what is a model?

[bnhpct]

[From Bruce Nevin (980811.1012 EDT)]

Marc Abrams (980810.2042)

I am saying the model does _NOT_ represent _any_
_particular_ _real_ data set. It _represents_ the components of the
control process and _how_ they interact. What is of concern and
interest in _this_ model ( i.e. coercion) is to see if the model
_structure_ produces the theorized behavior.

The "control process" with its elementary control systems (ECS) and their
perceptual input functions, comparators, and other components, is an
abstract model of what we observe living organisms doing, namely,
controlling their perceptual input. It is a statement of theory. You say
the spreadsheet is a model of the control process. That means it is a model
of an abstract theoretical model of a process that we impute to what we
observe in living organisms. I suspect that wasn't what you meant to say.

The spreadsheet in e.g. Coercion4c.xls is supposed to represent two control
systems in conflict, each represented by a single loop, one ECS, no
hierarchy. By setting the output strength of one ECS high and that of the
other ECS low we represent coercion in its simplest sense, overwhelming use
of force.

A model is a model of something. It is not a model of a model or a model of
a theory or a model of a theoretical construct. You model something that in
nature is to large or too complex or too embedded in its natural context to
understand or manipulate experimentally, and you reduce it to an analog
that is smaller or simpler or removed from its natural context to a
simulated environment, so that you can manipulate it experimentally and
come to understand it better. To verify that your new understandings are
valid, you have to verify that there is a valid relation of the parts and
relationships in the model to parts and relationships observed in the
natural object. When you are modelling behavior, you have to measure inputs
and outputs in nature and in the model and show that they are the same.

A PCT model is a model of the control of specified perceptions of a
controlled environment variable against measured disturbances as observed
in living control systems. I cannot speak as an expert in this, but this
appears to be the method. To specify the perceptions we measure values of
the identified environment variable. We measure the effect of specified
disturbances on the variable, and reduce the effects of unmetered
disturbances to zero if possible. We measure the organism's outputs
affecting the variable. We then build a model. It is not a model of the
control process--it is not a model of the theoretical model of control in
the abstract. It is a model of of the control system that we have observed,
controlling the input values and producing the output values that we have
measured. We adjust the internal parameters of the model until its inputs
and outputs match these observational data.

This explains why so much work is done with tracking experiments. The
inputs and outputs are easy to measure. In particular, you can eliminate
most unmetered disturbances (not all). In naturalistic observation of
living control systems this is not possible. To study coercion, we need
something like the tracking experiments. But before we abstract the
phenomenon, coercion, to the artificial simplifications of a tracking
experiment, paring the natural phenomenon down to its essentials, we need
to be clear about what is essential to the natural phenomenon.

Rick says the spreadsheet is a model of coercion. In justification, we see
that he made a theoretical interpretation of what he observes as coercion.
The "data" being modelled are some generalizations about naturalistic
observations. His interpretation is that two control systems are in
conflict over the state of an environment variable. Only one of them is
able to control the state of the variable, the other is so much weaker that
it has no effect on the state of the variable.

To test the proposed model, we go back to observations of living control
systems. A cat wants to go through the cat door that it normally uses. A
person is holding the cat door closed. The cat pushes its head against the
cat door, and it doesn't move. The model predicts that the cat increases
its output to the maximum of its capacity, which is not strong enough to
make the door move at all. The model predicts that the cat keeps doing that
forever. What does the cat do? The cat pushes once. It pushes again,
harder. It sits down and looks up at the door. It leans forward and sniffs
the cat door. It pushes again, not as hard. It looks up and miaows
plaintively. It sniffs at a flower pot next to the door. It turns around
and sits down again. It miaows several times, glancing up at the door. It
switches its tail. Abruptly, it gets up, walks to the side, and leaps up to
a windowsill. It looks in through the window, miaowing. Another occasion,
or another cat, and we observe different actions. In no case do we see what
the model predicts.

Perhaps it is not a good model. Perhaps it is incomplete. Perhaps we should
have looked at more observational data before making quick generalizations,
perhaps our generalizations were too much informed by our theory-based
expectations and not enough informed by observation of living control
systems in coercive conflict.

Coercion appears to be a special case of conflict, a limiting case on a
scale varying the balance of power between the conflicting control systems.
Here's a simple case that's been observed repeatedly with the rubber band
experiments. It's described at the beginning of Chapter 6 of Bill's new
book. One person is willing to break the rubber band, the other is not.
This means that the second person has an output limit (set by a
higher-order loop controlling "rubber band breaking" with a reference level
of zero), and the first person does not. The person without the output
limit is coercing the person with the limit. The knot comes closer and
closer to the mark that this unlimited person is trying to reach.

But then what happens? The person with the output limit (due to internal
conflict) stops controlling. Typically, they either protest that the rubber
band will break, or they start talking about the experiment and what it
means. In other words, they shift attention to the source of their
limitation (a weaker person in a tug of war might say "I'm not strong
enough! I give!") or they go up a level. Either way, they stop trying to
control what they have found that they cannot control.

From observational data like these (and I do not pretend these are complete

or adequate for generalizations) we can begin to design a model of the
simplest form of coercion, overwhelming use of force in a conflict
situation. It might turn out that the other forms of coercion would not be
so difficult to model, given that starting point.

So there are several steps away from naturalistic observation. First, we
identify what we believe is essential in what we are observing. For social
interactions it is not practical and often not ethical to measure inputs
and outputs directly in a natural setting. So we set up laboratory
experiments that replicate what we have postulated to be essential in the
natural setting. Then build a model that performs as the laboratory
participants did. Then test the model against performance data. Then verify
that the naturalistic observations are indeed a valid interpretation of the
artificial laboratory interactions. This must be done or the point of the
tracking experiments is lost to all but a small, sectarian audience,
essentially a modeller subset of the CSG. And it must be done or we risk
producing inadequate models and making interpretations of them that far
overreach what they validly show. This is what has happened with the
coercion spreadsheet.

  Bruce Nevin

[Marc_Abrams5]

From [ Marc Abrams (980811.1325) ]

[From Bruce Nevin (980811.1012 EDT)]

Marc Abrams (980810.2042)

I am saying the model does _NOT_ represent _any_
_particular_ _real_ data set. It _represents_ the components of the
control process and _how_ they interact. What is of concern and
interest in _this_ model ( i.e. coercion) is to see if the model
_structure_ produces the theorized behavior.

The "control process" with its elementary control systems (ECS) and

their

perceptual input functions, comparators, and other components, is an
abstract model of what we observe living organisms doing, namely,
controlling their perceptual input. It is a statement of theory. You

say

the spreadsheet is a model of the control process. That means it is a
model of an abstract theoretical model of a process that we impute to
what we observe in living organisms. I suspect that wasn't what you

meant >to say.

Thank you Bruce, that is _exactly_ what I was trying to say.

The spreadsheet in e.g. Coercion4c.xls is supposed to represent two
control systems in conflict, each represented by a single loop, one

ECS, >no hierarchy. By setting the output strength of one ECS high and
that of >the other ECS low we represent coercion in its simplest
sense, >overwhelming use of force.

Yes, It is a very basic model made to show the phenomenon of coercion
( control ) The hierarchy is _implied_. This model cannot show the
effects of coercion throughout the entire system. That was _not_ the
intent. you tried extending it, and _this_ model is inadequate for
those extensions. It doesn't mean your conjectures are untrue. It just
means, _that_ model is inadequate to explain it.

A model is a model of something. It is not a model of a model or a

model >of a theory or a model of a theoretical construct.

Yes, it is a "representation". If the PCT model is not a model of a
theoretical construct can you please tell me where in your body the
comparator lies. Can you tell me what the input function looks like?
Aren't these theoretical constructs? If not what are they? Physical
realities?

You model something that in nature is to large or too complex or too
embedded in its natural context to understand or manipulate
experimentally,

You also model things that you can't see. Like the interactions of
atoms, and the formation of stars. Your models are nothing more or
less then how you perceive the world. Seeing little orbits of neutrons
and protons around the nucleus gives us a "representation" of an atom.
It is _not_ how an atom _actually_ exists.

and you reduce it to an analog that is smaller or simpler or removed
from its natural context to a simulated environment, so that you can
manipulate it experimentally and come to understand it better.

This holds _only_ if you understand the _structure_ that produces the
behavior. Data is not used in validating _STRUCTURE_ . What you look
for, is to see that the graphs of the sim over time ( sim time )
_produce_ the same graphs as the phenomenon your looking at ( _any_
numbers will do ). That is the _FIRST_ step. Hopefully the model
( i.e. the structure )helps you determine _what_ kind of data you
might need to validate the model. That is not always so clear or
obtainable. That _doesn't_ invalidate the model. Experimentation
continues ( RESEARCH ) _using_ the model to help explore the
ramifications and see if the model _continues_ to produce the behavior
expected.

To verify that your new understandings are valid, you have to verify

that >there is a valid relation of the parts and relationships in the
model to parts >and relationships observed in the natural object.

What "natural objects are you referring to? What "natural objects" are
part of human behavior?

When you are modelling behavior, you have to measure inputs and
outputs in nature ...

Says who? How can you "see" what an "input" looks like, how do you
_know_ that what you see for instance is an "input". Maybe what you
see is an _output_ of some other function in your body.

and in the model and show that they are the same.

Yes, and I don't need _real_ data to show this.

A PCT model is a model of the control of specified perceptions of a
controlled environment variable against measured disturbances as
observed in living control systems. I cannot speak as an expert in

this, but >this appears to be the method.

A PCT model is a theoretical construct to show _a_ _representation_ of
the structure necessary to _produce the behavior. Nothing more,
nothing less. The basic model needs to be extended by RESEARCH not bs.

To specify the perceptions we measure values of the identified
environment variable.

No, not quite, We _conjecture_ what _we_ think the CV's are. If we
do the test right we might even have a better then 50% of coming
_close_ to what the CV's _actually_ are. Of course we will _Never_
know for sure. What values do you measure to specify perceptions &
CV's?

We measure the effect of specified disturbances on the variable,

How can you do that when you can never be sure _exactly_ _what the
variable is? Science demands _Proof_ ( right, Isaac) Whats the
data that validates the disturbance, the variable and the effect?

and reduce the effects of unmetered disturbances to zero if possible.

How can you _know_ this? What's _your_ data? How can you seperate the
"unmetered Disturbances" from the measured ones? How would you know it
if you _saw_ it?

We measure the organism's outputs affecting the variable.

Again, How can you seperate the outputs affecting the variable and the
outputs that are unintendend side effects. How would you _know_ what
you were seeing?

We then build a model.

Interesting. Based on what? All the "data" you were able to collect? I
don't think so.

It is not a model of the control process--it is not a model of the

theoretical >model of control in the abstract.

Thats exactly right. Truer words have never been spoken

It is a model of of the control system that we have observed,

It's BS. It's not PCT because you _can't observe the PCT process. Nice
coffee house chat that has zero basis of validity.

controlling the input values and producing the output values that we

have

measured.

Sounds like some great S ->R stuff. Exactly where does PCT come
into it?

We adjust the internal parameters of the model until its inputs
and outputs match these observational data.

You don't need PCT for this. You can generate this kind of stuff with
good 'ol behavioral research.

Exactly what kind of parameters do you have in your model?

This explains why so much work is done with tracking experiments. The
inputs and outputs are easy to measure.

I don't think thats the only reason. The output data ( graph over
time ) _SUPPORTS_ the _conjectured_ structure.

In particular, you can eliminate most unmetered disturbances (not

all).

This is irrelevant.

In naturalistic observation of living control systems this is not

possible.

I agree.

To study coercion, we need something like the tracking experiments.

What the tracking experiments show, is behavior that correlates to the
_expected_ or theoritized behavior one would expect _from_ a control
process. _Nothing_ has been _proven_. It's a theory, a conjecture.

But before we abstract the phenomenon, coercion, to the artificial
simplifications of a tracking experiment, paring the natural

phenomenon >down to its essentials, we need to be clear about what is
essential to the >natural phenomenon.

_Who's_ "simplifications" are you speaking of? What theory are those
simplifications based on? _Who_ determines the "essentials"?

Rick says the spreadsheet is a model of coercion. In justification,

we see

that he made a theoretical interpretation of what he observes as
coercion.

Not only Rick. I think it "represents" coercion. I also think Bill
feels this way.

The "data" being modelled are some generalizations about naturalistic
observations.

Huh !. The "data" are some numbers used to pass through the system to
produce an output graph. The data is _not_ meant to represent any
"naturalistic observations".

His interpretation is that two control systems are in
conflict over the state of an environment variable. Only one of them

is

able to control the state of the variable, the other is so much

weaker that

it has no effect on the state of the variable.

Exactly. I agree with this interpretation.

To test the proposed model, we go back to observations of living

control

systems.

You can _only_ observe _one_ aspect, the _output_. All the other
aspects of the _theoretical_ model are unobservable.

A cat wants to go through the cat door that it normally uses. A
person is holding the cat door closed. The cat pushes its head

against the

cat door, and it doesn't move. The model predicts that the cat

increases

its output to the maximum of its capacity, which is not strong enough

to

make the door move at all. The model predicts that the cat keeps

doing >that forever. What does the cat do? The cat pushes once. It
pushes >again, harder. It sits down and looks up at the door. It leans
forward and >sniffs the cat door. It pushes again, not as hard. It
looks up and miaows

plaintively. It sniffs at a flower pot next to the door. It turns

around

and sits down again. It miaows several times, glancing up at the

door. It

switches its tail. Abruptly, it gets up, walks to the side, and leaps

up to

a windowsill. It looks in through the window, miaowing. Another

occasion,

or another cat, and we observe different actions. In no case do we

see >what the model predicts.

The model cannot "predict" your little story. It is _not_ meant to
tell this story. Do you have any idea what it would take to model this
specific situation? Probably not. If _your_ little story could be
modeled PCT would be a done deal. We would fully understand all the
levels and how they interact, PCT would no longer be a theory. It
would be _fact_.

Perhaps it is not a good model. Perhaps it is incomplete. Perhaps we

No, it's a very good model it shows coercion. _ALL_ models are
incomplete.

should have looked at more observational data before making quick
generalizations,

And what data is that?

perhaps our generalizations were too much informed by our

theory-based

expectations and not enough informed by observation of living control
systems in coercive conflict.

Actually I think you got it turned around the wrong way

Coercion appears to be a special case of conflict, a limiting case on

a

scale varying the balance of power between the conflicting control
systems.

I agree, and the model _shows_ this

Here's a simple case that's been observed repeatedly with the rubber
band experiments. It's described at the beginning of Chapter 6 of

Bill's >new book. One person is willing to break the rubber band, the
other is not.

This means that the second person has an output limit (set by a
higher-order loop controlling "rubber band breaking" with a reference

level

of zero), and the first person does not. The person without the

output

limit is coercing the person with the limit. The knot comes closer

and

closer to the mark that this unlimited person is trying to reach.

Thats why I said before that you really need a 3 level model to _try_
to _begin_ to understand the ramifications.

But then what happens? The person with the output limit (due to

internal

conflict) stops controlling.

Lots of stuff can happen. Thats why you need research and models. How
do you observe _this_ data?

Typically, they either protest that the rubber
band will break, or they start talking about the experiment and what

it

means. In other words, they shift attention to the source of their
limitation (a weaker person in a tug of war might say "I'm not strong
enough! I give!") or they go up a level. Either way, they stop trying

to

control what they have found that they cannot control.

For every "observation" you make I can come up with 1,000 more. So?
What's the point?

From observational data like these (and I do not pretend these are
complete or adequate for generalizations) we can begin to design a

model

How? Your "observational data" can be modeled _any_ way you like. What
did you have in mind?

of the simplest form of coercion, overwhelming use of force in a

conflict

situation. It might turn out that the other forms of coercion would

not be

so difficult to model, given that starting point.

Really. You have a "simple" coercion model. Try doing some other
forms. Do you have something specific in mind. Hopefully my 3 level
System Dynamics model ( in Stella and Vensim ) will help you, me and
other interested people in exploring some of these things. But
"simple" it ain't and it's just a _beginning_.

So there are several steps away from naturalistic observation.

Again Bruce, we can only "observe" the output. _All_ other aspects are
hidden.

First, we identify what we believe is essential in what we are

observing.

I don't agree. PCT tells you that what we "observe" could ( and
usually does ) have very little to do with _why_ we produce the
actions we do. We can't "trust" what we see alone.

For social interactions it is not practical and often not ethical to

measure >inputs and outputs directly in a natural setting.

Besides, even if you could, which you can't :-). we don't have a model
that could handle that kind of data

So we set up laboratory experiments that replicate what we have
postulated to be essential in the natural setting. Then build a model

that >performs as the laboratory participants did.

I think you got it backwards again. How do you know _what_ experiments
to set up? By observing output?, Input?, Both, There are bunches of
"experiments" you can set up based on that. Check with the Psych lit.
Bruce Abbott could probably help.

Then test the model against performance data.

What performance data?

Then verify that the naturalistic observations are indeed a valid
interpretation of the artificial laboratory interactions.

How do you do that? By observing the output?, Input?, Both?

This must be done or the point of the tracking experiments is lost to

all but >a small, sectarian audience, essentially a modeller subset of
the CSG. >And it must be done or we risk producing inadequate models
and making >interpretations of them that far overreach what they
validly show. This is >what has happened with the coercion
spreadsheet.

I guess thats why S -> R is so popular. No General theory, no models,
just plain old observations. can't get much easier then that. Bill why
didn't you think of that?
Btw Bruce, _you_ and others ( Isaac, Ken, Tim, Bruce Gregory. Am I
leaving anyone out ) made those overreaching interpretations. Rick
simply tried sticking to the _basic_ phenomenon. That didn't seem to
be good enough. Too Bad.

Marc

[bnhpct]

[From Bruce Nevin (980811.1709 EDT)]

Rick Marken (980811.1215) --

I don't mean to pontificate about scientific method, just to remind because
in some posts (this was not a reply to one of yours, Rick) there seemed to
me to be some confusion about the relationships among models and
theoretical constructs and naturalistic observations and the data of
measurement. Because I was reminding and not writing a treatise on
scientific method or on experimental methodology (you've written a book
about that), perhaps I was too terse to be clear. It is altogether too easy
to be prolix.

I'm afraid I find this post very confusing; I think you are
saying that we can use PCT to understand a particular
kind of behavior only after we have tested the PCT model of
that behavior. Is this what you mean.

More like, you cannot legitimately claim that your model of "a particular
kind of behavior" enables you to understand it until you have tested the
model by comparing measurements of its performance with measurements of
organisms' performance. We improve our understanding by an iterative,
bootstrapping process. You start with observations. You make your best
guess at a model to account for what you are observing. You set up a
controlled environment to get some solid data to model. You build a model
to do what participants in the lab experiment are doing. You compare the
measured performance of the model with the measured performance of the
people in the experiment. You revise the model, or perhaps replace it with
another, as means for controlling the difference between the two sets of
performance data, model and organism. You cannot attain the next
improvement of your understanding until you have tested your current model,
studied the deficits in its performance, and come up with some hypotheses
to explain them. These hypotheses are your new understanding, still
tentative. The new understanding has to be verified by using it to modify
(or replace) the model and then test it.

In several posts I have compared spreadsheet performance with that of
living organisms engaged in coercion and described differences that I
observed. More detailed and precise testing of the spreadsheet is not
possible because there are no measurements of the performance of organisms.
However, even these impressionistic observations are enough to show that
the model needs to be improved. I have proposed some specific hypotheses to
explain the deficits in the performance of the spreadsheet, and have
sketched the kinds of improvements to the model that might be needed. Until
these are implemented and tested we don't know whether my hypotheses are
right or wrong. But we can say with confidence that, while the present
spreadsheet is a good start, it is not an adequate model of coercion. The
iterative process of science must continue.

  Bruce Nevin

[Marc_Abrams5]

From [ Marc Abrams (980811.2353) ]

[From Bruce Nevin (980811.2304 EDT)]

Let me try to clarify my position.

Marc, I don't understand much of the many things you seem to be

driving at

in this long post. I just can't seem to pull it together and make

sense of

it. Must be tired.

First. I believe I share with you and Isaac and Bruce Gregory and Ken
and Tim a _strong_ desire for _a_ PCT model that will give us the
necessary insights to answer _some_ of the questions you and others
have raised about the notion of coercion, cooperation and other
behavioral phenomenon. The basic model as currently configured
_cannot_ do this.

Second. Because the only _visable_ aspect of the control
"representation" ( model ) is the output. Identifying _where_ and
_how_ to get the data becomes an issue. ( i.e. What does the
comparator look like?, Where is it? How would I know it if I saw it? )
Since this is a problem it becomes difficult if not _currently_
impossible to Iteratively model - compare to existing data- modify
model- compare again. In a perfect world I agree that that would be
the method of choice. I do not think we currently have that choice.
The "data" is simply not currently available. Hopefully as our
knowledge expands these barriers will drop. So, What can we do in the
mean time? Well I feel we could develop more sophisticated models.
This is _not_ a trivial task. As evidenced by the lack of people
capable of doing them well. This is not limited to PCT modeling.
Modeling is both an art and science and difficult to master. Because
of lack of data modelling becomes _critical_ (in my opinion ) to our
understanding of PCT. But _only_ models that help answer the kinds of
questions people are interested in. This will take time and hard work.
I believe both Bill and Richard Kenneway are working on some _very_
interesting models. Things will get better

One thing I can respond to:

Btw Bruce, _you_ and others ( Isaac, Ken, Tim, Bruce Gregory. Am I
leaving anyone out ) made those overreaching interpretations.

Rick

simply tried sticking to the _basic_ phenomenon. That didn't seem to
be good enough. Too Bad.

The overreaching interpretations were made by Bill and Rick.

I stand corrected. The overreaching was done by both.

The spreadsheet was claimed to be a model that explains interactions
between nazi guards and compliant prisoners, between rapists and rape
victims, and so on. It is not.

I read it differently. I think the claim was that the spreadsheet
_showed_ the phenomenon of coercion. The pontification was all Rick,
not model. Rick as well as you and Tim an Ken have _very_ strong
feelings about coercion. The model was left at the starting gate by
both sides

That is why and how I got involved in the coercion discussion. Not
because of anything to do with RTP, but because from a simple model
broad social interpretations were being extrapolated that the model
could not support.

I know this and agree a 100%. So with that in mind I will ask both
Rick and the others, Without a more sophisticated model, what can
further learned from this discussion?

Hoped this helped to clear up the post a little

Marc

I'll look at your post again. Maybe I can figure out something else

to say.

···

       Bruce Nevin

[Marc_Abrams5]

From [ Marc Abrams (980812.0925) ]

[From Bruce Nevin (980812.0713 EDT)]

Marc Abrams (980811.2353) --

The epistemological difficulty of doing science is much deeper than

you

have said. It is an extension of the general problem of imputing an
external reality from our perceptions, when all we have is our

perceptions.

Consistency is a cornerstone.

No disagreement here. What if we are consistently wrong? How do we
know?

When you create a model, and its performance--its inputs and

outputs-->are virtually identical to those of the organism being
modelled, there is >good grounds for imputing to the organism the
internal structure that you >can observe in the model, even though you
cannot (yet) observe it in >action in the organism.

Bruce, this is the heart of my argument. There are many different
structures that can take your inputs and produce your outputs. Skinner
was able to "show" certain inputs resulted in certain outputs. Thats
one of the reasons PCT is such a difficult sell. It's _not_ only about
what we 'see' going in and coming out I don't think there is a
disagreement between us about the "fact" of control. The spreadsheet
model on coercion "showed" the _basic_ phenomenon of coercion between
_two_ control systems. I don't think Rick was wrong in stating what
coercion was. I think it was incorrect to say that the model
"explained" the _entire_ continuum of coercion, But I don't think that
was Rick's intent. When Rick tried to make an emphatic point ( by
using the rape and Nazi examples ) all hell broke loose. Not because
there was a disagreement about _what_ coercion was. But _when_ could
the behavior be called coercion. _That_ the model could _not_ answer.
trying to justify "either" position with that model was and is
meaningless. So instead, terrific stories were created and agrued,
none of them justifiable _based_ on a working model. This is "coffee
house chat" Interesting ( at times ) but _infinite_.

Have you read Phil Runkel's book, _Casting nets and testing

specimens_?

Yes I have. A terrific read.

Have you read Bill's "Epistemology" paper in LCS II? His "Applied
epistemology" paper in LCS I?

Not yet, But I will.

Marc

[bnhpct]

[From Bruce Nevin (980812.0713 EDT)]

Marc Abrams (980811.2353) --

The epistemological difficulty of doing science is much deeper than you
have said. It is an extension of the general problem of imputing an
external reality from our perceptions, when all we have is our perceptions.
Consistency is a cornerstone.

When you create a model, and its performance--its inputs and outputs--are
virtually identical to those of the organism being modelled, there is good
grounds for imputing to the organism the internal structure that you can
observe in the model, even though you cannot (yet) observe it in action in
the organism.

Have you read Phil Runkel's book, _Casting nets and testing specimens_?
Have you read Bill's "Epistemology" paper in LCS II? His "Applied
epistemology" paper in LCS I?

  Bruce Nevin

[Marc_Abrams5]

From [ Marc Abrams (980812.1148) ]

[From Bruce Nevin (980812.0713 EDT)]

Good. Because these statements are not true:

There are many different
structures that can take your inputs and produce your outputs.

Skinner

was able to "show" certain inputs resulted in certain outputs.

That's

one of the reasons PCT is such a difficult sell.

This is _not_ about competing theories.

This is about _modeling_. Given a black box ( i.e. any number of
different model structures ) you can, given certain input, produce
certain output. It has _nothing_ to do with any underlying theory. My
point is, that you are taking on faith the mechanisms that are
generating the behavior. There is no proof that a comparator exists or
any other aspect of the Control process model. Ptolemy was able to
"explain" the movement of the planets. He even had a very elegant
epicycle model. The point here is _not_ to challenge the basic
precepts of PCT. I simply believe it is _dangerous_ ( although a lot
of fun ) to _extrapolate_ from a model information it does not
contain.

Marc

[bnhpct]

[From Bruce Nevin (980812.0713 EDT)]

Marc Abrams (980812.0925) --

When you create a model, and its performance--its inputs and

outputs--are virtually identical to those of the organism being
modelled, there is good grounds for imputing to the organism the
internal structure that you can observe in the model, even though you
cannot (yet) observe it in action in the organism.

Bruce, this is the heart of my argument.

Good. Because these statements are not true:

There are many different
structures that can take your inputs and produce your outputs. Skinner
was able to "show" certain inputs resulted in certain outputs. That's
one of the reasons PCT is such a difficult sell.

You believe this after reading Runkel?

There are no alternative theories or methodologies that reproduce, in
action, the behavior of an individual living organism with near 100%
fidelity to performance data. Count them: zero, none. PCT is the only
contender.

If you want to pursue this, others should take it up with you. Rick's been
in the thick of it for many years, and knows the other methodologies well,
he wrote a textbook on psych methods before learning about PCT. Ask Bruce
Abbott, I think he'll make some strong distinctions between what PCT can do
and what Skinner could do. (I don't pretend to understand the EAB issues
that he's been involved with.) Read Bill's papers on behaviorism in LCS,
the Spadework paper, the "fallacy of behaviorism" paper in LCS II.

The reason PCT is a "difficult sell" is not that other theories do as good
a job. Their criteria of adequacy have slipped so low into marginal
statistical correlations, because they can do no better, that they cannot
imagine that 99+% correlation, individual by individual (not across
populations), can possibly be true. Isaac can tell you about the bright new
faculty member who looked at demos on Tom's PC. "Do you know what this
means?" Yup, says Tom. "This means the end of psychology as we know it."
Yup, says Tom. "This means that the last hundred years of psychological
research is mostly a waste of time." Yup, says Tom. The guy borrows the
demos. Tom can see him mightily disturbed by them in his office over the
next days. Then he returns the demos, refuses to have anything more to do
with them, and becomes Tom's bitterest opponent on the faculty. Conflict
resolution? The career cost apparently was too great. Could this make PCT a
hard sell?

I'll stop there. Others can speak to this issue far better than I.

  Be well,

  Bruce Nevin

[bnhpct]

[From Bruce Nevin (980812.1210 EDT)]

Marc Abrams (980812.1148) --

If you construct a model that controls its inputs by means of the
environmental effects of its outputs the way a living organism controls and
the way a good PCT model controls, and then when we look inside the black
box it is structured differently from the way we now construct PCT models,
I think that will be an important contribution to PCT. I look forward to
seeing it.

  Bruce Nevin