How can perceptual hierarchy be tested?

CSGnet Archive CSG2003
1

[Martin Taylor 2003.11.29.0912]

[David Goldstein (11,29.2003.0637)]
How does one determine that two experiences are at different levels?...

How does one determine that one experience is necessary for another
experience to take place? Even if an experience B is at a higher level,
does experience A evoke experience B as a component evokes a whole? Or
are we talking about an arbitrary memory association thing?

Even if A is necessary for B, it doesn't follow that they are on
different levels. B may be necessary for C, and C for A in a loop.
Conversely, A may be a component of a higher-level B, but not be
necessary for B. A may facilitate B. And the same can happen in
loops. A may facilitate B and produce B if X is present, B may
facilitate C if Y is present, and C may facilitate A if Z is present.
Such loops can be self-organizing (see
Mutuality Home Page for a more extended
discussion of how this can happen)

For there to be a strict hierarchy, there must be no loops. If one
weakens the idea of "hierarchy", all elements of a loop must be at
the same level, but there are no connections that go from higher
levels to lower levels in the actual formation of a perception. Some
time ago, I did put forward the proposition that the mechanism of the
"category level" of such a "weak hierarchy" was the self-organized
formation of such mutually facilitatary loops at a single level. A
"category" perception was presumed to occur when a loop was
sufficiently excited by data coming from lower levels to be
self-sustaining.

A still weaker notion of a level-based structure (which probably no
longer deserves the name "hierarchy") is to count, for any
perception, the minimum number of processing stages that must
separate the sensory input from that perception. In such a structure,
other processing pathways with more stages may be involved in a
particular perception, so one could ask the question in the other
way--what is the maximum number of processing stages in any necessary
pathway? In a strict hierarchy, those two questions give the same
answer. Otherwise they don't.

So David's question is a real one: "How does one determine that two
perceptions are at different levels." I think also that one must ask
"How can one investigate whether the hierarchy is strict, in whole or
in part?" It could be, for example, that there is a strict hierarchy
below the barrier (assuming there is one) between analogue levels and
logical levels, but that the barrier consists of self-facilitating
loops of perceptions at the same level. Such a structure would be a
mixture of a strict and a weak hierarchy.

I don't know how one would gather evidence to define the structure of
the putative hierarchy. Bill's approach is good for setting up a
hypothesis that is presumably testable, but as he has repeatedly
stated, that's all it has ever been intended to be. The question is
how to go about actually testing it.

I don't have an answer for that.

Martin

2

from [Marc Abrams (2003.11.29.1116)]

[Martin Taylor 2003.11.29.0912]

Even if A is necessary for B, it doesn't follow that they are on
different levels. B may be necessary for C, and C for A in a loop.
Conversely, A may be a component of a higher-level B, but not be
necessary for B. A may facilitate B. And the same can happen in
loops. A may facilitate B and produce B if X is present, B may
facilitate C if Y is present, and C may facilitate A if Z is present.
Such loops can be self-organizing (see
Mutuality Home Page for a more extended
discussion of how this can happen)

Martin where have you been? :-). In my view this is _extremely_ important
material. There are possible alternatives to the hierarchy I think are
worthwhile of discussion.

_EVERYONE_, please go to Martin's website and review his material. This
material should provide us with a very strong base for discussion.

Martin, when was this material discussed on CSGnet? What ever became of it?

For there to be a strict hierarchy, there must be no loops.

Yep.

A still weaker notion of a level-based structure (which probably no
longer deserves the name "hierarchy") is to count, for any
perception, the minimum number of processing stages that must
separate the sensory input from that perception. In such a structure,
other processing pathways with more stages may be involved in a
particular perception, so one could ask the question in the other
way--what is the maximum number of processing stages in any necessary
pathway? In a strict hierarchy, those two questions give the same
answer. Otherwise they don't.

Yes, this would provide a test for the hierarchy. Wonderful Martin. It would
also provide a blueprint for how a network might be configured for each.

A corrollary to Martin's statement above is, _exactly_ what _kinds_ of
'inputs' are needed. There are ,many more than our 5 senses.

So David's question is a real one: "How does one determine that two
perceptions are at different levels." I think also that one must ask
"How can one investigate whether the hierarchy is strict, in whole or
in part?" It could be, for example, that there is a strict hierarchy
below the barrier (assuming there is one) between analogue levels and
logical levels, but that the barrier consists of self-facilitating
loops of perceptions at the same level. Such a structure would be a
mixture of a strict and a weak hierarchy.

Certainly a possibility and I think a strong one.

I don't know how one would gather evidence to define the structure of
the putative hierarchy. Bill's approach is good for setting up a
hypothesis that is presumably testable, but as he has repeatedly
stated, that's all it has ever been intended to be. The question is
how to go about actually testing it.

I don't have an answer for that.

What about this initial statement in this post;

For there to be a strict hierarchy, there must be no loops. If one
weakens the idea of "hierarchy", all elements of a loop must be at
the same level, but there are no connections that go from higher
levels to lower levels in the actual formation of a perception. Some
time ago, I did put forward the proposition that the mechanism of the
"category level" of such a "weak hierarchy" was the self-organized
formation of such mutually facilitatary loops at a single level. A
"category" perception was presumed to occur when a loop was
sufficiently excited by data coming from lower levels to be
self-sustaining.

Wouldn't this provide a 'test' or at least provide a way of examining the
efficacy of a pure hierarchy?

A wonderful post. Thanks Martin.

David G, Now Martin T. Bruce G, you made it happen. A big wet sloppy kiss to
you. :slight_smile:

_THIS_ folks, is what I have been after. NOTHING more.

Marc

3

[From Bill Powers (2003.11.29.0931 MST)]

Martin Taylor
2003.11.29.0912–

Such loops can be self-organizing
(see

http://www.mmtaylor.net/PCT/Mutuality/index.html
for a more extended

discussion of how this can happen)

A wonderful presentation. You know me, I’m not much for spin-glasses and such physicists’ fantasies, but you have a wonderful way of organizing ideas and the whole presentation is worth anyone’s time to read and ponder. What a great document! Of course I downloaded all the pieces before I read the line saying that I could get them all at once.
I suggest that you try simulating those flip-flops.
There’s one approach to the hierarchy which is (after one sketchy reading) apparently omitted fron your paper: the evolutionary processes which result in there being partially-organized brains at birth. My intent with the “ll levels” was to find relationships among perceptions that seem so utterly necessary that we have to take them for granted – and that are the same for all human beings. This would not seem to relate to phenomena of reorganization in a single lifetime, though it might constrain the possibilities for reorganization.
I have, by the way, checked with a Chinese philosopher-psychologist (Zhiang Huaxia) who is going to publish B:CP in Chinese (South China Normal University), and allowing for translations, he says that the definitions of levels in MSOB are quite in line with Chinese experience. As I say, allowing for translations, that’s rather encouraging.
If we ever do find levels of organization that are universal among human beings, it seems to me that there would be only two ways to explain them. One would be that the objective universe is organized in exactly the same terms as human perceptions, and the other would be that evolution long ago produced tendencies to organize in these 11, or however many, ways, because pragmatically they are sufficient to allow survival. If you take the long view on reorganization, these two possibilities may be one, or at least they may be closely related.
One of the main criteria in defining my levels is that a perception of a given level could not exist at all without perceptions from at least one lower level. This “inclusionary” principle doesn’t say anything about other types of influences among perceptions, which I am happy to consider. It just provides a sort of minimum requirement that must be satisfied before we can get more complicated. I think that in most of the pairs of levels, this requirement is met, although there remain a few fuzzy places.

Another subject a physicist should consider is “fine structure.” It is possible that within some of the levels I propose there can be sublevels, all sharing the main characteristic of the level but having a hierarchical (or other) structure within the level. An example is the category class, where there can be categories of categories, overlappping categories, and so on. They’re all categories, but there is internal organization at that level. The same thing happens with programs, obviously: programs, sub-programs, and so on. All programs, but structured within that concept. Even configurations seem to have this sort of internal structure. All this, of course, purports to be in the realm of observation, not modeling. Your musings on reorganization might be very relevant to the development of fine structure within major types of perceptions.

A still weaker notion of a level-based structure (which probably no

longer deserves the name “hierarchy”) is to count, for any

perception, the minimum number of processing stages that must

separate the sensory input from that perception.

I did this very roughly in writing B:CP. Ranson and Clark’s neuroanatomy book (1947 edition as I recall!) is organized in some places in terms of “Nerve 1, nerve 2, nerve 3 …” for each modality, with synapses doing the separating. If we consider a “synapse” in this chain to represent a sensory nucleus (which generally it does), we come out pretty much like the first four levels in the model, not entirely by accident.

In such a structure,

other processing pathways with more stages may be involved in a

particular perception, so one could ask the question in the other

way–what is the maximum number of processing stages in any necessary

pathway? In a strict hierarchy, those two questions give the same

answer. Otherwise they don’t.

There is pretty good agreement through the first four levels in all modalities, although the anatomical locations get pretty scattered. Counting synapses aligns the levels pretty well, I think, if you ignore collaterals.

So David’s question is a real one: "How does one determine that two

perceptions are at different levels."

I caution that identifying levels in a model is very different from “going up levels” in the Method of Levels. The last thing we think about in the MOL is whether the 11 levels are properly identified – the model-levels simply aren’t well-enough established to rely on in therapy. We’re interested only in relative level, and even then only in the sense that the viewpoint one was operating from becomes the object of observation.

I think also that one must ask

"How can one investigate whether the hierarchy is strict, in whole or

in part?" It could be, for example, that there is a strict hierarchy

below the barrier (assuming there is one) between analogue levels and

logical levels, but that the barrier consists of self-facilitating

loops of perceptions at the same level. Such a structure would be a

mixture of a strict and a weak hierarchy.

I agree with all this. Our descendants will have to work it out.

Best,

Bill P.

4

[From Rick Marken (2003.11.29.0940)]

Martin Taylor (2003.11.29.0912) --

So David's question is a real one: "How does one determine that two
perceptions are at different levels."

Relative timing is the main approach to determining whether perceptions
are controlled at different levels. See the example of a reaction time
experiment described on pp. 245 and 246 of B:CP. Another example of the
reaction time approach is shown in my "Levels of control" demo at
Levels of Control. A somewhat
different approach is described in the "Hierarchical behavior of
perception" paper in _More Mind Readings_. In these experiments,
control is measured as a function of the speed of presentation of the
environmental events of which controlled perceptual variables are a
function. A computer implementation of these experiments can be found
at http://www.mindreadings.com/ControlDemo/HP.html but, unfortunately,
it is not machine independent because, contrary to the Java manual, the
time commands are not in milliseconds.

One final way to study the hierarchy is through the use of models. We
are implicitly testing the hierarchy by finding the simplest working
model (fewest systems, levels and parameters) that can account for
observed data. Bill's three level "little man" model of pointing is, to
the extent that it predicts actual pointing trajectories, evidence that
three levels of control are involved in pointing at a moving target.

I think also that one must ask
"How can one investigate whether the hierarchy is strict, in whole or
in part?"

The answer to that question will presumably fall out after many years
of investigation of the hierarchy.

I don't know how one would gather evidence to define the structure of
the putative hierarchy.

I've described three approaches above: reaction time, "perceptual"
timing and modeling.

Bill's approach is good for setting up a
hypothesis that is presumably testable, but as he has repeatedly
stated, that's all it has ever been intended to be. The question is
how to go about actually testing it.

You just seem to have missed the parts of B:CP where Bill explains
this. As the resident lapdog I think it's my responsibility to note
that Bill is not just a brilliant theorist. He's also a fine
researcher. It was as much Bill's skills as a researcher as his
ingeniousness as a theorist that attracted me to PCT. The PCT model is
eminently testable, and some of us have actually been testing it.

Best

Rick

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

5

[Martin Taylor 2003.11.29.1409]

[From Rick Marken (2003.11.29.0940)]

Martin Taylor (2003.11.29.0912) --

So David's question is a real one: "How does one determine that two
perceptions are at different levels."

Relative timing is the main approach to determining whether perceptions
are controlled at different levels.

Good, but different sense modalities have different timings, as does
any one modality when discriminations are difficult (e.g. vision in
low light). I remember many years ago playing table-tennis by the
light of the full moon. It took me quite a few shots to learn that I
had to hit the ball at the moment when perceptually it was only just
arriving at the net.

A naive interpretation of this would be that the level at which the
hitting event was controlled would be at a lower level than the
perception of the ball's position and velocity. Subjectively, it was
a conscious effort to use relatively long-range prediction of what
the perceptions would become (as compared to the predictive time
scale necessary in the "ballistic" part of the stroke under normal
lighting conditions).

There's another problem with relative timing, which is that it does
not handle the multiple-path question very well.

One final way to study the hierarchy is through the use of models. We
are implicitly testing the hierarchy by finding the simplest working
model (fewest systems, levels and parameters) that can account for
observed data. Bill's three level "little man" model of pointing is, to
the extent that it predicts actual pointing trajectories, evidence that
three levels of control are involved in pointing at a moving target.

Yes. That's what I've been trying to do with my tracking complexity
study, but I don't think my data are clean enough or have enough
degrees of freedom to do it.

However, one also has to recognize that Ptolemy's epicycles described
the movements of the planets extremely accurately, and it was only
the emergence of a simpler counter-theory that was equally accurate
that led astronomers to abandon them. Suppose someone devises a
two-level "little man" that equally well describes the trajectory?
Would that constitute proof that humans do NOT use a three-level
control system? What Bill's "Little Man" shows is that a three-level
control system is adequate for the job, not that humans don't use a
seven-level control system or an entirely different approach.

Bill's approach is good for setting up a
hypothesis that is presumably testable, but as he has repeatedly
stated, that's all it has ever been intended to be. The question is
how to go about actually testing it.

You just seem to have missed the parts of B:CP where Bill explains
this.

Missed or forgotten. I lent my copy of B:CP to someone a long time
ago, and I haven't seen it since. At one time I may have remembered
Bill's explanation, but I don't now. (And I haven't bought a
replacement copy, sorry, Bill).

As the resident lapdog I think it's my responsibility to note
that Bill is not just a brilliant theorist. He's also a fine
researcher. It was as much Bill's skills as a researcher as his
ingeniousness as a theorist that attracted me to PCT. The PCT model is
eminently testable, and some of us have actually been testing it.

Yes.

Martin

6

[From David Goldstein(2003.11.29.1802 EST)]
[Rick Marken (2003.11.29.0940)]

Hi Rick,
I wonder if your suggestion of relative timing could be used in the
clinical situation to decide if topic A and a topic B are at different
relative levels.
I don't think so but maybe.
David
David M. Goldstein, Ph.D.

···

-----Original Message-----
From: Control Systems Group Network (CSGnet)
[mailto:CSGNET@listserv.uiuc.edu] On Behalf Of Rick Marken
Sent: Saturday, November 29, 2003 12:38 PM
To: CSGNET@listserv.uiuc.edu
Subject: Re: How can perceptual hierarchy be tested?

[From Rick Marken (2003.11.29.0940)]

Martin Taylor (2003.11.29.0912) --

So David's question is a real one: "How does one determine that two
perceptions are at different levels."

Relative timing is the main approach to determining whether perceptions
are controlled at different levels. See the example of a reaction time
experiment described on pp. 245 and 246 of B:CP. Another example of the
reaction time approach is shown in my "Levels of control" demo at
Levels of Control. A somewhat
different approach is described in the "Hierarchical behavior of
perception" paper in _More Mind Readings_. In these experiments, control
is measured as a function of the speed of presentation of the
environmental events of which controlled perceptual variables are a
function. A computer implementation of these experiments can be found
at http://www.mindreadings.com/ControlDemo/HP.html but, unfortunately,
it is not machine independent because, contrary to the Java manual, the
time commands are not in milliseconds.

One final way to study the hierarchy is through the use of models. We
are implicitly testing the hierarchy by finding the simplest working
model (fewest systems, levels and parameters) that can account for
observed data. Bill's three level "little man" model of pointing is, to
the extent that it predicts actual pointing trajectories, evidence that
three levels of control are involved in pointing at a moving target.

I think also that one must ask
"How can one investigate whether the hierarchy is strict, in whole or
in part?"

The answer to that question will presumably fall out after many years of
investigation of the hierarchy.

I don't know how one would gather evidence to define the structure of
the putative hierarchy.

I've described three approaches above: reaction time, "perceptual"
timing and modeling.

Bill's approach is good for setting up a
hypothesis that is presumably testable, but as he has repeatedly
stated, that's all it has ever been intended to be. The question is
how to go about actually testing it.

You just seem to have missed the parts of B:CP where Bill explains this.
As the resident lapdog I think it's my responsibility to note that Bill
is not just a brilliant theorist. He's also a fine researcher. It was as
much Bill's skills as a researcher as his ingeniousness as a theorist
that attracted me to PCT. The PCT model is eminently testable, and some
of us have actually been testing it.

Best

Rick

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

7

[From Rick Marken (2003.11.29.1630)]

David Goldstein(2003.11.29.1802 EST)--

Hi Rick,
I wonder if your suggestion of relative timing could be used in the
clinical situation to decide if topic A and a topic B are at different
relative levels.

My intuition is that the MOL approach to getting a higher level topics
is the way to go in clinical settings. The relative timing thing works
best in controlled experimental situations where confounding variables,
like the ones mentioned by Martin Taylor (2003.11.29.1409), can be
carefully factored out.

Best regards

Rick

···

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