time scales

[Jeff_Vancouver2]

[from Jeff Vancouver (2002.10.23.1650)]

I have a question about the issue of time scales and the hierarchy. On pg.
52 in B:CP Bill begins to describe the idea that higher level control units
must operate slower than lower level units or else the system will be
unstable. He then states that the way this is accomplished is via the
averaging over time of inputs, where the amount of time is a function of the
level in the hierarchy (and what I will label as lag). I am sure if this is
creating a disturbance for anyone, they will let me know. Anyway, what I am
unclear about is whether it is the control unit that has to have the
increasingly slower lag as one moves up the hierarchy, or whether changes in
the functions (i.e., reorganization) must have increasingly slower lags? Or
even whether it is simply likely that subgoals take less time to achieve
than superordinate goals such that the timing comes from the lags in the
environment and not the lags in the control unit functions.

Hence, question number one is, would a hierarchical system of control units
really need to be time organized? Question 2 is, if they do need such order,
how does it come about? Specifically, if I acquired a control unit that
sought a perception of 20 widgets built, from whence would that unit acquire
its time scale? Question three is, what is the likely time scale of the 11th
level? Are we talking seconds, minutes or days?

Looking for help,

Jeff

[Isaac_Kurtzer1]

i.kurtzer (2002.10.23.1815EST)
Jeff, I think the short answer is no one knows, put a grad student on it.
The less obnoxiuos, but true, response is that the control processes have
to be slower so much that they stably drive the references to lower order
systems, program it.
As for the very theoretical "11th order" question,
ask yourself how often you have changed religions. Once every couple
seconds, hours, years, or maybe twice.
I'd guess the last one. I'd guess significant high level changes are very
rare and are preceded by chronic and/or large errors.
You could find this out a bit more by asking 30 people if they ever changed
religions and what what going on beforehand. I'd goes noone would say
something like, "I love kosher food so i became Jewish". That could be
taken care of by going to a deli in any civilized city, , and
that control resides at a
much lower level, program/etc. Rather, a
system concept change would occur when all the lower evels couldn't take
care of business; see war, famine, bad things happen to good people, i
don't want to die, etc.
That seems right to me.

Isaac

···

On Wed, 23 Oct 2002, Jeffrey B. Vancouver wrote:

[from Jeff Vancouver (2002.10.23.1650)]

I have a question about the issue of time scales and the hierarchy. On pg.
52 in B:CP Bill begins to describe the idea that higher level control units
must operate slower than lower level units or else the system will be
unstable. He then states that the way this is accomplished is via the
averaging over time of inputs, where the amount of time is a function of the
level in the hierarchy (and what I will label as lag). I am sure if this is
creating a disturbance for anyone, they will let me know. Anyway, what I am
unclear about is whether it is the control unit that has to have the
increasingly slower lag as one moves up the hierarchy, or whether changes in
the functions (i.e., reorganization) must have increasingly slower lags? Or
even whether it is simply likely that subgoals take less time to achieve
than superordinate goals such that the timing comes from the lags in the
environment and not the lags in the control unit functions.

Hence, question number one is, would a hierarchical system of control units
really need to be time organized? Question 2 is, if they do need such order,
how does it come about? Specifically, if I acquired a control unit that
sought a perception of 20 widgets built, from whence would that unit acquire
its time scale? Question three is, what is the likely time scale of the 11th
level? Are we talking seconds, minutes or days?

Looking for help,

Jeff

[Isaac_Kurtzer1]

Yikes, I should've said the nth level reference(s) will change as the error
signal(s) of the nth+1 level changes.

Isaac

···

On Wed, 23 Oct 2002, Marc Kurtzer wrote:

i.kurtzer (2002.10.23.1815EST)
Jeff, I think the short answer is no one knows, put a grad student on it.
The less obnoxiuos, but true, response is that the control processes have
to be slower so much that they stably drive the references to lower order
systems, program it.
As for the very theoretical "11th order" question,
ask yourself how often you have changed religions. Once every couple
seconds, hours, years, or maybe twice.
I'd guess the last one. I'd guess significant high level changes are very
rare and are preceded by chronic and/or large errors.
You could find this out a bit more by asking 30 people if they ever changed
religions and what what going on beforehand. I'd goes noone would say
something like, "I love kosher food so i became Jewish". That could be
taken care of by going to a deli in any civilized city, , and
that control resides at a
much lower level, program/etc. Rather, a
system concept change would occur when all the lower evels couldn't take
care of business; see war, famine, bad things happen to good people, i
don't want to die, etc.
That seems right to me.

Isaac

On Wed, 23 Oct 2002, Jeffrey B. Vancouver wrote:

> [from Jeff Vancouver (2002.10.23.1650)]
>
> I have a question about the issue of time scales and the hierarchy. On pg.
> 52 in B:CP Bill begins to describe the idea that higher level control units
> must operate slower than lower level units or else the system will be
> unstable. He then states that the way this is accomplished is via the
> averaging over time of inputs, where the amount of time is a function of the
> level in the hierarchy (and what I will label as lag). I am sure if this is
> creating a disturbance for anyone, they will let me know. Anyway, what I am
> unclear about is whether it is the control unit that has to have the
> increasingly slower lag as one moves up the hierarchy, or whether changes in
> the functions (i.e., reorganization) must have increasingly slower lags? Or
> even whether it is simply likely that subgoals take less time to achieve
> than superordinate goals such that the timing comes from the lags in the
> environment and not the lags in the control unit functions.
>
> Hence, question number one is, would a hierarchical system of control units
> really need to be time organized? Question 2 is, if they do need such order,
> how does it come about? Specifically, if I acquired a control unit that
> sought a perception of 20 widgets built, from whence would that unit acquire
> its time scale? Question three is, what is the likely time scale of the 11th
> level? Are we talking seconds, minutes or days?
>
> Looking for help,
>
> Jeff
>

[Bill_Powers1]

[From Bill Powers (2002.10.23.1908 MDT)]

Jeff Vancouver (2002.10.23.1650)]

I have a question about the issue of time scales and the hierarchy. On pg.
52 in B:CP Bill begins to describe the idea that higher level control units
must operate slower than lower level units or else the system will be
unstable. He then states that the way this is accomplished is via the
averaging over time of inputs, where the amount of time is a function of the
level in the hierarchy (and what I will label as lag). I am sure if this is
creating a disturbance for anyone, they will let me know. Anyway, what I am
unclear about is whether it is the control unit that has to have the
increasingly slower lag as one moves up the hierarchy, or whether changes in
the functions (i.e., reorganization) must have increasingly slower lags?

I would have said then that the control units themselves must have
increasingly slower responses. The reason has do to with phase shifts and
gains, which must satisfy specific criteria if a control system is to be
both dynamically stable and capable of accurate control. I still think that
is the case, though I wouldn't say any longer than the only place where the
required time filtering can be done is on the input side.

However, your guess is also relevant, as Isaac points out. Higher systems
are, I would guess, slower to reorganize, not for any reasons having to do
with dynamic stability, but simply because all the lower systems involved
must also be readjusted as a superordinate system changes. The higher the
system, the more parameters have to be adapted to the higher-level changes.
You don't, as Isaac said, change religions overnight. And once you have
nominally changed such a thing, the implications at lower levels can take a
long time to be recognized and worked out. The latter readjustments are not
necessarily the e-coli type of reorganization -- they could be systematic
and rational -- but they must still happen before all inner conflicts can
be done away with.

>Specifically, if I acquired a control unit that

sought a perception of 20 widgets built, from whence would that unit acquire
its time scale?

I presume you mean 20 widgets in a specific length of time, like per hour.
The time scale is determined by the most rapid disturbance that must be
counteracted and the most rapiod changes that must be created, and the
fundamental properties of the existing system at lower levels. It is
possible that the fundamental delays in already-organized control systems
are such that the new control system simply cannot be adjusted to react
rapidly enough without becoming dynamically unstable -- going into
self-sustained oscillations which might even reach destructive magnitude.In
that case you simply can't make widgets as fast as you hoped, at least in
the way you were thinking of doing it. Maybe you need two different control
systems making 10 widgets per hour each.

Question three is, what is the likely time scale of the 11th
level? Are we talking seconds, minutes or days?

I probably didn't make this clear in B:CP, but the only time scales we can
generalize about are the _shortest_ ones at each level. Given control
systems with certain inherent delays up to a given level, the next added
level (in a given context) can't operate faster than some limit which is a
little slower than all the existing systems below it in the relevant "tree"
of subsystems. However, there is nothing to say that a higher system may
not operate much slower than that. Just consider two sequence control
systems, one which controls the sequence of sounds we call
"shave-and-a-haircut, six bits", taking about three or four seconds to
complete, and another sequence of sounds that we call Beethoven's Fifth
Symphony, which takes, I suppose, something like 40 minutes to an hour to
produce, not counting rehearsals. There is, I am sure, a limit on how fast
we can produce and correct errors in the shortest possible sequence, say a
sequence of two notes, and this speed will, I am equally sure, be slower
than the fastest speed at which we can correct errors in the simplest
possible relationship, like cursor on target. However, some sequences might
take a great deal longer to carry out and control, and it is also possible
that some very complex relationships might take a lot longer to control
than a simple sequence would take. The rule I had in mind is simply that in
any specific case, we will not find a higher-level element being controlled
more rapidly than any of the lower-level elements on which it depends. If
you think about it, that's almost self-evident.

If we're talking only about the simplest possible control processes at each
level, I would say that it might require only a second, plus or minus, to
recognize and react to a disturbance at the highest level, system concepts.
However, this could occur only if the particular processes involved at
lower levels, from principles on down, could react even faster than that.
It could not occur if the highest control process were something like
seeking justice for the oppressed, involving very complex and numerous
principles and procedures.

There are both simple and complex control processes going on at every
level.The temporal considerations apply only to _minimum_ times, not to all
possible control processes.

Best,

Bill P.

[Bill_Powers1]

[From Bill Powers (2002.10.24.0420 MDT)]

Jeff Vancouver (2002.10.23.1650)--

I think that in the discussion of time scales in the hierarchy, we have
been overlooking something of far more importance than the fact that higher
level systems operate more slowly than the lower systems with which they
are involved. Back when I was first exploring these relationships, the
reason that these time scales seemed important was that when they were
observed, they could be used as evidence about the apparent fact that some
perceptions depend on others. If perception B depends on some set of
perceptions A1, A2, and so on, then clearly perception B could not possibly
change any faster than A1, A2, and so on changed. Therefore control of
perception B would have to occur on a time scale slower than that on which
the A perceptions changed. The stability considerations were really
secondary to that basic fact.

An example is the relationship-perception we call distance. The distance
between two objects depends on their individual positions in a common
spatial framework.It is quite impossible for the distance between the
objects to change without any change in the configuration containing the
two objects. As a physicist, I had always taken this situation for granted
as a property of the universe in which I lived. Considered as a fact of
experience, however, it is clearly not about the external world so much as
about the way human perceptions are organized. The perception of distance
clearly exists, and so do perceptions of positions within a spatial
configuration.But what sort of existence does distance have apart from the
objects or positions that are distant from one another? None at all!
Distance is never perceived without the things that have distance between
them. Therefore distance is a _derived perception_ and not something we
sense directly in the external world. It is derived, somehow, from our
perceptions of positions or configurations and could not exist without them.

That is what I was looking for in looking for hierarchical relations:
perceptions that depended on other perceptions and existed only if the
other perceptions existed. The fact that such dependencies can be found is,
if you pause to think about it, rather astonishing. Have you ever read
about these dependencies in any textbook of psychology or any treatise on
perception? I'm sure that in the old Gestalt studies there were things
being talked about of a similar nature, and surely it has been recognized
that perceptions range from the concrete to the abstract, but have you ever
read anywhere that the perception of distance itself, as an aspect of the
experienced world, _depends on_ the perception of position, another aspect
(apparently) of the experienced world? I never have.

Always, before these ideas developed, when I saw a relationship like
distance, I assumed I was perceiving something in the real world outside
me, directly. It never occurred to me that I had to perceive or imagine
positions before it was even possible to perceive distance. Yet now that I
had had this idea, it was inescapable: distance is clearly something
derived not directly from the outside world, but from lower-level
perceptions of that world. And if that is true of this one example, what of
those lower-level perceptions? Are they, too, derived from some still lower
kind of perception? Are there, in fact, ANY perceptions that are derived
directly from contact with the outside world? The end of that chain was
the discovery of the intensity level, which seems to be the lowest of all.

It was the dependence of some perceptions on others that struck me as being
a fact of human nature that had not been recognized as such in psychology.
It was this dependence that let to the concept of a hierarchy of control,
in which control of one perception required varying the reference levels of
systems that controlled the lower perceptions on which the higher one
depended, of which the higher one was a function. The whole structure of
HPCT follows from this hierarchical dependency of some perceptions on
others. In B:CP I spent a lot of time trying to show examples of this kind
of dependency, but somehow I think the real message got lost among the
details. I may not even have realized fully what I was trying to do, to lay
out a new system concept of the phenomenon of perception itself, and thus
of our relationship to what we think of as the physical universe.

In fact, as I reflect on the development of that book, I wonder whether I
was ever conscious of the real idea that was driving it. It could be that I
have only become aware of its importance over the intervening years, so
that now the message seems to stand out vividly against the background of
detail -- detail that no longer seems as important as it once did.

Best,

Bill P.

[Richard_Marken2]

[From Rick Marken (2002.10.24.1130)]

Bill Powers (2002.10.24.0420 MDT)–

It was the dependence of some perceptions on others
that struck me as being

a fact of human nature that had not been recognized as such in psychology.
OK. Make that two Nobels. One for showing that behavior is control
(and all that implies: controlled perceptual variables, behavioral illusion,
the test for controlled variables, etc.) and the other for showing the
hierarchical nature of perception. Bravo!!
For those who are interested in “seeing” the perceptual hierarchy, I
refer you to my “Hierarchy of perception and control” demo at http://home.earthlink.net/~rmarken/ControlDemo/HP.html.

This demo works better on some computers than on others. But I think
the “Medium” and “Slow” presentations work on all computers. I you press
the “Medium” button you should be able to see objects (squares and circles
o different sizes) apparently moving clockwise and then counter clockwise.
So at this speed of presentation you can see diction of motion, which might
qualify as a “transition” perception. It turns out that the objects
are occurring in a sequence of different sizes; small-medium-large
and vice versa. But at the medium speed of presentation the sequence perception
is not yet visible. If you then press the :Slow" button the sequence perception
is easily visible (though it was there all the time, at medium speed; it
was just occurring at too fast a rate for the sequence perception/control
systems.

The demo doesn’t really illustrate the dependance of higher level
on lower level perceptions. But it does illustrate the temporal consequences
of this dependence. Since the perception of sequence depends on the ability
to perceive a transition (from one configuration to another) then can be
no perception of sequence before there is a perception of transition.
If the transition occurs at a fast enough rate (as is the case when the
speed is “Medium”) the transition perception can exist before the sequence
perception can be constructed. So by appropriate varying the speed of perceptual
components we can look at separate layers of the perceptual hierarchy.

You can read more about this demo and the concept of a perceptual
hierarchy in chapter 3 of More Mind Readings.

Best regards

Rick

···

–

Richard S. Marken, Ph.D.

The RAND Corporation

PO Box 2138

1700 Main Street

Santa Monica, CA 90407-2138

Tel: 310-393-0411 x7971

Fax: 310-451-7018

E-mail: rmarken@rand.org

[Jeff_Vancouver2]

[From Jeff Vancouver (2002.10.25.0955 EST)]

To Issac, Bill, Rick and whoever else is interested,

It seems that we have to unpack this hierarchy concept a bit. Bill says:

[From Bill Powers (2002.10.24.0420 MDT)]

It was this dependence that let to the concept of a hierarchy of control,
in which control of one perception required varying the reference levels of
systems that controlled the lower perceptions on which the higher one
depended, of which the higher one was a function. The whole structure of
HPCT follows from this hierarchical dependency of some perceptions on
others.

This is the primary answer I expected from reading B:CP. It dictates that
the level in the hierarchy of a focal unit is DEFINED by the source of the
inputs to the input function of the unit (and hence, this is how the
hierarchy is determined). For example, a 2nd level unit is 2nd level because
it creates its perception from first level signals. Given that it might be
possible that a 1st level perception AND a 2nd level perception are combined
in some focal unit, we might use the highest signal source (2nd level in
this case) to define the level of the focal unit (3rd).

If I understanding the implications of this, I do would not need to model
some lag into the input function due to level. They could all simply operate
at the same speed inherently, but the signals would be paced by the fact
that they are passed at some speed across the neural net (i.e., the
wetware).

On the other hand, given that the perceptions are refering to variables with
environmental lags (some things take longer to change than others
physically) plus the gain of the system can determine speed of change (i.e.,
a powerful engine can counter drags more readily than weak engines) we would
want the input function to account for this or else the system would be
unstable. Right? This corresponds to the Widget example Bill elaborated on.
The issue is that this property is designed into engineered systems, but how
is it built into human systems? Perhaps for now the answer is, we do not
know.

Some more unpacking:

Now another issue is that the hierarchy is often described in terms of the
output side. That is, to meet some perception, some other perception must be
met. So for example, in order to perceive myself receiving a diploma, I must
perceive myself taking classes and doing well in them. But there is a
disconnect between the first way of defining the hierarchy and this second
one. Specifically, one can physically see the diploma being received, and
understand what that means, even though the perception of some step towards
it (e.g., some class one thought was needed, but was not actually) was not
met. The perceptions needed to perceive "receiving diploma" and the meaning
that diploma represents do not really need to be slower than the time it
takes to see oneself passing a class.

Indeed, the example in B:CP is questionable. I could perceive an insult the
second I receive it (not a millisecond, because it will take time for the
vibrations to be translated into sound, into words, into meaning, into
insult, but this will be a pretty rapid process).

More generally, the issue is that the recipient control systems of reference
signals are not necessarily the source of the perception signals from which
the focal control system derives its perception. Hence, if both means are
used to define the hierarchical level of a focal system, one might find
conflicting values. It seems perception should take precedent here, but the
important point is that I do not think we can interchange the means of
determining hierarchy.

Finally, there is the issue of change to the control system functions (e.g.,
a change in one's religion). In this case the examples described why a
system might not want to change higher level ones before determining if
changes to lower level systems can reduce total system error (or even a
focal unit's error). While this seems correct, it speaks to an intelligent
design (or long term trial and error), but not to how the system "know" to
do this itself. Thus, I return to the original question of how does the
control system acquire its lag in terms of reorganization order? Perhaps for
now the answer is, we do not know.

Jeff