sixth or seventh order and upward, by my
guesstimate.

Note the time scales we’re talking about. It takes years to learn the
skills needed to design, build, and operate such systems. Even after they
are built, we’re talking about control processes that span many seconds,
and are almost incapable of counteracting unexpected disturbances that
last for times less than minutes or even hours. Accurate predictive
control systems of any complexity are generally SLOW. And predictive
control systems using only unaided human senses, human mental capacities,
and human motor skills are only marginally useful. Often they get us far
enough off the track that it would have been better not to have used them
at all – observe the consequences of acting inadequately on the basis of
half-worked-out predictions derived from the fuzzy observing procedures
called military intelligence.

Best,

Bill P.

[Martin_Taylor2]

[Martin Taylor 2005.02.15.16.21]

[From Bill Powers (2005.02.15.0816 MST)]

... What do we mean by "predicting?" One meaning is surely to make a statement or form an expectation about something that hasn't happened yet. But another is simply to act on the basis of the extrapolated future state of some variable -- i.e., its first derivative times some adjustable constant, added to its present state. That's how "anticipation" works in heating and cooling systems, and elsewhere. The effect of prediction or anticipation in this latter sense is, as we often find in control systems, the opposite of what intuition suggests. The action is slowed rather than accelerated.

In several messages you have talked as if the gunner is predicting the slew rate of the gun so as to lay it on a stationary target. Most predictors would, I think, want to determine where the target is most likely to be at the time the shell arrives, and lay the gun to aim at that point. Adding a constant times the derivative to the perception of the target (i.e., to the reference signal) is equivalent to subtracting it from the perception of the cursor.

Even intuitively, adding an illusory advance to magnitude of the controlled variable seen by the comparator would seem obviously to warrant slowing the actual controlled variable to bring it more in sync with the reference. The action is slowed if you believe you've already done it!

Whether by computation or just "by eye", some benefit is likely to be had by using the probable future location of the target rather than its present location. I don't think intuition is wrong in this.

Time scales do matter, but the same logic applies whether the time scale is decades or parts of a second. If you can judge better than chance what is likely to be the case if you continue to act as you are, you have an opportunity to improve the effects of your actions in bringing the situation to what you expect to want it to be at that future time. It's why we (or some of us) worry now about global warming that isn't likely to be a disaster until our grandchildren are grown.

Martin

[Bill_Powers1]

[From Bill Powers (2005.02.16.0220 MST);

[Martin Taylor 2005.02.15.16.21]

In several messages you have talked as if the gunner is predicting the slew rate of the gun so as to lay it on a stationary target. Most predictors would, I think, want to determine where the target is most likely to be at the time the shell arrives, and lay the gun to aim at that point. Adding a constant times the derivative to the perception of the target (i.e., to the reference signal) is equivalent to subtracting it from the perception of the cursor.

I'm not sure what I said that sounded like that. I would say the gunner (or rather, the gun-control computer) is computing where the target is going to be at the time the shell reaches the same distance and altitude. This is clearly a problem in simultaneous nonlinear differential equations, because where the shell is going to be at various times depends nonlinearly on where the gun is pointed, and where the gun is pointed will depend nonlinearly on where the shell and target will be at closest approach. The solution will be that pointing angle and firing time at which the shell and target will both be in the same place at the same time. I don't think any gunner in the world can solve that problem in his head with any accuracy.

Even intuitively, adding an illusory advance to magnitude of the controlled variable seen by the comparator would seem obviously to warrant slowing the actual controlled variable to bring it more in sync with the reference. The action is slowed if you believe you've already done it!

Yes, that is how rate feedback works. This came up specifically in the conversations with Flach in which he mentioned stopping a car at a specific place as an example of control that requires calculating required actions and then executing the plan. I immediately replied with a model of a driver stopping a car in which the driver controlled the sum of rate of approach to the target point and actual distance (assuming the perceptions were accurate) by varying the brake pressure. I added a second control system that controls the maximum deceleration (deceleration can be separately sensed) so we could see the effects when the driver thought the road was slippery. Bruce Abbott expanded on that model and improved it. Increasing the multiplier on velocity feedback does indeed slow the approach to the final location. The reference level determines how close to the target position the braking begins, and the rate gain determines the shape of the negative-exponential braking curve. A wide range of characteristics allows the car to come to a stop very close to the selected position, along various space-time trajectories, controlling nothing more than the sum of distance and rate of change of distance.

We can say that the rate of change control is a member of the class "predictions", but that does not share many characteristics with other sorts of predictions, like predicting that the Red Sox will win the pennant. Certainly there is a vast difference in complexity (and correctness, all years but one).

Whether by computation or just "by eye", some benefit is likely to be had by using the probable future location of the target rather than its present location. I don't think intuition is wrong in this.

I agree, if we're talking about controlling for a collision using a ballistic projectile. This is not, however, the way to design a homing missile. A homing missile should try to find the heading angle at which the target appears to remain at a constant bearing in x and y. No prediction is needed -- or perhaps more accurately, no prediction other than the action of the whole control system is needed. For this kind of homing system is actually solving the simultaneous differential equations by means of maintaining a constant bearing angle. I suppose it's an implementation of the method of variations -- that doesn't quite sound like the right name, but maybe you know what I mean. It's the way they prove that light travels from source, through lens, to focus in the minimum time. I think this method of homing solves those equations using only present and past data, rather than looking at the whole trajectory including the parts that haven't happened yet.

Time scales do matter, but the same logic applies whether the time scale is decades or parts of a second. If you can judge better than chance what is likely to be the case if you continue to act as you are, you have an opportunity to improve the effects of your actions in bringing the situation to what you expect to want it to be at that future time. It's why we (or some of us) worry now about global warming that isn't likely to be a disaster until our grandchildren are grown.

Yes, and this is a good example of how predictive control by computing actions and consequences takes a lot of time, and is not very accurate. It helps if we make the target area as large as possible, like the area within the boundaries of the tennis court, or the range of future conditions that could be called a disaster. But it helps a lot more if we can aim the projectile roughly in the right direction, and then turn on its guidance system which keeps making course corrections ten times per second until it hits the target. Controlling global warming would work a lot better than taking some spasmodic action and then hoping it corrects the problem 100 years from now, ballistically.

When the time scale is down to seconds or fractions of a second, ordinary closed-loop control is by far the best choice. Of course I have trouble getting used to elevating simple rate feedback to the status of "predictive control", because all that's being controlled is the sum of x and dx/dt. If x + dx/dt = constant, you get a nice exponential approach to x = constant. What's to predict?

Best,

Bill P.

[BUSINESS_GAVIN_HERMA]

( Gavin
Ritz 2011.04.07.13.26NZT)

Gavin:

Everything
is some control of energy in a living organism. There is nothing else.

If it was not then E=mc² is nonsense.
Tells us all matter is a form of energy.

AM: I’ve heard of similar interpretations of the E=mc2, but I’ve
also heard of different ones (from mainstream physics).

Sure. There are many interpretations. Energy
is the basis of our scientific thinking, all our thinking is around the laws of
energy, this is inescapable.

If we just equate matter
with energy,

We don’t just equate it. My comment
above says matter is a form of energy.

then there are no ways of
distinguishing them. If matter is

energy, we could say
everything is matter.

That’s if you equate it, science does
not do that.

Then, how would we name
the potential for moving?

Similarly - if everything
is energy, how do we name matter?

If we look at F = ma,
that does not mean that force * is
mass, or is acceleration.

I was using F=ma as an example of a linear
equation.

The equation just shows
how quantities of energy or mass can be calculated, not that they are “the
same”.

They are not the same, they are different forms
of energy. We can convert from one form of energy to the other. The first law
of conservation of energy.

The second law, tells us the ability of
the energy to be used for work, the entropy law. It’s like a growth law.

This is the basis of scientific thought.

(Gavin Ritz 2011.04.07.12.13NZT)

PCT is not a predication tool. Because we don’t
know exactly what each and every individual controls for.

Predication is all good under linear equilibrium
situations. Like F=ma, E=mc2, E=hf, E=mgh.

But non equilibrium situations is a big problem,
it’s a major problem facing all of science.

Why do you write “predicAtion”?

Sloppy spelling, it’s my spell checker,
sorry.

And I kinda disagree.
There would be no use for PCT if it couldn’t predict anything. You can always
predict

that a control system
will work toward it’s goal and if you know the goal - you know what will
happen.

Yes, but that’s not much of a predication.
It’s a heuristic, more like.

You might

not be able to predict
exactly how someone will come to the goal, but just knowing that is also useful

• you

know it might be random
or creative.

Useful yes, predictive no.

If you study the tracking
experiments, you’ll see that tracking in a new situation could be predicted
with a 99% accuracy.

Yes, but in the context of the social
affair predication is nigh impossible, simply because you do not know what the
other person is controlling for.

Regards

Gavin

···

[amatic]

( Gavin
Ritz 2011.04.07.13.26NZT)

Sure. There are many interpretations. Energy
is the basis of our scientific thinking, all our thinking is around the laws of
energy, this is inescapable.

We don’t just equate it. My comment
above says matter is a form of energy.

That’s if you equate it, science does
not do that.

Who do refer to when you say “our scientific thinking”?

You seem to be talking about an alternative quantum mechanics interpretation. I don’t see how it

connects to a few orders of magnitude greater systems, like humans or animals.

Yes, but that’s not much of a predication.
It’s a heuristic, more like.

Useful yes, predictive no.

Yes?

Why do you study PCT then?

Yes, but in the context of the social
affair predication is nigh impossible, simply because you do not know what the
other person is controlling for.

So, your suggestion is to try an “energy approach” which also does not predict anything (not that I agree that PCT doesn’t)?

Adam

···

On Thu, Apr 7, 2011 at 3:43 AM, Gavin Ritz garritz@xtra.co.nz wrote:

[BUSINESS_GAVIN_HERMA]

( Gavin
Ritz 2011.04.08.9.31NZT)

(Gavin Ritz 2011.04.07.13.26NZT)

You seem to be talking
about an alternative quantum mechanics interpretation.

There is no alternative approach to QM.

In terms of Non equilibrium thermodynamics,
try Prigogine, de Lange, De Groot and Mazur.

I don’t see how it

connects to a few orders
of magnitude greater systems, like humans or animals.

Not exactly sure what you are referring to
here. Magnitude of what?

Yes, but
that’s not much of a predication. It’s a heuristic, more like.

Useful yes, predictive no.

Yes?

Why do you study PCT
then?

I enjoy theories of mind thoroughly; it’s
a great hobby of mine.

Why do you study PCT?

Yes, but
in the context of the social affair predication is nigh impossible, simply
because you do not know what the other person is controlling for.

So, your suggestion is to
try an “energy approach” which also does not predict anything

Not sure how you came to this conclusion.

Prediction with linear models is fine;
with complex models it’s quite different. (See Complexity; try Kaufmann,
Bak, Jantsch, Mewes etc)

(not that I agree that
PCT doesn’t)

An energetic approach does have some sort
of prediction. It tells us clearly when a living system will be able to absorb
objectives, (e.g. The Diedeskein) if it will work and under what conditions. Using
the Gibbs Free Energy concept from Chemical thermodynamics.

Ask yourself this.

Where do you think the electrical signal
of the reference signal comes from?

If you answered that, ask this then, where
do the electrical signals from these higher reference signals come from, what are
their sources.

Does this explain everything we need to
known about a living organism’s mental life and mental conditions?

Can we answer this?

Adam, if you have a genuine and honest interest in understanding these
concepts you may contact me on my private email.

Regards

Gavin

?

Adam

[amatic]

There is no alternative approach to QM.

In terms of Non equilibrium thermodynamics,
try Prigogine, de Lange, De Groot and Mazur.

Not exactly sure what you are referring to
here. Magnitude of what?

I just don’t see how analysing thermodynamics is useful in studying human psychology.

I enjoy theories of mind thoroughly; it’s a great hobby of mine.

Why do you study PCT?

I enjoy it too.

I study PCT because it’s a complete theory of brain with lots of evidence supporting it and as far as

I have studied, zero evidence refuting it.

By complete, I mean it connects neurons and synapses, the neurobiology of the brain with the

cognitive processes, the thinking and feeling; in a very straightforward, no-nonsense way which is

very different from many other theories I learn about (being a psychology student).

It allows me to understand myself, other people and brain functioning much better.

I don’t see those properties in the energetic approach you write about.

An energetic approach does have some sort
of prediction. It tells us clearly when a living system will be able to absorb
objectives, (e.g. The Diedeskein) if it will work and under what conditions. Using
the Gibbs Free Energy concept from Chemical thermodynamics.

Ask yourself this.

Where do you think the electrical signal
of the reference signal comes from?

If you answered that, ask this then, where
do the electrical signals from these higher reference signals come from, what are
their sources.

As I understand it, there doesn’t have to be a refference signal on the highest level. The perceptual signal can represent the error signal at

the same time (that is, the reference signal is always zero. No comparator, just an inverter). Perhaps the reference signals are “read out” from

genes. Perhaps there is a different solution.

Does this explain everything we need to
known about a living organism’s mental life and mental conditions?

Well, it’s a part of it. Of course, a lot of other things require explanations. Where does energy come into play?

Adam, if you have a genuine and honest interest in understanding these
concepts you may contact me on my private email.

I’m still trying to figure out where do those concepts connect to brain functioning or PCT.

Best

Adam

···

On Fri, Apr 8, 2011 at 12:14 AM, Gavin Ritz garritz@xtra.co.nz wrote:

[BUSINESS_GAVIN_HERMA]

(Gavin Ritz 2011.04.09.9.07NZT)

I just don’t see how
analysing thermodynamics is useful in studying human psychology.

Adam

What I will do is put a drawing together
with the track analyze demo as an example as you asked to show how it all
works.

I’m just a bit snowed under at the moment.

When you see the drawing it will make
sense.

In the meanwhile if you can get hold of “Energy
in Nature and Society” by V Smil, (not the abridged version) the complete
version is an introduction to the subject, it’s well covered with a lot
of data and good information.

In the early 90’s I was introduced
to (Engpass Konzentrierten Strategie-EKS) Energy Bottleneck Strategie, written
by Wolfgang Mewes and very high profile German Cybernetician. Almost unknown outside
Germany.

In this massive work he showed how to (amongst
how to do strategy in business) unleash market opportunities and how psychological
warfare is waged (and what not to do here) with detailed accounts and blueprints
of how to do these things.

When things started to work extremely well
for me both personally and financially I wanted to understand it better, hence
this is now my hobby.

Regards

Gavin

···

On Fri, Apr 8, 2011 at 12:14 AM, Gavin Ritz garritz@xtra.co.nz wrote:

[amatic]

Adam

What I will do is put a drawing together
with the track analyze demo as an example as you asked to show how it all
works.

I’m just a bit snowed under at the moment.

When you see the drawing it will make
sense.

Looking forward to that,

Best

Adam

···

On Fri, Apr 8, 2011 at 11:27 PM, Gavin Ritz garritz@xtra.co.nz wrote: