Feelings; TC; adaptive control texts

[From Bill Powers (940525.0915 MDT)]

Wolfgang Zocher --

The e-mail address for Marcos Rodrigues, mar@aber.ac.uk, has
always worked for me. Perhaps the system is down in Aberystwyth.
You can try sending your material to me for forwarding.

Tom Bourbon (940525.0824) --

Bill interpreted the question as a request for a report on his
feelings earlier, _at the time when he was acting to control_,
not later when his attempts to control were _pointed out_.

I guess I did. At the time Rick pointed out that I was controlling,
my only thought was "Sure, that's right." Is there anything we do
that isn't controlling for something?

I took Rick's question to pertain to how we feel about "controlling
people." I worded my reply to make it clear that I am not
controlling Mr. Thigh Cream, but only a variable that he is
disturbing. I will push back on Mr. Thigh Cream until I can be sure
he will not continue to disturb that variable. And I am hoping that
others who may be thinking of creating similar disturbances will be
paying attention, and draw the logical conclusions about the hassles
involved in using the internet for unwanted commercial advertising.

ยทยทยท

-------------------------------------
Mary suggested that the French (or Italian) approach might work on
the internet: make commercials available to those who want to see
them, in a way that does not disturb those who do not. When the
information highway is open for traffic, there could be "channels"
that offer services paid for by commercial advertisers. There are
already shopping channels on satellite TV; anyone who tunes in is
quite aware that the whole program will be nothing but commercials.
What the people who tune in get in return is the opportunity to buy
schlock at inflated prices, with a good part of the price contained
in "shipping and handling" charges. This is a perfect way to allow
suckers to select themselves and advertisers to pour on the
hyperbole and misdirection for a willing audience: a match made in
heaven, that bothers no one else.

The freedom of speech issue will no doubt be raised on the
information highway. But there is a complementary freedom: the
freedom not to listen to noise.
---------------------------------------------------------------
Hans Blom, Oded Maler, Martin Taylor (various posts) --

After a 2.5 month delay, I have finally received interlibrary loans
of two books: Kokotovic (Ed), _Lecture notes in control and
information sciences_, and Harris and Billings (Eds), _Self-tuning
and adaptive control_. As Sam Snead once said of Jack Nicklaus,
these people are playing a game with which I am not familiar.

I have to admit immediately that most of the mathematics is over my
head -- not always because of the mathematics, but often because of
the underlying assumptions about the kinds of processes being
controlled and the aims of the designer, which are barely mentioned
in these mostly mathematical texts. I am often totally lost, not
understanding what the object of all the mathematical manipulations
is. It would not surprise me, therefore, if what I say will be
viewed as a claim that the grapes are sour anyway, so who cares if I
don't understand? That might even be true, but all I can do is
comment out of what little I do understand about control.

There are two main features of PCT modeling that are missing in
these books: the reference signal and the disturbance, as they are
seen in PCT. There is, of course, a reference input in most of these
models, but it is not seen as a standard against which the "output
of the plant" (the perceptual input to the control system) is
judged. It is simply an input, from which the plant output is
generated.

Adaptive control is not, apparently, seen as a method for making the
controlled variable match the reference signal as closely as
possible, but as a way of maximizing or minimizing some extraneous
function of the controlled process's state and the output variable
-- some sort of "cost function." It is explicitly stated more than
once that a central problem of optimization is that of designing the
cost function, for the criterion of optimal control has to be
specified by the _user_ of the system. The simple criterion of
making the controlled variable match the reference signal as closely
as possible seems not to be the main idea.

In designing industrial control systems, this is probably a
(somewhat) realistic way of optimizing, but it is not realistic as a
way of understanding the organization of an autonomous system.
Actually, in industrial systems the improvements of control (over
that of the "classical control system") achievable by this method
are quite modest, ranging from a few shining examples of 6-to-1
improvements, down to a more common 10% improvement, and going as
low as a 1% improvement (in the speed of a larger tanker ship). Such
modest gains may impress a bookkeeper, and provide a commercial
justification for the labor involved, but to me they seem hardly
better than what a seat-of-the-pants engineer like me could achieve
just by fiddling around with the system.

The disturbance as I think of it is also missing. Instead, we have
noise inputs that affect the process randomly. These noise inputs
are dealt with stochastically, not as variables that have specific
values and rates of change at each moment. Nowhere in the
mathematics can we find a simple result such as a statement that the
effector output will systematically oppose the effects of a
disturbance on the controlled variable. The optimization criterion
seems to be based on expected future values of variables rather than
on present-time minimization of an error signal. The whole treatment
of what is to be optimized and why is, to me, surprisingly vague.

I have a strong feeling that "control of output" is still an
influential idea in these texts. But it's really futile for me to
try to make any case here in terms that the authors of this approach
would consider worth listening to. There is no way I can deal with a
hundred pages of uninterrupted abstract mathematics on the subject
of "robust control." There is so little mention of any real control
problems that I find nothing to convince me of the relevance of all
the theorems and lemmas. It seems to me that the basic concepts of
control are so poorly defined that the mathematics is simply an
exercise done for its own sake. I see a mountain laboring to produce
a mouse.

All that is probably just a roundabout way of making my ignorance
crystal clear. But there's no help for it. I will leave this subject
for others better equipped to handle, and will concentrate on trying
to devise simple control models of simple systems -- living systems,
not steering systems for ships or control systems for distillation
columns. Judging from these texts, I have chosen the simpler
problem.
---------------------------------------------------------------
Best to all,

Bill P.