trying to make a connection

CSGnet Archive CSG1999
1

Bruce Nevin wrote:

Maybe you're bowing out because of other commitments of time and energy. I
can understand that, and just turn me down if that's the case. I'm trying
to understand your position. The best way to demonstrate (verbal)
understanding is paraphrase. This is my attempt to paraphrase what I think
you are saying.

To: "Control Systems Group Network (CSGnet)"
<CSGNET@POSTOFFICE.CSO.UIUC.EDU>,
        CSGNET@POSTOFFICE.CSO.UIUC.EDU
Subject: Re: signals and experience
Cc: bn@cisco.com

[From Bruce Nevin (990803.1148 EDT)]

i.kurtzer (990802.2035)--

>where is the evidence for heirarchical control? Just saying, "well,
>we know it has to be" is not evidence. [...] The rest of
[Bill Powers (990802.0035 MDT)]
>post is a nice proposal, but if that counts as evidence then PCT is not a
>scientific theory.

i.kurtzer (990803.1000)--

>I agree that behavioral theories are
>analyzable with behavioral data. But this "one control system uses another
>as a means of control" is a proposal.

Me (990803.0957 EDT)--

>Could you give a
>specific example of a science and something about how researchers work in
>that field, illustrating what you believe is missing from HPCT?

i.kurtzer (990803.0957)--

>How about synaptic trasmission. [...]
>Does HPCT have any line of research? Maybe the Plooij's. Thats it.

I think you are saying that what counts for you as a line of research is
specifically neurophysiological research showing that control systems are
implemented in nervous systems, and that within nervous systems one control
system uses another as a means of control. Am I right?

Naturalistic, observational research into the phenomenon of control does
not suffice for you--that's mostly what we talk about here. For example,
for you, doing one of Gary Cziko's or Bill's demos of the phenomenon of
hierarchical control does not count as a line of research in science.

Demos are not lines of research, but starting points. Great starting points,
but still starting points.

It is possible that even constructing and testing a model whose measured
outputs replicate the measured outputs of a living control system (under
the same limiting conditions) do not count for you as a line of HPCT
research, because there is no evidence that the structures found to be
working in the model are in fact working in the same way in the living
organism. Is this your view also? [There's been little enough of this done.]

Is this getting close to what you mean?

i think that behavioral theories can stand on their own terms, with behavioral
data. I gave a physiological example since any behavioral example other than
PCT would be contentious.

i.kurtzer (990803.1200)

>i bow out of the conversation.

I hope you will reconsider. I know you have ideas about the nature of
science and the status of HPCT that are important to you. I would like to
understand them better.

We will continue our relationship. I don't have the energy to deal with someone
i respect squirming around and bungling things up. I think its a major issue to
stick with what is the strongest evidence and not confuse that with someone's
pet idea, especially outdated and unsubstatiated pet ideas.

i.

2

Maybe you're bowing out because of other commitments of time and energy. I
can understand that, and just turn me down if that's the case. I'm trying
to understand your position. The best way to demonstrate (verbal)
understanding is paraphrase. This is my attempt to paraphrase what I think
you are saying.

···

To: "Control Systems Group Network (CSGnet)"
<CSGNET@POSTOFFICE.CSO.UIUC.EDU>,
        CSGNET@POSTOFFICE.CSO.UIUC.EDU
Subject: Re: signals and experience
Cc: bn@cisco.com

[From Bruce Nevin (990803.1148 EDT)]

i.kurtzer (990802.2035)--

where is the evidence for heirarchical control? Just saying, "well,
we know it has to be" is not evidence. [...] The rest of

[Bill Powers (990802.0035 MDT)]

post is a nice proposal, but if that counts as evidence then PCT is not a
scientific theory.

i.kurtzer (990803.1000)--

I agree that behavioral theories are
analyzable with behavioral data. But this "one control system uses another
as a means of control" is a proposal.

Me (990803.0957 EDT)--

Could you give a
specific example of a science and something about how researchers work in
that field, illustrating what you believe is missing from HPCT?

i.kurtzer (990803.0957)--

How about synaptic trasmission. [...]
Does HPCT have any line of research? Maybe the Plooij's. Thats it.

I think you are saying that what counts for you as a line of research is
specifically neurophysiological research showing that control systems are
implemented in nervous systems, and that within nervous systems one control
system uses another as a means of control. Am I right?

Naturalistic, observational research into the phenomenon of control does
not suffice for you--that's mostly what we talk about here. For example,
for you, doing one of Gary Cziko's or Bill's demos of the phenomenon of
hierarchical control does not count as a line of research in science.

It is possible that even constructing and testing a model whose measured
outputs replicate the measured outputs of a living control system (under
the same limiting conditions) do not count for you as a line of HPCT
research, because there is no evidence that the structures found to be
working in the model are in fact working in the same way in the living
organism. Is this your view also? [There's been little enough of this done.]

Is this getting close to what you mean?

i.kurtzer (990803.1200)

i bow out of the conversation.

I hope you will reconsider. I know you have ideas about the nature of
science and the status of HPCT that are important to you. I would like to
understand them better.

  Bruce Nevin

3

i.kurtzer

Bruce Nevin wrote:

>i think that behavioral theories can stand on their own terms, with
>behavioral data.

Yes, you said

> i.kurtzer (990803.1000)--
>
> >I agree that behavioral theories are
> >analyzable with behavioral data. But this "one control system uses another
> >as a means of control" is a proposal.

Why is the assertion "one control system uses another as a means of
control" not a valid part of a valid behavioral theory? That is, a theory
that accounts for behavior such as that seen in the arm drop demo? Put
another way, any theory is a proposal--the flat earth idea is a theory that
happens to be unaccepted because it is controverted by evidence, a
disproven theory.

Oh yeah, I think its a great specific proposal of a general theory. But it has
not been put to any tests. Saying it looks like someone else's data does not
remove it from the game but that cannnot, cannot, cannot be convincing. "Looks
like" must be given a rigorous treatment. He hasn't and he has stopped trying to
submit it, that was once to Science!!, to where others could give it the criticism
it deserves, like all hypotheses. He has turned down offers for lead articles in
Behvioral and Brain Research, which is very widely read, and he is effectively
hocking this to people that don't have the math to criticize. Other people have
used him as part of their ouput function and have meanwhile neglected to develop
their own input functions, thats includes rick. and although that is consistent
with bill's leftist postion of inclusion i think it is wreaking far more harm than
good. greg williams, a mit trained engineer, who used to go through bill's
equations to check them--who does that now?--left because of bill making broad
strokes, not differentiating between his pet ideas and the core theory and
obviously because of rick being a major asshole. tom left principally because of
rick. i can think of other quality people. and what we have left is kent
mcclelland who models imaginary people--not tied to any data--cziko hocking books
and the iaact quacks..

Are you challenging the "testing specimens" metatheory that if a model
behaves faithfully and reliably the way the organism behaves, then the
organization of the model gives us valid insight into the organization of
the organism? This can only assert that there is a control loop, or a
hierarchy of control loops (depending on the specific model), it can't say
thing about how those control loops are implemented. If you do accept this
as valid science, then HPCT has some good evidence as a behavioral theory.

HPCT is a specical hypothesis that sounds neat, maybe "on the right track", but
that needs to be tested. The tests are not there demos are not cumulative tests
but demonstrations of principle.

Isn't it the proposal that control systems are implemented in nervous
systems in the ways that are sketched in B:CP that you are objecting to?

No. Of course that has about zip data too. I am objecting to smearing the line
between confidently established conjectures and unestablished conjectures. If
anyone loses sight of that than they have stopped doing science.

This started with your objection to identifying "perception" with "neural
signal". Except for a few cases like the stretch reflex there is no
confirmation of Bill's conjectures about neural signals, comparators
implemented in neurons, and so on. Is that the focus of your objections?

I complained on that because i would like them to tell me why one signal is
magically percieved in the experiential sense once it is in a relation to another
signal. Why are reference signals perceived or if they are why differently, or
error signals or the signal than just sits there. these are not immediate
question to a working scientist and would never expect them to be embroiled in
it. But to substitue a working hypothesis for "it has to be" becuase bill said it
in pg 3 chapt 5 is so upsetting. and that bizarre rebuttal came from both rick
and bill. Then comes the rate coding hypothesis which is a specific hypothesis
tangential to PCT and so out of whack from what computtational neurosceince has
been doing for 20 years. but that doesn't matter since they not engineers and we
know they all do _ _. but the engineers disagree with him and he now hates
modern contro, theory and who can distinguish whether their claims are accurate?
rick? me? you? no becuase we don't have the background..we have to take it from
bill. what happens when he dies..who will we take it from..i don't think we
should all become engineers but something has developed when one of the world's
experts in PCT has to look right to bill when he does his own modelling..

two asides that just take the cake. first, remeber the "univeersal error curve"
that bill invented to explain coersion..that is such an oxymoorn and anithetical
to everything i thought bill stood for that i will leave it.
second, thsi economic control system of ricks that he presented at the
conference..isn't that the same as the social control systems that we have
criticized for years..

When you say there is no evidence for HPCT, do you mean that there is
unsufficient evidence for a hierarchy of control loops in the nervous
system? That there is no evidence for HPCT as a neurophysiological theory,
but you accept that there is good evidence for it as a behavioral theory?

No there is almost no evidence as a behavioral theory either.

Bill is defending the validity of HPCT as a behavioral theory. He thinks
that's what you're challenging.

I am as a substantiated claim, which its not. I think its a great proposal and
have some good demos of principle but that is not a test of a specific
hypothesis.

The original objection to perception=signal
has been forgotten.

Thats tangential, but he thinks he has invented that idea too.As far as the firing
rate hypothesi that is again a specific proposal, not of PCT, but of a coding
scheme. Its that type of loosness that will make him look like a fool to anyone
who has loked at the material. Of course, all that we know is ridiculous since it
wasn't done with the Test. The loop closes and bill is left as the only one with
the answer.
That my problem.
i.

···

At 12:42 AM 08/04/1999 -0400, you wrote:

4

i think that behavioral theories can stand on their own terms, with
behavioral data.

Yes, you said

i.kurtzer (990803.1000)--

>I agree that behavioral theories are
>analyzable with behavioral data. But this "one control system uses another
>as a means of control" is a proposal.

Why is the assertion "one control system uses another as a means of
control" not a valid part of a valid behavioral theory? That is, a theory
that accounts for behavior such as that seen in the arm drop demo? Put
another way, any theory is a proposal--the flat earth idea is a theory that
happens to be unaccepted because it is controverted by evidence, a
disproven theory.

Are you challenging the "testing specimens" metatheory that if a model
behaves faithfully and reliably the way the organism behaves, then the
organization of the model gives us valid insight into the organization of
the organism? This can only assert that there is a control loop, or a
hierarchy of control loops (depending on the specific model), it can't say
thing about how those control loops are implemented. If you do accept this
as valid science, then HPCT has some good evidence as a behavioral theory.

Isn't it the proposal that control systems are implemented in nervous
systems in the ways that are sketched in B:CP that you are objecting to?

This started with your objection to identifying "perception" with "neural
signal". Except for a few cases like the stretch reflex there is no
confirmation of Bill's conjectures about neural signals, comparators
implemented in neurons, and so on. Is that the focus of your objections?
When you say there is no evidence for HPCT, do you mean that there is
unsufficient evidence for a hierarchy of control loops in the nervous
system? That there is no evidence for HPCT as a neurophysiological theory,
but you accept that there is good evidence for it as a behavioral theory?

Bill is defending the validity of HPCT as a behavioral theory. He thinks
that's what you're challenging. The original objection to perception=signal
has been forgotten.

Perception=signal is not necessarily an assertion about HPCT as a
neurophysiological theory -- though it is usually taken as such, taking
Bill's conjectures as a working hypothesis that is weakly supported by
limited data. ("Modelling tells us there must be a loop structure there,
and no other methodology even comes close; neural transmission data fit
pretty well with timing of behavioral data, so far as that has been looked
at. No other reasonable contenders in sight, looks like a pretty good
working hypothesis.") So are you saying "we can't assert this as true, but
only as a somewhat plausible working hypothesis"? A working hypothesis is
useful for organizing research. Much research into the neurophysiology of
control is needed. You aim to contribute to this research.

If you accept this, then I think that you and Bill are actually in violent
agreement. You got at crossed purposes or unintended equivocation between
HPCT the behavioral theory where perceptions are signals (however
implemented) and HPCT the neurophysiological conjecture where perceptions
are signals as neural firing rates.

Unrecognized ambiguity is the chief source of misunderstanding and
argument. That's why we define technical terms and use them carefully as
defined. HPCT has these two meanings, grounded in two sorts of evidence,
behavioral and neurophysiological. Modelling can be the bridge to unify
them eventually, but we need more data of the latter sort.

  Bruce

···

At 12:42 AM 08/04/1999 -0400, you wrote:

5

Bruce Nevin wrote:

>I think its a great specific proposal of a general theory.
>But it has not been put to any tests.

So General Relativity and Quantum Mechanics were only proposals, not
theories, until after they were tested? Retroactively accounting for
observations "explained" by prior theories doesn't count for anything?

I don't have a problem with that.

Do you get your use of terms like theory vs. proposal from some philosophy
of science standard that I could look up?

Are you being serious? The situation is far from comparable. The scientists in
that field would have loved to put their predictions to a test, but couldn't.
Here its been TOLD it matches..have you seen how it matches, has anyone but
bill..

>Saying it looks like someone else's data does not
>remove it from the game but that cannnot, cannot, cannot be convincing.

I guess you mean e.g. the arm model matching the MIT measurements? Is there
something wrong with it being someone else's data? I should think that
would enhance the case for it being replicable and "objective." Or are you
referring to something else? (Your jab jab jab style leaves a lot to be
filled in by imagination.)

If someone is touting a proposal, drawn from a general theory, to an audience
that wouldn't know the difference than its not science.

What constructive suggestions have you? What would constitute a valid test?
What kind of work is needed for HPCT to graduate from "proposal" to "theory"?

There is not a valid test per se. But we need a cumulative research program with
HPCT as its heursitic. Demos won't subsitute for that.

Being convincing is another matter, going outside of science to the social
and political dynamics of networks (not communities!) of scientists and
others. I won't get into that here, in particular not the social dynamics
of CSG, and I'm not asking you to try to find a solution for that, which
obviously aggrieves you (and me). Just the question of test and validation.

        Bruce

Validation is within a series a questions. Not a demonstration. The Plooij's
chimp and infant studies were a good start.. Almost any subject will do. I am
working on motor control. I am applying it to my work as best I can. I wouldn't
expect anything more from anyone. Just work on out own stuff. But that does not
mean that anyone else has done the job. They haven't. Ask bill about all the
empirical paers on HPCT. His cognitive control model!! that shows that we can
write 2 with a big or small pen!!Jesus what type of test is that.

6

I think its a great specific proposal of a general theory.
But it has not been put to any tests.

So General Relativity and Quantum Mechanics were only proposals, not
theories, until after they were tested? Retroactively accounting for
observations "explained" by prior theories doesn't count for anything?

Do you get your use of terms like theory vs. proposal from some philosophy
of science standard that I could look up?

Saying it looks like someone else's data does not
remove it from the game but that cannnot, cannot, cannot be convincing.

I guess you mean e.g. the arm model matching the MIT measurements? Is there
something wrong with it being someone else's data? I should think that
would enhance the case for it being replicable and "objective." Or are you
referring to something else? (Your jab jab jab style leaves a lot to be
filled in by imagination.)

"Looks like" must be given a rigorous treatment.

[...]

HPCT is a special hypothesis that sounds neat, maybe "on the right track",
but that needs to be tested. [...] demos are not cumulative
tests but demonstrations of principle.

[...]

there is almost no evidence [for HPCT]
as a behavioral theory either.

What constructive suggestions have you? What would constitute a valid test?
What kind of work is needed for HPCT to graduate from "proposal" to "theory"?

Being convincing is another matter, going outside of science to the social
and political dynamics of networks (not communities!) of scientists and
others. I won't get into that here, in particular not the social dynamics
of CSG, and I'm not asking you to try to find a solution for that, which
obviously aggrieves you (and me). Just the question of test and validation.

  Bruce