[From Bill Powers (2003.11.29.0710 MST)]
Marc Abrams (2003.11.28.2024)–
No, and let me try to explain
why. Besides being incomplete, to say that
Rick’s model is a working HPCT model I believe is intellectually
Why do you have to use a belligerent term like “intellectually
I don’t think you know what is meant by a “working model,” or
what the other kind of model would be. A working model is one which
actually runs on a computer (or one that is actually constructed, of
course), starting from a known state and then producing changes in all
the variables through time. The other kind of model is a
conceptual model, which may be drawn with boxes and arrows, but
which is not specified fully enough to operate. The conceptual
model just shows some proposed relationships among variables, but it
doesn’t demonstrate what would happen in a real system organized as the
model suggests. People may claim they know how a system organized like a
conceptual model would behave, but they really don’t – the only way to
find out for sure is to develop the conceptual model into a working model
and run it. Another name for working model is “simulation,” if
you mean a computer model.
Rick’s spreadsheet hierarchy is clearly a working model, not a conceptual
model. When the variables are set to initial conditions and the program
is started, they change through time, showing what the model’s
organization would actually produce if a real system were organized that
“Working model” does not mean the same kind of thing that
“working hypothesis” means. The term does not mean a model that
is good enough to work with, as opposed to the final version of the
model. I have seen a few comments indicating that this usage of
“working” is how some people have taken this term. That is
simply a misunderstanding, perhaps indicating that the writer hasn’t
realized that a simulation actually operates in imitation of a
(hypothetical) real system, its variables changing value as time
proceeds. And the changes are not preprogrammed as in an animation; they
grow out of the relationships among variables according to the equations
showing how each variable depends on others from one instant to the next.
The programmer usually doesn’t know exactly what behavior will result
when the model is run; often it is totally surprising.
To say that one has a working model is not to say that the model behaves
like some real system. It just says that the model behaves, as opposed to
a conceptual model which does not produce any behavior. If the real
system were measured under conditions like the simulated conditions in
the model, the behavior of the working model could be compared with the
behavior of the real system, allowing the adequacy of the model to be
assessed. It is possible to have a working model of a system, and
discover that it is completely incorrect. It is still a working model. A
conceptual model, on the other hand, can’t be compared with real behavior
at all: one simply claims without proof that the conceptual model
represents the real system.
You seem to show some appreciation of this distinction, though the
questions that follow seem to deny that.
Sure it’s a ‘working’ model.
The question becomes, of what? Which
levels in the hierarchy does Rick’s model purport to represent? Does
‘real’ HPCT hierarchy have 3 or 400 levels?
These questions are completely irrelevant to the question of whether
principles of HPCT have been demonstrated with working models. Why are
you even asking whether the levels in Rick’s hierarchy are like levels in
the real system? Do you know what the levels in the real system are? I
don’t say this very often, but that’s a dumb question. The whole point is
to develop a working hierarchical model so we can have something to
compare with experiments using real organisms. Until we understand how
hierarchies of control systems would operate, we don’t have anything to
compare with real behavior.
Do you think that just drawing a diagram is sufficient to show how a
model would actually operate? You have to try out various possible
organizations as working models, and run them, and see how they
actually behave as opposed to how you thought or hoped they would behave.
Rick’s model showed what we suspected and hoped might be the case – that
it is possible for multiple control systems working at three levels to
control their own perceptions relative to arbitrary reference levels
without conflict, where all the systems act through a shared environment.
We hoped that would be true, but neither we nor anyone else knew
it was true until the working model proved that it could be
So you see that your impatient questions about how many levels there
really are, and what their order is, or even what their names are, are
totally premature. It’s like asking Neils Bohr, “Why haven’t you
invented lasers yet?” You can’t go any faster than the models are
developing. And the first step is the hardest: showing that a proposed
model would actually do something vaguely resembling the way Nature
actually works. We’re a little past “vaguely,” but not very
Bill has hypothesized 11 [levels].
Has he been able to validate that?
Another – uh – unreasonable question. Why do you suppose we are trying
to develop working HPCT models in the first place? So we can think of
experiments in which a model’s performance can be compared with a real
person’s performance, which is our only way to judge what is right and
wrong about the model. The 11 levels result from an analysis of
experience, not from a model. They, too, are tentative, but
they’re not a model.
I would love to be able to test the hypothesis that there is a level of
human control concerned with configurations, as experience seems strongly
to suggest. The only problem is that I don’t know how to construct a
working model that can monitor the states of configurations in real time.
Controlling them would be relatively easy; the problem is finding a way
to perceive them. Of course I don’t mean a way for me to
perceive them; I do that very easily. But how can I make the model
perceive them? That’s the hangup. And that’s only level 3.
You show your lack of understanding of these problems in other ways: I
can set up a model for any one level, and run it and match it to
behavior, but we have to cheat on the perceptual input function: it
creates a perceptual variable that corresponds to what we can observe in
environment. We can’t show how that perception is derived from
aspects of the environment. To do that, we would have to know a lot
than anyone now knows, anywhere".
and you said
I believe this is true only if you
limit yourself by trying to build
perceptions using Bill’s current hierarchy. Perceptions are composed of
types and kinds of inputs, and environmental inputs for a large number
perceptions play almost no role at all. So who says perceptions can
built this one way?
Suppose we forget the proposed hierarchy, and just try to build a model
that can perceive and control some simple configuration, like the
orientation of a cube in space. The input would be two two-dimensional
retinal arrays of light and dark, and the output would be a signal
representing some dimension of the orientation of the cube, say the angle
between one of its faces and a vertical plane (the controller wants to
look at the cube face-on). Does that problem get any easier if you say
this perception is part of a network rather than a hierarchy? Or if you
propose other sets of inputs? Remember, of course, that we must start
with the real inputs, the images on the retinas. You don’t have a
free choice of inputs.
To me Bruce, having a ‘working’
HPCT model means having a model that can
validate the theory, or at least it’s major
You have that exactly backward. Before you can validate a theory, you
have to present the theory in the form of a model that actually produces
– and thereby predicts – behavior: a working model. Only then can you
compare the behavior of the model with the behavior of the real system,
to see where the theory is wrong or inadequate. Of course some theories
are presented as mathematical equations which can be used for
predictions. But that is basically what working models are, with the
addition of moving pictures or time plots describing the solutions of the
To me, Rick’s HPCT spreadsheet
model simply acknowledges that theoretically,
a hierarchy is possible. I certainly have no quarrel with
Why not? Did you know already that such a hierarchy could actually work?
And if you did, how did you know that? And if you do know that, why are
you proposing a network rather than a hierarchy? Do you also know that a
network would work? Do you even know the difference between a network and
why are you saying I feel that the HPCT model is adequate and
First, I am not saying. I’m asking.
I’m asking because I really don’t know
what you do or don’t want to consider with regard to HPCT. You might
said what you did in MSOB but when you accused me of attacking you
malice, what am I supposed to think?
When you tell me that the last good idea I had was 30 years ago, how am I
to perceive that remark as being anything but nasty and malicious? Your
mouth runs away from your brain sometimes. As long as you let it do that,
you can expect to catch some return fire. I could claim that your ideas
have a long way to go before they catch up to where I was 30 years
And when Bruce Gregory asked a
perfectly innocent question about the hierarchy you went over the
Bruce Gregory hardly ever asks a perfectly innocent question – do you,
Bruce? But I find that all I have to do is call his bluff, and he
turns back into the pussycat he really is.
Perhaps it’s the idea of a
hierarchy of control that you object to,
whatever levels are proposed.
Huh?, You haven’t read my posts. If you did this would be one thing
would not question.
Why not, when you said your were doubtful about the hierarchical approach
and planned to work with Levin on a SD model of a network? Do you read
your own posts?
The hierarchies that are currently
talked about in neuroscience are
physical ones, that is, from the spinal cord (lower level) to the
mid-brain to the thalamus and than to the cortex (higher
So are the hierarchies in HPCT, and if you read the neurology in B:CP you
will see that I tie the two together, even roughly localizing many (not
all) specific proposed levels.
of course many different hierarchies in the CNS, including one
only with the 6 levels of the cortex. There are no hierarchies that I
aware of that build perceptions with the property levels you
Of course not. Neuroscience is about 40 years behind PCT in proposing
models to test against reality. Mostly neuroscientists measure stimuli
and responses, showing that they’re more like 90 years behind.
I would be most interested in
reading about any such hierarchy you might be able to refer me
Just read the chapters in B:CP where the functional levels are related to
neuroanatomy and functionality in the CNS. Since you weren’t interested
in that when you first read the book, you may have skipped those parts.
You sure sound as if you skipped them.
Why is it a good place to start?
and What is going to change it? This was
started 30 years ago. What needs to happen in order for it to
That mouth is flapping again. Tend to it.
Look at the neurological evidence in B:CP. It’s not a circuit
diagram, but there are strong
indications of a hierarchical arrangement
(and no evidence at all of an arrangement in which every part of the
communicates directly with every other part). I will vigorously
concept of a control hierarchy, though I am much less convinced
particular levels I have proposed.
Bill, this passage baffles me. Where do you believe we differ
Well, you just said a few inches above that you doubted any connection
between my proposed hierarchy and actual physical hierarchies in the
brain. Do you read your own posts through a keyhole, so you can see only
the sentence before and after what you’re looking at?
You should be thankful that I’m in a good mood.