Modelling PCT: (was Re: The Great Maturana: a PCT Heretic?)

[Martin Taylor 2016.09.26.12.45]

[From Rick Marken (2016.09.25.1750)]

Rupert Young (2016.09.24 1700)

The problem with that is that it is hard to prove a negative by just

modelling. You may fail to show that PCT can explain the situation
simply because you are modelling control of the wrong variable, or
have not found suitable ranges of parameter values. Failure doesn’t
mean PCT can’t explain, even though success shows it can. In most
sciences, modelling and theory go hand in hand. PCT science should
BE a science. Theory CAN show why certain sets of assumptions cannot
explain some things, and need to be expanded or amended.

A new definition of "perfectly"? And of "every" at the same time?

Huh? What version of CSGnet do you get in your alternate Universe?

The challenge has never been to PCT as a theory in its present form.
The challenge is to find the correct controlled variable(s) and
parameter values for a working model that produces the power law
appropriate to the situations under which it is observed. Nobody has
done that yet – or if they have, they haven’t posted it. That
doesn’t mean PCT needs to be changed.

Look up the definition of "chutzpa" in the Random House dictionary.

I think this paragraph is a good example, given the multitude of
ways that it has been shown that the model is not even a model of
the situation, and that the maths behind the analysis is wrong
anyway. You may have had the right to be astonished the first time
Alex showed you how it was wrong, but not over the subsequent months
of being shown in ever varying ways and levels of detail how wrong
it was.

Look, modelling is great. When used effectively with theory, it's

even better. But a model that is supposed to explain an observation
ought to be at least a little like the situation in which the
observation is made. Rick’s “model” that “accounts for the
power-law” phenomenon isn’t.

Rick's model is of a 2D tracking task, like chasing a butterfly over

an open field. How fast the butterfly goes determines how fast and
where you go (assuming you aren’t intending to end the chase by
catching it). If Rick’s “butterfly” followed the power law relation
between speed and path curvature, and his tracker was good at its
job, its tracker object (a.k.a. “cursor”) also had to follow the
same power law, didn’t it? I don’t think anyone used those words
yet, but wouldn’t it be perfectly fair to call that a “put-up job”?

The problem at hand is more like "Here's a road. How fast will you

drive around the different curves you see on the map (or that you
see before you)?" Or, “Here’s a drawing. How fast will you go over
its parts when you trace it accurately with a pencil?” That’s not a
bit like chasing a butterfly to see where and how fast it goes.

Any model of the actual task has to take as input an entire curving

line, not some point that moves along the curve. The curve could be
a closed figure such as an ellipse. The line could be input as
disturbance or as reference. The model has to control perceptions so
that something moves, and so that the moving object stays close to
the input curving line when it does move. That’s a minimum
requirement for a working model of the situation.

And who is most recently responsible for that? Might it possibly be

someone who consistently refuses every attempt over a period of
months to counter scientific criticisms of a pet theory?

You could help bring CSGnet back to "scientificity" by creating a

model that meets the minimum requirements for a model that might
explain the power law. If it’s a good model, it will generate 1/3,
1/4, and no power law at all by changing the value of one parameter,
or maybe two. You are a good modeller, and to solve this problem
would be a feather in your cap.

Martin
···
                      AGM:

Indeed. I am interested in the implications of
the PCT asymmetry where P could stand for
Procrustean, namely, do we shrink the genius
of Maturana or are we willing to be heretic
enough to conceive an extension of PCT…? :wink:

                RY: Like any theory PCT is open to revision or

extension. So, if you think something is missing by
all means make a suggestion.

                RY: In this instance could you explain: What it is

you are proposing? What do you think is missing from
PCT? How does the proposal fill the hole? What will
the revised theory explain that it didn’t
previously?

        RM: Thank you Rupert. This is what has been missing from

discussions of PCT on CSGNet: an explanation of what any
proposed revision or extension of PCT would explain that PCT
does not currently explain. In order to do this you have to
had observed some phenomenon and then show, using modeling,
that PCT in its current form doesn’t explain it.

        RM: It's important to understand,

however, that in PCT “modeling” is not just curve fitting;
it’s getting a working instantiation of the model to behave
in the same way as does an organism placed in the same
circumstances. Working PCT models are usually instantiated
as computer programs but now, even better, as robotic
systems.

        RM: The PCT approach to modeling is

described in Chapter 2 of B:CP (both editions). It’s an
approach to modeling – and testing models against the
results of experimental test – that drew me to PCT. What
attracted me wasn’t just that PCT seemed like a nice theory
of behavior; it was that every prediction of the theory
could be tested quantitatively against observation; and
every prediction held up perfectly.

        ...

        RM: The modeling approach to

evaluating PCT has been conspicuously missing from
discussions on CSGNet. The power law discussion is the most
recent example. The power law is an observed phenomenon
whose explanation supposedly presents a challenge to the PCT
model in its present form.

        But this claim was made without

determining whether a working PCT model, behaving in the
same circumstances as those in which the power law is
observed, could produce the power law. It was simply
asserted that it couldn’t. So when I produced a PCT model of
one example of behavior that results in the power law and
found that the model accounts for the power law, I was
astonished to find that it was dismissed as a “put up job”
or something like that.

        RM: So it seems that CSGNet is no

longer (if it ever was) a forum for dealing with PCT
scientifically;

[From Rick Marken (2016.09.26.1600)]

···

Martin Taylor (2016.09.26.12.45)

MT: The problem with that is that it is hard to prove a negative by just

modelling.

RM: Proving is a mathematical, not a scientific concept. What I am proposing is that we try to account for behavioral phenomena using the current version of the PCT model before proposing changes to the model. And don’t worry, Martin. I don’t expect you to participate in this venture; I was just saying what I would like to see because that is what I have always been doing with PCT.

MT: Huh? What version of CSGnet do you get in your alternate Universe?

The challenge has never been to PCT as a theory in its present form.

RM: We were challenged to provide the PCT explanation of the power law.

MT: The challenge is to find the correct controlled variable(s) and

parameter values for a working model that produces the power law
appropriate to the situations under which it is observed.

RM: Right, the challenge was to provide the PCT explanation of the power law, which would, of course, involve including in the model a controlled variable and parameter values that produce movement trajectories that corresponded to the power law.

MT: Nobody has

done that yet

RM: Actually, two people did it; Kent McClelland and me.

MT: – or if they have, they haven’t posted it.

RM: I did post it, as did Kent. My model was rejected; there was some interest in Kent’s model until I noted that it was the same as mine, then… crickets.

MT: That

doesn’t mean PCT needs to be changed.

RM:That’s true. But since you (and everyone else) rejected my PCT model I assumed that you thought some changes to the PCT model were needed.

MT: Look up the definition of “chutzpa” in the Random House dictionary.

RM: Hey, you mishuga, my people invented the word.

MT: I think this paragraph is a good example, given the multitude of

ways that it has been shown that the model is not even a model of
the situation, and that the maths behind the analysis is wrong
anyway.

RM: The model is a perfectly good model of one situation in which the power law is found – the one where people move their finger around in an elliptical pattern. And the model produces data just like the data produced by people. And the analysis of the data results in a power law. And the math analysis is exactly right, as demonstrated by the multivariate regression analysis.

MT: You may have had the right to be astonished the first time

Alex showed you how it was wrong, but not over the subsequent months
of being shown in ever varying ways and levels of detail how wrong
it was.

RM: And yet, it power laws;-)

MT: Look, modelling is great. When used effectively with theory, it's

even better.

RM: Modeling is used effectively with data, Martin, not theory. Models are theory.

MT: But a model that is supposed to explain an observation

ought to be at least a little like the situation in which the
observation is made. Rick’s “model” that “accounts for the
power-law” phenomenon isn’t.

RM: Please explain how the model is not like the situation in which the power law is observed, Here’s a diagram of the model again so you can point out exactly how it goes awry:

MT: Rick's model is of a 2D tracking task, like chasing a butterfly over

an open field.

RM: No tracking involved in this model. But it wouldn’t matter anyway if it were.

MT: How fast the butterfly goes determines how fast and

where you go (assuming you aren’t intending to end the chase by
catching it). If Rick’s “butterfly” followed the power law relation
between speed and path curvature, and his tracker was good at its
job, its tracker object (a.k.a. “cursor”) also had to follow the
same power law, didn’t it? I don’t think anyone used those words
yet, but wouldn’t it be perfectly fair to call that a “put-up job”?

RM: So you’re saying that the power law “doesn’t count” when people track patterns that result in the power law? What if it also doesn’t produce the power law with patterns that don’t produce the power law? Isn’t that evidence that the model is behaving like the person? But let’s just go with the non-tracking version of the model shown above; if some patterns of reference movement produce a power law in both people and model and other patterns don’t, isn’t this evidence that the model accounts for the power law. Actually, there’s a nice PNAS paper by Huh and Sejnowski entitled “Spectrum of power laws for curved hand movements” that shows that you get different power laws for different curved hand movements. My model produces this result also and my “incorrect” math accounts for why this is found.

MT: The problem at hand is more like "Here's a road. How fast will you

drive around the different curves you see on the map (or that you
see before you)?"

RM: No, it’s not like that at all. That’s a very different situation than the one where a person moves their finger in different patterns in the air (or water).

MT: Any model of the actual task has to take as input an entire curving

line, not some point that moves along the curve. The curve could be
a closed figure such as an ellipse. The line could be input as
disturbance or as reference. The model has to control perceptions so
that something moves, and so that the moving object stays close to
the input curving line when it does move. That’s a minimum
requirement for a working model of the situation.

RM: Since you know this you should be able to build the model and show us how it works. I’m catching a cold from the wind created by your waving hands (and it’s 100 F outside).

MT: And who is most recently responsible for that? Might it possibly be

someone who consistently refuses every attempt over a period of
months to counter scientific criticisms of a pet theory?

RM: We just have very different ideas about what science is.

MT: You could help bring CSGnet back to "scientificity" by creating a

model that meets the minimum requirements for a model that might
explain the power law.

RM: I already did. And I set up a way to collect data to test the model. But you don’t really care for data, do you?

MT: If it's a good model, it will generate 1/3,

1/4, and no power law at all by changing the value of one parameter,
or maybe two. You are a good modeler, and to solve this problem
would be a feather in your cap.

RM: The current model does precisely that by changing just one parameter – variations in r.

Best

Rick

Martin


Richard S. Marken

“The childhood of the human race is far from over. We
have a long way to go before most people will understand that what they do for
others is just as important to their well-being as what they do for
themselves.” – William T. Powers

        RM: Thank you Rupert. This is what has been missing from

discussions of PCT on CSGNet: an explanation of what any
proposed revision or extension of PCT would explain that PCT
does not currently explain. In order to do this you have to
had observed some phenomenon and then show, using modeling,
that PCT in its current form doesn’t explain it.

        RM: The modeling approach to

evaluating PCT has been conspicuously missing from
discussions on CSGNet. The power law discussion is the most
recent example. The power law is an observed phenomenon
whose explanation supposedly presents a challenge to the PCT
model in its present form.

        RM: But this claim was made without

determining whether a working PCT model, behaving in the
same circumstances as those in which the power law is
observed, could produce the power law. It was simply
asserted that it couldn’t. So when I produced a PCT model of
one example of behavior that results in the power law and
found that the model accounts for the power law, I was
astonished to find that it was dismissed as a “put up job”
or something like that.

        RM: So it seems that CSGNet is no

longer (if it ever was) a forum for dealing with PCT
scientifically;

[Martin Taylor 2016.09.28.11.25]

[From Rick Marken (2016.09.26.1600)]

You still don't understand either why your model is said to be wrong

or why the explanation of how you misuse mathematics is not “like a
Trump lie” [From Rick Marken (2016.09.26.1255)]. At the end of this
message, I offer you a new way to explain why I am mistaken about
the model (I don’t think “lie” is a word that suits a scientific
dispute).

I think there is only one person on CSGnet who does not know why

your model was rejected. I am under the impression that Kent knows
why his motion-following 2D tracker was not a model of the situation
about which the question was asked. But I remain,as always, ready to
be corrected, along with the rest of the readership.

To the so-called “PCT model”, yes. To PCT, no.

Well, if there isn't, I have misunderstood your model for the last

several months. I thought the pencil tracked the movement of a
target point along the squiggle. I guess you have to explain it more
carefully than you did originally or subsequently.

The main points to clarify are how the long vectors (sampled

waveforms) r(t), p(t) and e(t) are represented, and where in the
model there is a reference to time (the “t” variable for the
waveform samples). As I understood it all these months, time was
represented by a point that moved along the reference squiggle, with
its x-y position sampled at defined moments. The vectors were
represented by the successive x-y sample values of the three
variables. I guess I was wrong, but that leaves the question wide
open. Where is the time (or equivalently velocity) specified, and
how are the vectors represented if they are not represented by the
values of x and y for the three variables at successive sample
moments?

Martin

···

Martin Taylor (2016.09.26.12.45)

            MT: The problem with that is that it is hard to prove a

negative by just modelling.

          RM: Proving is a mathematical, not a scientific

concept. What I am proposing is that we try to account for
behavioral phenomena using the current version of the PCT
model before proposing changes to the model. And don’t
worry, Martin. I don’t expect you to participate in this
venture; I was just saying what I would like to see
because that is what I have always been doing with PCT.

            MT: Huh? What version of CSGnet do you get in your

alternate Universe? The challenge has never been to PCT
as a theory in its present form.

          RM: We were challenged to provide the PCT explanation

of the power law.

            MT: The challenge is to find the

correct controlled variable(s) and parameter values for
a working model that produces the power law appropriate
to the situations under which it is observed.

          RM: Right, the challenge was to provide the PCT

explanation of the power law, which would, of course,
involve including in the model a controlled variable and
parameter values that produce movement trajectories that
corresponded to the power law.

MT: Nobody has done that yet

          RM: Actually, two people did it; Kent McClelland and

me.

                    RM: Thank you Rupert. This is what has been

missing from discussions of PCT on CSGNet: an
explanation of what any proposed revision or
extension of PCT would explain that PCT does not
currently explain. In order to do this you have
to had observed some phenomenon and then show,
using modeling, that PCT in its current form
doesn’t explain it.

                    RM: The modeling approach

to evaluating PCT has been conspicuously missing
from discussions on CSGNet. The power law
discussion is the most recent example. The power
law is an observed phenomenon whose explanation
supposedly presents a challenge to the PCT model
in its present form.

            MT: -- or if they have, they

haven’t posted it.

          RM: I did post it, as did Kent. My model was rejected;

there was some interest in Kent’s model until I noted that
it was the same as mine, then… crickets.

            MT: That doesn't mean PCT needs to

be changed.

          RM:That's true. But since you (and everyone else)

rejected my PCT model I assumed that you thought some
changes to the PCT model were needed.

          RM: The model is a perfectly good model of one

situation in which the power law is found – the one where
people move their finger around in an elliptical pattern.
…;

          RM: Please explain how the model is not like the

situation in which the power law is observed, Here’s a
diagram of the model again so you can point out exactly
how it goes awry:

            MT: Rick's model is of a 2D

tracking task, like chasing a butterfly over an open
field.

          RM: No tracking involved in this model. But it wouldn't

matter anyway if it were.