[Martin Taylor 2018.03.04.15.24]
I don't understand this comment. If you don't know that a pendulum
is not a control system, what do you know about what constitutes a
control system? For the pendulum, you start out knowing that your
disturbance will be resisted by a gravitational force that varies as
the sin of the pendulum angle from the vertical. Then you do your
test, and you find out that what you already knew is in fact true,
but only then do you “discover” that the pendulum is not a control
system. I doesn’t make sense to me. Maybe what you are saying is
that you don’t at first know that it is a pendulum, though it looks
like a pendulum and moves like a pendulum. I’m not clear why you
wouldn’t know that it is a pendulum, but I suppose I could grant the
possibility that what you are saying is “I don’t know whether this
thing that looks like a pendulum is indeed a pendulum, and that’s
what my test is going to be finding out”. In that form, it makes
more sense, though it is still a bit bizarre.
To some degree, that is indeed so. Changes in the value of the
variable in question will be uncorrelated with the variations in the
disturbance if the disturbance has no influence on the variable. But
to require you to know the function relating the value of the
disturbance to the value of the change in the variable if it is
uncontrolled smacks to me of the modelling approach to control,
requiring unnecessary modelling of the environment.
What you do need to know (or discover) is simply that your test
disturbance actually can influence the value of the variable, and in
what direction. You don’t need to know the function that describes
how much in order to find out whether control is involved.
No control can be perfect. If it was in some instance, there's no
way you could distinguish control from an environmental stolidity. I
push upon a rock. It doesn’t move at all. Do I conclude that it is
being controlled perfectly by something that really, really, wants
the rock to be where it is? Because control is not perfect, if the
variable is controlled or not, you will be able to tell whether your
test disturbance influences it, especially if you try a rapid change
to start with, because the variable will initially move under the
influence of the disturbance alone at least for the duration of the
transport lag in the control loop.
Â
You have to know these things if you want to calculate control
quality. You don’t have to know them if you just want to see whether
the variable is controlled.
Yep.
Not necessary, though to know how and how much the disturbance would
affect the variable if it were not controlled might be interesting
for other reasons (such as I noted above, determination of the
control quality).
How do you do that magic, to know how the disturbance would
influence the variable without knowing the environment that
determines how the disturbance would influence the variable?
True, but it doesn't change the problem. You still have to know the
relevant environment.
The environment in this case is the program that determines how the
physical mouse position translates into a number to be added to a
number determined by the program, and then how that sum is
represented on screen. If you don’t know these things, you can’t
compute how much the cursor would change with the disturbance. But
you could still determine the correlation between physical mouse
movements and cursor on-screen deviations from stability, once you
notice that moving the mouse abruptly results in a correspondingly
abrupt movement of the cursor.
Same argument applies.
The one does not preclude the other. The fact that it is a ***good***
explanation of the controlling done by living organisms is what
allows its use as a tool for many purposes relating to what living
organisms are observed to do. I seem to remember a long time ago you
reported that you thought that understanding PCT helped you to
become a better person (or maybe it was husband). That’s using PCT
as a tool. Anyway, even if I misremember, it describes my
experience.
Both true. It was true in my case, even having previously published
papers that unknowingly used that fact (Layered Protocol Theory, and
one in 1973 that used hierarchic perceptual control to explain why
active (haptic) touch provides perceptions of objects whereas
passive touch produces perceptions of variations in touch
sensation). But my prior flirting with the idea did probably help me
to accept PCT wholesale very quickly once my attention was pointed
to it by someone on the System Dynamics mailing list. I think others
with a less conformable background may well find it more difficult
to get their heads around PCT and its wide ramifications.
I don't believe this ordering is correct. It's much easier to see
the fact of control once one understands what it means in PCT, even
in a single control loop. Lots of people do think that other people
control, and most of them think that to do so is a bad character
trait. If you understand the PCT version of control, you won’t think
control is a bad character trait. It’s just what being alive means.
And then you can go on to see the wider implications of hierarchic
control, and control more generally in interpersonal interactions,
and in the behaviour of trees and flies.
[Aside] Do you have an explanation for why this last is in a smaller
font and a different colour on my screen? I’ve seen this effect in
various messages, but I don’t know whether it is at source or
something my e-mail client (Thunderbird on Mac) does.
RM: And I control (also with very high gain)
for the correct application of PCT to behavior, especially by
people who purport to be experts in PCT. And that perception has
deviated from its reference for a long time due to your failure
(and that of way too many others on CSGNet) to see the power law
as a perfect example of a behavioral illusion. So the disturbance
to my perception is bigger than yours;-)Â
In order to try to cool the interpersonal situation, I'm not going
to comment on the rest of Rick’s message, except to point out that I
don’t think anyone on CSGnet argued whether the curvature effect is
or is not a behavioural illusion. Maybe it is, maybe it isn’t, but
that’s never been the issue. The issue has been the use of faulty
logic in the interpretation of experimental results.
The logic in question is parallel to this: *
I see a house with a red front step, therefore all houses have red
front steps.*
In the paper, the logic is: *
We saw an expression that was used to compute a velocity that
turned out to be V; therefore the same velocity V can be
substituted for the expression* * in any formula in which it
occurs*.
I know it sounds ridiculous, but it resulted in two publications for
Rick – the original paper and the forthcoming rebuttal. That’s the
only thing this year of trouble and ill-feeling has been about. It
has never had anything to do with PCT or with the behavioural
illusion.
Here's a small excerpt from my published comment:
···
[Rick Marken 2018-03-04_12:02:38]
[Martin Taylor 2018.03.02.23.45]
MT: You forgot the step that you need before your
step 1: Determine that you are looking at a pendulum,
(my step 1). Your steps 2,3,4,and 5 are my steps 2 and
4.
RM: We knew we were looking at a pendulum. This
discussion started with the question of how to
discriminate an equilibrium system (like the pendulum)
from a control system. So I described how the test could
be used to determine whether or not the pendulum is a
control system.
RM: I think this is a very poor trope on
what I actually said. But here’s another
description of how to test to determine
whether or not the bob of a pendulum is
under control.Â
MT: Yes, it is a good paraphrase, and I
apologise.
RM: Accepted.
Â
MT: On p 76,
however, Runkel describes a better “Test”, one that
requires no pre-determination of what the variable would
do if uncontrolled. You also have in the past pointed
out the value of that test, and which you use to
determine the controlled variable in demos: The
variation of a possibly controlled variable correlates
highly with the disturbing influence if it is not
controlled, but shows almost no correlation with the
disturbing influence if the variable is controlled. You
can use that test any time you can be reasonably well
guaranteed that the reference value will be reasonably
stable.
RM: Actually, you still have to determine what the
hypothesized controlled variable would do – how it would
be affected by disturbances – if it is not controlled.
RM This is a brief paraphrase of the
steps involved in doing the test for the
controlled variable as described in Phil
Runkel’s * People as Living Control
Systems* (p. 77).\
In a compensatory tracking task, for example, you have
to know what effect the computer generated disturbance
would have on cursor position if cursor position were not
being controlled. In the pendulum test, you have to know
what effect forces on the position of the bob would
be if the position of the bob were not being controlled.
This means that
you  can't pick just anything out of the air to be
the disturbance you use to test to determine whether a
variable is controlled. You have to know that the variable
would affect the hypothetical controlled variable if it
were not controlled
– and how it would affect it
…
Â
MT: The correlation
test does not require that you know the physical
environment of the variable before you start, and you do
have to know that in order to see whether the influence
of the disturbance is less than you would otherwise
expect.
RM: You don't really have to know the physical
environment of a hypothetical controlled variable before
you start the test. All you have to know is how a
disturbance variable would affect the hypothetical
controlled variable if it were not under control.
And the effect of the disturbance on a hypothetical
controlled variable is not necessarily physical.
... But in compensatory tracking, the effect of the
computer generated disturbance on the position of the
cursor if the cursor is not controlled is known from the
laws of mathematics: cursor position = mouse position +
disturbance value.
And the effect of the computer generated change in the
sequence or program that is occurring if the sequence or
program is not being controlled in my demo of sequence and
program control is known from the laws of logic.
MT: What newcomers
to PCT need to do is learn how to use for control the
tools with which they are supplied already, and to learn
to use whatever new tools are particularly suited to
understanding control. What is necessary to drop is the
idea that any one approach is the only approach, and
what is necessary to learn is that once you understand perceptual
control, so many other things become clear that
previously seemed to belong to different domains
entirely. Perceptual Control Theory, like any
foundational theory, is itself a tool, a powerful tool
with many uses.
RM: I disagree with this. PCT is not a tool; it is an
explanation of the controlling done by living organisms.
What newcomers to PCT need to do first is understand
the phenomenon that PCT explains; it is virtually certain
that no newcomer comes to PCT aware of the fact that the
behavior of living systems is a process of control.
Once a newcomer understands the phenomenon that PCT
explains he or she is ready to learn how the theory
explains this phenomenon.
Certainly some of the previous tools a person has
learned can help them understand PCT. And some tools are
probably essential, such as basic math (at least to the
level of algebra), research methodology and computer
programming.