Computing output, model-based control

[From Rick Marken (970222.1120)]

Bill Powers (970222.0843 MST) --

I'm afraid that the term "computing output" has become a code
expression in PCT that threatens to substitute for thought.

Thanks for another inspired post.

I just want to add somthing that I tried to suggest yesterday, viz.,
that the term "model based control" has also become a code expression
that threatens to substitute for thought. What the expresssion was
originally meant to refer to was a particular control architecture:
one in which an adaptive mechanism is used to tune an "internal
model" of the environmental function that connects system output
to plant output. The model doesn't control a perception of plant output
(as does the PCT model); rather, it controls a modeled version of what
this perception _would be_ if the current version of the "internal
model" were a correct representation of the environmental function.

This narrow referent of "model based control" has been expanded (in
usage) to include many examples of good old fashioned control of
perception. For example, just yesterday Martin used the term to
refer to control of a perception relative to a variable reference;
the variable reference was taken to be a "model" of the temporal changes
of a variable in the environment. I also recall uses of the
term "model based control" to refer to control of an imagined perception
(as described in B:CP. p. 222); again, the imagined perception was taken
to be a "model" of some possible real perception. I have even heard the
term "model based control" used to refer to good old, plain vanilla
control of perception; in this case the controlled perception itself is
taken to be a "model" of some objective state of affairs (such as the
speed at which the car is moving down the road).

All of these uses of of the term "model based control" turn on the
notion that a "model" is a representation of something else. In the
"model based control architecture" the model is a representation of the
actual environmental (plant) function. In control of imagination the
imagination is a representation of past perceptions. In control
of perception the perception is a representation of some external state
of affairs. So all these uses of "model based control" are perfectly
appropriate-- just as it is perfectly appropriate to say that the PCT
model "computes output".

I don't think we going to make any progress toward understanding the
real differences between PCT and other explanations of behavior until we
can agree on what we are referring to when we say things like "computed
output" and "model based control". I think Bill Powers has been very
clear about what he means by both terms and I agree completely with his
meanings. I hope some others are willing to
agree to these meanings as well.

Hopefully

Rick

[Martin Taylor 970224 12:30]

Rick Marken (970222.1120)]

I just want to add somthing that I tried to suggest yesterday, viz.,
that the term "model based control" has also become a code expression
that threatens to substitute for thought. What the expresssion was
originally meant to refer to was a particular control architecture:
one in which an adaptive mechanism is used to tune an "internal
model" of the environmental function that connects system output
to plant output. The model doesn't control a perception of plant output
(as does the PCT model); rather, it controls a modeled version of what
this perception _would be_ if the current version of the "internal
model" were a correct representation of the environmental function.

This narrow referent of "model based control" has been expanded (in
usage) to include many examples of good old fashioned control of
perception. For example, just yesterday Martin used the term to
refer to control of a perception relative to a variable reference;
the variable reference was taken to be a "model" of the temporal changes
of a variable in the environment.

I did NOT. I used it to refer to the enhancement of control by the use
of a predictive _model_ of the kind you like to call an "internal model"
of the environmental function (which, as we know, is indistinguishable
from an "internal model" of environmental function and predictable
disturbance). Your control simulation _used_ a veriable reference that
predicted the disturbance, using a built-in "internal model." If the
variable reference had been an arbitrary variable reference, the tracking
error wouldn't have been affected at all by it (or hardly at all, and
not for the better). But it wasn't arbitrary; it was the output of a model
of the disturbance, 200 or 300 msec in advance of the changes in the
disturbance.

I think Bill Powers has been very
clear about what he means by both terms and I agree completely with his
meanings. I hope some others are willing to
agree to these meanings as well.

You haven't been listening (reading), have you?

We went through this "model seen as confusing terminology" discussion only
a couple of weeks ago. What we agreed was to try to avoid the use of the
term "model" in both cases, and replace it with terms like simulacrum
or internal replica, on the one hand, and mould or inverse on the other.

Yours was a simulacrum or internal replica.

Martin

[From Bill Powers (970224.1135 MST)]

Martin Taylor 970224 12:30]
Rick Marken (970222.1120)]

Rick, I think you'd better explain your model of the sine-wave stuff again.
I was under the impression that you were using rate-plus-proportional
perceptual feedback in the second level of control, which of course would
put a phase advance into the reference signal of the lower-level system. The
phase _advance_ would compensate for the phase _lag_ in the leaky-integral
output function. But you'd better straighten us out on this.

Martin, I have no recollection at all of this:

What we agreed was to try to avoid the use of the
term "model" in both cases, and replace it with terms like simulacrum
or internal replica, on the one hand, and mould or inverse on the other.

I said I was going to use the term "simulation". A simulation, to me, is a
set of computations that mimic the processes in a physical system as a set
of equations that create the same input-output (or other) effects. A mass on
a spring in a damping medium would be simulated as a differential equation,

m*(d2x/dt^2) - k*dx/dt - s*x = f(t),

the solution(s) of which would give us the position and velocity of the mass
as a function of the driving force f(t). Each term on the left refers to a
specific aspect of the physical situation: the inertial force, the viscous
drag force, and the spring constant of the restoring spring. The term on the
right represents the pattern of applied force.

That is all I agreed to; I left the uses of "model" up to you, since you
seemed to insist on using it to mean a lot of different things. If you like
you can use it to mean simulacrum or replica or mould or inverse or
perceptual function or output function or reference signal or anything you
like; it no longer has any technical meaning, by my understanding of the
agreement. Simulation, however, does. Not "simulacrum" or "replica" but
"simulation." That's the only term to which I respectfully lay claim.

Best,

Bill P.

ยทยทยท

I just want to add somthing that I tried to suggest yesterday, viz.,
that the term "model based control" has also become a code expression
that threatens to substitute for thought. What the expresssion was
originally meant to refer to was a particular control architecture:
one in which an adaptive mechanism is used to tune an "internal
model" of the environmental function that connects system output
to plant output. The model doesn't control a perception of plant output
(as does the PCT model); rather, it controls a modeled version of what
this perception _would be_ if the current version of the "internal
model" were a correct representation of the environmental function.

This narrow referent of "model based control" has been expanded (in
usage) to include many examples of good old fashioned control of
perception. For example, just yesterday Martin used the term to
refer to control of a perception relative to a variable reference;
the variable reference was taken to be a "model" of the temporal changes
of a variable in the environment.

I did NOT. I used it to refer to the enhancement of control by the use
of a predictive _model_ of the kind you like to call an "internal model"
of the environmental function (which, as we know, is indistinguishable
from an "internal model" of environmental function and predictable
disturbance). Your control simulation _used_ a veriable reference that
predicted the disturbance, using a built-in "internal model." If the
variable reference had been an arbitrary variable reference, the tracking
error wouldn't have been affected at all by it (or hardly at all, and
not for the better). But it wasn't arbitrary; it was the output of a model
of the disturbance, 200 or 300 msec in advance of the changes in the
disturbance.

I think Bill Powers has been very
clear about what he means by both terms and I agree completely with his
meanings. I hope some others are willing to
agree to these meanings as well.

You haven't been listening (reading), have you?

We went through this "model seen as confusing terminology" discussion only
a couple of weeks ago. What we agreed was to try to avoid the use of the
term "model" in both cases, and replace it with terms like simulacrum
or internal replica, on the one hand, and mould or inverse on the other.

Yours was a simulacrum or internal replica.

Martin

[From Rick Marken (970224.2240 PST)]

Bill Powers (970224.1135 MST) --

Rick, I think you'd better explain your model of the sine-wave
stuff again. I was under the impression that you were using
rate-plus-proportional perceptual feedback in the second level
of control, which of course would put a phase advance into the
reference signal of the lower-level system.

The sine wave stuff I am thinking of was done at least four years
ago -- maybe five. I don't even have the programs anymore; I think
I was writing this stuff as HyperCard stacks. All I remember about
it was that I was trying to think (not very effectively, as it
turned out) of a way to show two levels of control (in a behaving
subject) happening at the same time. I was writing the two level models
to see if there was some way to tease out two levels of performance in
the model; I would then test the subject with the method that worked on
the model.

That's all I remember; I think my ideas never got less vague than that.
I think I played around with sine wave disturbances of
different frequencies applied to perceptions controlled by
different levels of the model. I don't think I ever got to the
point where I successfully derived the varying reference input to
the lower level system from the error in the higher level system.
I seem to remember discussing how to do this (using rate plus
proportional feedback in the higher level system, possibly) when
you were visiting here several years ago, but I don't think
I got the program to work. Maybe I did. I don't remember. All I remember
is that I did not succeed in developing the kind of experiment I was
after -- one where it was possible to
_continuously_ monitor two levels of control simultaneously.

Now that I've explained the sine wave model, maybe you can explain
the following comment to Bruce Abbott:

I rather pounded on you in my post; got carried away (I visited
Rick only a week ago).

Did you mean:

a) I had to be polite when I was visiting Rick so I had a lot of
pent up pounding to use up.

b) Since I visited Rick only a week ago his pounding lessons were
fresh in my mind.

c) I mistook you for Rick because I had been visiting him so
recently (a "recency" effect;-))

d) all of the above

e) none of the above

f) refuse to answer ... :wink:

Love

Rick