[From Bill Powers (2002.09.25.0100 MDT)]
Rick Marken (2002.09.24.1950)--
> This equation holds only when the reference signal is constant.
Right. The complete equation is, approximately:
qo ~ r - 1/f (d)
where r and d have _independent_ effects on qo. So variations in r have no
effect
on the causal relationship between qo and d, a fact that is demonstrated
in my
"Control of Behavior" demo at
http://home.earthlink.net/~rmarken/demos.html. So it
is still quite correct to say that disturbances to controlled variables are an
independent cause of behavioral variations, qo.
> >What control theory glasses let us see is that the causal connection
between
> >external circumstance and behavior is through the _environment_ (the
feedback
> >function, f), not the organism.
>
> The behavioral "reaction" to a disturbance depends on where the reference
> signal is set.
Not true. The effects of disturbance and reference on qo are _independent_.
I see your point: in terms of changes, the change in action opposes the
change in disturbance independently of the change in reference signal.
However a positive change in action can show up as a decrease in an action
in the opposite direction. Suppose you're steering a car around a left bend
in the road. Your action is to apply a force twisting the steering wheel
to the left. Now if a crosswind arises that is blowing toward the left, the
car would turn too fast, and you would have to _reduce_ your steering
effort to keep the car on the road. If the wind blew just right, you could
let go of the steering wheel (for a second or two).
Of course if you were on Bonneville Salt Flats, you might be intending to
turn either to the right or to the left. In that case, whether the
crosswind blowing to the left resulted in a _decrease_ or an _increase_ in
your steering effort would depend on whether you wanted the car to be
turning to the left or to the right. While the algebraic change in steering
effort would be the same, the absolute magnitude of the effort could either
increase or decrease, depending on the reference condition.
This sort of situation arises whenever the action that brings a controlled
variable to its reference condition must be sustained in order to maintain
the state of the controlled variable. The car, when caused to deviate from
straight-line motion, is subject to a centrifugal force caused by the
deviation, and the driver has to maintain a steering force just to overcome
it. This is not an independent disturbance, since it is a function of the
steering effort. In fact it's better just to think of the force as what is
required to maintain the controlled variable in a given condition -- part
of the definition of the external feedback function.
This can also apply in situations where physical dynamics are irrelevant.
Suppose we have a joint bank account, and the bank is applying service
charges every month because the balance is too low. If I am paying those
charges, you could say I am exerting a certain amount of effort to maintain
the account at a certain level. If you start paying those charges, I will
decrease my payments to keep the level the same. The change in my action
has an effect that is opposed to your disturbance, but in fact my action
decreases when you apply the disturbance. Of course if I wanted the bank
account to decrease, I might switch to making withdrawals to counteract
your payments of the service charges. Silly example.
Let's try for a more directly relevant example, one in which my reaction to
your comments will depend on my reference levels for various controlled
variables. Let's say I'm trying to impress someone with my knowledge of
control theory, so I am putting out a lot of effort to answer questions,
correct misinformation, and so forth. I want correct information to be
given, but I also want to be the one who gives it. You tell me, "That's not
the right explanation, you've made a mistake." Of course this goes against
one of the things I'm trying to do, present myself as an authority, so I
will increase my effort to give correct answers and set the record
straight, and I will probably also argue with you that I was correct all
the time, even if I end up changing my explanation in the way you pointed out.
But suppose I reduce the setting of the reference level for being an
authority, so I am not aiming to give such an impression of correctness and
authority and so on. Suppose I reduce that particular reference level to
zero. Now what will happen when you come up with a criticism of something
I've said? Now my desire to see a correct answer given will be satisfied by
your correction, and I will step aside and let you explain in my place. It
will not matter to me which of us gives the explanation as long as it is
correct. In fact, if you now try to defer to me and say I'm really the
authority and so on, I will resist your suggestions because my reference
level for being the authority is now zero. So I still resist your
disturbance, but now by decreasing my effort to be the authority and in
fact exerting a _negative_ effort. You're helping me toward my old goal of
being the authority, but now I actively resist because my goal has changed.
I've probably brought in considerations that make this not quite a parallel
to the original case, but you see what I'm getting at.
Setting the authority reference signal to zero does not mean I stop
controlling that variable. I will resist if you try to make me seem either
less of an authority (an anti-authority trying to give false information)
or more of an authorithy. But I can, alternatively, cease to consider being
an authority to be of any importance one way or the other. I could say,"If
I know more about a specific subject than anyone else, for example what I
ate for breakfast, then that makes me the authority on that matter. On the
other hand, if someone else knows more than I do -- say they have a
videotape of what I actually ate -- then they become the authority. So
what?" In that case I will not oppose anyone's effort to increase or
decrease my standing as an authority -- it won't matter to me what anyone
thinks on that score. My loop gain for that variable will be zero.
> As a result, we are free to choose not to react by reducing the
importance of
> an error to zero (I no longer care what you do)
Well, free within the context of one's other goals. You can choose not to
react by
functionally relinquishing control of a perception. But control of that
perception
may matter to other systems inside you. So your choice to not react may
result in
other systems trying to unchoose your choice, ie. internal conflict.
True. Exactly. If you're trying to rid yourself of reactions that have
become unwanted, you have to explore those other reasons for continuing to
have them. A young man trying to get a PhD has every reason to want at
least certain people to recognize his authoritative knowledge about
_something_. And those reasons can persist long after they've become
irrelevant.
> A strict relationship between disturbance and action occurs only when the
> reference signal and the loop gain are being held constant.
I believe this is true for loop gain but not for the reference signal.
I think I've introduced a second factor, which is absolute magnitude of
effort as opposed to sign of effort. To sum up, in algebraic terms, the
change in effort remains opposed to the change in size of the disturbance
regardless of the setting of the reference signalm, as you say (I'm
ignoring quibbles about nonlinearities). However, the absolute magnitude of
the effort may either increase or decrease in response to a given
disturbance, depending on where the reference signal is set.
I am very appreciative of Bill Williams' efforts to find and stick to the
optimum path through these thickets. I am sure he is reviewing his remarks
on various subjects as the rest of us are trying to do.
Best,
Bill P.