# Controlling means and ends

[From Rick Marken (940903.2300)]

Bruce Buchanan (940903) --

I said:

I'm controlling for good health care for everyone at the lowest
possible cost. I am not controlling for the means used to achieve it.

Bruce replies:

The trouble with most general statements of this kind is that they are
sufficiently imprecise that no one can resonably disagree.

Let me try to be more precise. When one controls a variable (say
"quality of health care") the actions (means) used to keep that variable
in its reference state must vary in order to compensate for changes in
the specification of the reference state for that variable, changes in the
effects of disturbances to that variable and changes in the effects that
the actions themselves have on that variable. If the actions themselves
are under control, it will be impossible to vary these actions, as
necessary, to keep the controlled variable (quality of health care) under
control. This statement is "general" in the sense that it is _always_ true
in all control situations. It is true whether the perception being
controlled is the position of a cursor on a computer screen, the quality
of one's relationship with one's spouse or the quality of health care in
the US. It is embodied in one of the two basic "laws" of perceptual
control theory:

o = p* - d

which says that, in order to keep a perceptual variable, p, matching a
reference variable, p*, the actions (o) of the system doing the
controlling must be directly proportional to variations in the reference
variable, p*, and negatively related to variations in the net disturbance,
d, to the controlled variable. If o (the means used to control p) is
itself under control, then it will very likely be controlled at a value
that is not equal to (p*- d). Since the only way to control p is to
keep o nearly equal to (p*- d), control of o would prevent control of p.

This is all just a fancy way of saying that, if you have a preferred way of
holding the handle in a tracking task, you won't be able to keep the
cursor on the target; you can't control the cursor and the means used to
control it (handle position) at the same time.

In addition, while it is hardly conceivable that the ends do not control
for at least some possible means

I take this to mean that p* (the end to which p, the controlled variable,
is constantly being brought) influences (not controls, of course) the
actions involved in the control of p, and, indeed, this is true. Actions
are determined by both p* and d, though d is by far the main influence
on actions when p* is constant.

The point of such questions here is to illustrate the need to recognize
that vaguely stated general principles are just that.

I hope I have stated the general principle less vaguely. Again, the
general principle is that the actions (o) that are involved in the control
of a perceptual variable (p) must be free to vary (they must be
uncontrolled) in order to keep p (the controlled variable) in a (possibly
varying) reference state, p*, against disturbance, d.

Best

Rick

[Rick Marken (940903.2300)]

Rick wrote:

I'm controlling for good health care for everyone at the lowest
possible cost. I am not controlling for the means used to achieve it.

Bruce replies:

The trouble with most general statements of this kind is that they are
sufficiently imprecise that no one can reasonably disagree. . . .(etc.)

To which Rick responds:

Let me try to be more precise. When one controls a variable (say
"quality of health care") the actions (means) used to keep that variable
in its reference state must vary in order to compensate for changes in
the specification of the reference state . . .[some material deleted] . . .
This statement is "general" in the sense that it is _always_ true
in all control situations. . . .

Perhaps I should have been clearer. The imprecision and generality on
which I commented did not have to do with the principles of PCT.

What I intended to address was the immense difficulty of formulating the
controlling variables - ideas, agencies - that operate at higher levels of
abstraction with reference to many of the real world problems that most
interest us. Even then it is often the case that nothing much different can
be done in a practical way. However, our hope might be that, with more
accurate understanding of the real nature of the actual problems, some new
strategic insights may be found.

What I think needs emphasis is that it is no advance to simply relabel
familiar general concepts with the terminology of PCT. Much more basic
analysis in relation to levels of control and meaning is required, since
the real problems relate to actual situations and controlling agencies -
constituencies, governments and otherwise (hence the importance of sound,
accepted data).

Resistance to disturbances is an almost universal phenomenon which has been
described in many ways. My point is that, to be more useful, possibly new
descriptions of specific societal conflict situations and policy problems
may require not only a change in terminology but more precise insights and
understanding of the variables and relationships involved.

I am not suggesting any answers at this point, but rather that the real
nature or operational characteristics of the problems in situ must be
recognized. Only then can anything approaching useful discussion and
modelling begin.

Or so it seems to me, at any rate! (This is just a clarification, for I

Cheers!

Bruce B.

[Martin Taylor 940904 17:20]

Rick Marken (940903.2300)

Again, the
general principle is that the actions (o) that are involved in the control
of a perceptual variable (p) must be free to vary (they must be
uncontrolled) in order to keep p (the controlled variable) in a (possibly
varying) reference state, p*, against disturbance, d.

Actually, this is true only under the special circumstance that there is only
one means to the desired end. In a control hierarchy, it is normally the
case that several lower-level perceptions contribute to any single higher
level one. In such a case, all that needs to be assured is that the
number of degrees of freedom controlled at lower levels is at least one
less than the number available. (By "controlled at lower levels," I mean,
of course, that the reference signals at those levels are set independently
of the higher-level control system under consideration).

You can control for being at a certain place at a certain time AND for