Brownian control

[From Rick Marken (960212.0810)]

Me:

... can you tell me whether or not the envinronment controls
behavior using this meaning of "control"?

Hans Blom (960212) --

This question deserves a somewhat longer answer than a simple yes or
no.

I think a simple "no" would have done fine.

Note that in this experiment I do not need a "living" organism; it is
sufficient that the material properties of the particle are such that
its size varies with the concentration of some chemical. Yet it looks
as if the particle "controls" its position. The Test will demonstrate
this: pick up the particle and put it somewhere else. As soon as you
let it go, it will start to move towards the chemical's concentration
maximum again.

What do we have here, "control" or simply the result of a combination
of some laws of physics? Both, I would say.

I would say it's pretty unlikely that there is any control at all going on
here. But you have to do The Test correctly in order to find out. If you've
got this set up as a computer simulation just add a continuous disturbance to
the positions of the particles and see what happens; I bet that the
disturbance will be completely effective.

The version of The Test for control that you describe above is like pushing
on a stationary pendulum, watching it return to it's vertical position
(after swinging for w while) and concluding that the pendulum controls its
position. There is no control; just cause (gravity, friction) and effect
(pendulum line moves to vertical position). When there is control, resistance
to disturbance is happening _while_ the disturbance is affecting the
controlled variable!

In this case, what we call control is the result of the interaction of
physical laws and material properties.

I think you're one of only a few here who is prepared to call this "control".
_We_ don't call this control unless there is continuous resistance to
disturbance to a variable.

Why this extensive discussion? To show you how I prefer to look at control:
as an abstract, emergent phenomenon. The question "what controls what" is
meaningless for me.

I'm not sure this answer will satisfy you ;-).

It satisfies me because it helps me understand why you have been unable to
understand or accept PCT after all these years.

Control is a very clear and real phenomenon; it is only superficially similar
to stability phenomena like the one you described. If the question "what
controls what" is really meaningless to you (and I'll take your word that
it is) then PCT is definitely NOT for you.

PCT explains how living systems control perceptual representations of
environmental variables. It includes a method for determinig which perceptual
representations are controlled and which are not. This method also makes it
possible to determine what is controlling what; what is controller (the
control systems) and what is controllee (controlled variable). If the
distinction between controller and controllee is really meaningless to you,
then PCT must be meaningless to you, too.

Why study a theory of control if the phenomenon that the theory explains
(control) is meaningless (ie. non-existent; a word has meaning if it refers
to something; apparently the word "control" refers to nothing, from your
point of view)?

Best

Rick

[Martin Taylor 960212 15:30]
(Rick Marken 960212.0810) to Hans Blom

Control is a very clear and real phenomenon; it is only superficially similar
to stability phenomena like the one you described.

Maybe so, but your presentation doesn't make that distinction clear. Hans
presented a situation rather like a ball in an angel-food cake tin with a
sharp bottom. Any Test you want to apply would look pretty much the same
as would a Test on a "real" control system (especially since the "predicted"
nature of the movement is not very intuitive).

When there is control, resistance
to disturbance is happening _while_ the disturbance is affecting the
controlled variable!

Yeah. This also is true of the pendulum. Gravity "resists" the movement
away from the equilibrium point _all the time_.

To me, at least, the heavyweight distinction between equilibrium and control
systems is that a control system needs a power source _independent_ of the
disturbance source. A control system is a cooling device. It makes the
CEV thermodynamically cooler than it would be under the influence of the
disturbance. The "perceiving system" is what Maxwell's Demon needs to perform
the feat of entropy reduction. But, like Maxwell's Demon, that entropy
reduction is at a cost of entropy increase elsewhere, and is possible only
by using an external source of energy.

The particle under the influence of Brownian disturbances is at thermodynamic
equilibrium with its surroundings. A particle controlled against the effects
of molecular bombardment is not. It is cooler than it "should" be. And
_that_ is the crucial distinction.

Martin

[From Rick Marken (960212.2200)]

Me:

Control is a very clear and real phenomenon; it is only superficially
similar to stability phenomena like the one you [Hans] described.

Martin Taylor (960212 15:30) --

Maybe so, but your presentation doesn't make that distinction clear.
Hans presented a situation rather like a ball in an angel-food cake tin
with a sharp bottom. Any Test you want to apply would look pretty
much the same as would a Test on a "real" control system

The Test for the controlled variable is designed to distinguish control
systems from cause-effect systems. If the particles in Hans' Brownian
movement demonstration are cause-effect systems (which they are) then
this fact will be quickly and unambiguously revealed by The Test. Hans
suggested that particles act like they are controlling their average
position (in two space); it is easy to Test (and reject) the hypothesis
that this variable is under control.

When you say that "any Test you want to apply [to Hans Brownian motion
demo] would look pretty much the same as would a Test on a "real"
control system" you are saying that The Test doesn't work; that it
cannot (at least in some special cases) disciminate a cause-effect system
(like Hans' particles) from a control system (like Hans). As you must
know by now (having done all the PCT demos) this is crap; The Test
can always readily distinguish a relatively high gain control system
from a no gain cause-effect system.

Where do you come up this this stuff?

Me:

When there is control, resistance to disturbance is happening _while_
the disturbance is affecting the controlled variable!

Martin:

Yeah. This also is true of the pendulum. Gravity "resists" the
movement away from the equilibrium point _all the time_.

Does the resistance of gravity increase as you push harder on the bob,
forcing it away from vertical? Of course not. When you apply a force
disturbance to the bob it produces exactly the expected displacement.
The Test will immedialtely reveal to any non-practicing sophist that
the angle of the pendulum line with respect to the ground is not
controlled; gravity is not a control system; it only looks like one
to people who are desperate to see nothing fundamentally different about
the purposive behavior of living systems and the non-purposive behavior
of non-living systems.

To me, at least, the heavyweight distinction between equilibrium and
control systems is that a control system needs a power source
_independent_ of the disturbance source

This is a theoretical distinction -- and an important one. But why
would you suspect that a system has a "power source _independent_ of
the disturbance source"? Why, if you found what appears to be such a power
source, would you suspect that the system _uses_ it to actively resist
disturbances to a controlled variable?

As usual, you are putting the theoretical cart before the observational
horse. We know a system must have a "power source _independent_ of the
disturbance source" if The Test reveals that the system is a high
gain, negative feedback control system (the power source would be needed
to give the system the requisite gain). If you push on the pendulum bob
and it moves far less than would be expected given the applied force
(expected movement calulated from f=ma), then there must be something
actively opposing that force. In order to generate a force that actively
opposes another force, there must be a power source. But before you
start talking about power sources that are independent of "disturbances"
you must determine (by _observation_ using The Test) that there is
something to "disturb" and not just "influence"; you must deterine that
there is a variable is under control.

Best

Rick

[Martin Taylor 960213 12:30]

Rick Marken (960212.2200)

Let's not argue about which piece of the elephant we ought to look at. The
animal is all of a piece.

As you must
know by now (having done all the PCT demos) this is crap; The Test
can always readily distinguish a relatively high gain control system
from a no gain cause-effect system.

That wasn't Hans's problem, and neither was it mine in my posting, "as you
must know", having (presumably) read the postings in question.

"The Test" has several component parts.

It includes the notion that the force returning the disturbed
entity toward its initial position is caused by something that relates
to a sensed variable. One of the elements of "the Test" is to determine
whether a proposed sensing system exists and is used. But in the initial
stages of determining whether a particular system is in fact a control
system, an _inability_ to determine whether the possibly controlled state
is sensed cannot be taken as evidence that it is not sensed.

"The Test" also includes the notion that you may be able to deduce a
predicted effect of a known influence on the influenced variable. This is
in fact the case for a pendulum, and the effect that is observed is
just what is predicted. The pendulum fails this test for control, though
you are wrong in saying:

Does the resistance of gravity increase as you push harder on the bob,
forcing it away from vertical? Of course not.

It does, in fact.

When you apply a force
disturbance to the bob it produces exactly the expected displacement.

That also is true.

As I read Hans's Brownian motion thought experiment, he set up a situation
in which there was NO prediction about the influence of the disturbance.
Before Einstein, nobody could guess what was causing the movemment of the
particle, but after Einstein its parameters could be calculated. Using
Einstein's equations, Hans was able to predict, using no control ideas,
that there would be an equilibrium solution to the average position of the
particle. He then used this as an example to ask how someone who knew
nothing of Einstein would determine whether the system was a control system.

In Hans's system, the observer was assumed not to know the source of the
disturbance, and therefore could not calculate any expected degree of
movement. The observer could see that there was a disturbance, and that
wherever the particle was, the probability of its being disturbed in all
directions was the same. Nevertheless, the particle returned to its original
radial position. Likewise, in Hans's system, the observer could not tell
whether something inside the particle was perceiving the changes in position
(which in fact it was, altering the particle's bulk accordingly).

Since the observer could detect neither the source or magnitude of the
forces nor whether the particle perceived its position, the observer was
limited to the aspect of the Test relating to disturbance resistance.
And here all that could be seen was a more or less stiff resistance of
the particle to a disturbance of its radial position. In that, it could
not be distinguished from a low-gain control system.

Now let's see whether Hans's particle is even closer to being a control
system. Remember its mechanism for retaining its radial position against
disturbance: when it moves into a region of higher concentration it gets
bigger, and when it moves into a region of lower concentration it gets
smaller. When it is small, it is hit more erratically and moves faster
(in any direction), but when it is large, the hits are smoothed out and
it moves slower. It senses the chemical concentration, and does a very
close analogy of tensing--bulking up--a muscle. So, it would be quite easy
to see the system as a control system with a fixed reference level, the
controlled perception being of chemical concentration, the output gain
being represented by the change in bulk as a function of the change in
concentration.

To me, at least, the heavyweight distinction between equilibrium and
control systems is that a control system needs a power source
_independent_ of the disturbance source

This is a theoretical distinction -- and an important one. But why
would you suspect that a system has a "power source _independent_ of
the disturbance source"? Why, if you found what appears to be such a power
source, would you suspect that the system _uses_ it to actively resist
disturbances to a controlled variable?

If you thought that a system that acted against a disturbance was a control
system, you would look to see whether it has such an independent power source,
just as you would look to see whether it had a means of perception and a
means of generating output to counter the disturbance. You would look to
see whether the disturbance provided the energy that was used in returning
the entity to its undisturbed state. In the pendulum, it is precisely the
energy provided by the disturbance that allows the pendulum to return to
vertical after the disturbance ends. The pendulum is not a control system.

In Hans's situation, the disturbance provides all the energy used by the
particle in its movements. The particle is not a control system (though
I'm a little unhappy here, since there is a differential aspect associated
with the change of chemical concentration that alters how the energy is
extracted from the environment as a function of radial position).

As usual, you are putting the theoretical cart before the observational
horse. We know a system must have a "power source _independent_ of the
disturbance source" if The Test reveals that the system is a high
gain, negative feedback control system (the power source would be needed
to give the system the requisite gain). If you push on the pendulum bob
and it moves far less than would be expected given the applied force
(expected movement calulated from f=ma), then there must be something
actively opposing that force.

You need both the horse and the cart. Most discussions of the philosophy
of science put successful predictions from theory above successful
theoretical descriptions of observations, so I'm not sure which is
horse and which is cart--or even if that's an appropriate metaphor.

But I'm not clear where your "knowledge" comes from that a specific high-
gain negative feedback system (that has passed your application of the Test)
therefore needs a power source independent of the disturbance, if not from
fundamental theory. Unless for that system you can compute the forces
involved and thereby determine that there is an output from the system
that requires an external power source, how do you distinguish a control
system from a stiff spring or a v-shaped bowl? You _need_ those other
components of "The Test."

Incidentally, if, as you suggest, you used "f=ma" to compute the expected
movement of the pendulum, you would come to the definite conclusion
that "there must be something actively opposing that force", and according
to you there must be an external power source. This simple illustration
suggests how critical the correct _a prior_ model is, if you want to use
the "expected deviation" criterion for determining whether something is being
influenced by a control system.

As you must
know by now (having done all the PCT demos) this is crap; The Test
can always readily distinguish a relatively high gain control system
from a no gain cause-effect system.

That was never Hans's question, as you must have known when you wrote that.
He wanted to distinguish one kind of feedback system from a control system.
Or he wanted to show that a simple physico-chemical system can be said to
be a control system. Either way, neither Hans nor I are as simplistic as
you make us out to be.

Where do you come up this this stuff?

From trying to deal with postings _as they are written_ rather than ignoring

what is written so as to impose my _a priori_ views about what they must say.
It's not too hard. You should try it some time.

Martin

[Hans Blom, 960214b]

(Martin Taylor 960212 15:30)

To me, at least, the heavyweight distinction between equilibrium and
control systems is that a control system needs a power source
_independent_ of the disturbance source.

Is this necessarily so? Remember that physicists sometimes call
thermal energy "free energy", with the connotation that it can be
_used_, at least if something is clever enough to have invented a
method to do so. Plants have, in that they rely on the "free energy"
of light; some bacteria do something similar. Control theory, at
least the PCT version of it, usually disregards power sources, power
conversion and power consumption, being more concerned with analysis
of the informational aspects of "the loop". I consider that an
oversight. As you note, an autonomous control system needs energy.
Where does it come from?

A control system is a cooling device.

Replace "is" with "can be viewed as", and I'm with you. Similar
formula's apply, and similar theoretical considerations can describe
either control or cooling. But the same goes for control and
learning. Yet I would hesitate to say that learning IS control. I
would rather say that learning "might be viewed as" similar to
control. The identity of learning and cooling makes little sense
anymore (maybe ;-).

Greetings,

Hans

[Hans Blom, 960214b]

(Martin Taylor 960212 15:30)

To me, at least, the heavyweight distinction between equilibrium and
control systems is that a control system needs a power source
_independent_ of the disturbance source.

Is this necessarily so? Remember that physicists sometimes call
thermal energy "free energy", with the connotation that it can be
_used_, at least if something is clever enough to have invented a
method to do so. Plants have, in that they rely on the "free energy"
of light; some bacteria do something similar. Control theory, at
least the PCT version of it, usually disregards power sources, power
conversion and power consumption, being more concerned with analysis
of the informational aspects of "the loop". I consider that an
oversight. As you note, an autonomous control system needs energy.
Where does it come from?

A control system is a cooling device.

Replace "is" with "can be viewed as", and I'm with you. Similar
formula's apply, and similar theoretical considerations can describe
either control or cooling. But the same goes for control and
learning. Yet I would hesitate to say that learning IS control. I
would rather say that learning "might be viewed as" similar to
control. The identity of learning and cooling makes little sense
anymore (maybe ;-).

Greetings,

Hans