[From Bill Powers (930822.1000 MDT)]
Michael Fehling (930822) --
... it appears that I may be more reluctant than you to declare
a schism between those intersted in biological control and
those concerned with the control of (other) mechanisms.
I'm not declaring a schism, I hope. Biologists and control
engineers who understand control theory understand all the basics
they need to know to grasp PCT. But there is an underlying
approach that gets in the way, for those who stick too closely to
the customary engineering ways of analyzing control systems, as
most people outside PCT do. It's partly that in PCT we draw the
diagram differently, which upsets people used to the conventional
way of drawing it. All the functions and signals are the same,
but we arrange everything so you can't confuse a reference signal
with a sensory input (in engineering diagrams, the reference
signal is very often labelled "input", which causes no end of
misunderstanding among non-engineers who think input means
sensory input).
The biggest problem is rather silly when you come down to it. We
say that control systems control their own inputs, and more
particularly their own perceptions which are all they can know of
the external variables at their physical inputs. In substance
this is absolutely identical to the way control engineers design
their systems, except that control engineers call the variable
that is sensed and controlled the "output." They don't really
mean output, of course; they mean some consequence of the actual
effector output. They mean shaft angle or load position, not
motor torque. We simply reserve "output" to mean the first
physical variable directly affected by an outgoing signal in the
control system, via an output transducer. An "input" variable is
one that is directly monitored by the senses of the control
system. This separation of output from controlled input has
profound meanings in applications of control theory to living
systems. One thing it means is that the "behavior" we see people
producing is often linked only very loosely to what they are
doing -- what they are controlling.
Any engineer who studies our diagrams and sees why we distinguish
output from controlled variable should have no trouble at all
making the necessary translations. Some, indeed, have done so and
think the distinction makes sense. But others have found this
shift in terminology too much too handle. One very nice fellow
who tuned in to the net finally tuned out again, saying he had
always called the controlled variable an output and preferred to
go on doing so. He never got past that problem to see what PCT is
all about.
The biggest problem, which I have been thinking about a great
deal lately, is that control engineers and those who have
followed their lead really don't understand control of the kind
that an organism carries out. Control engineering textbooks spend
very little time developing an organized approach to control;
they dive right into the mathematics of stability, and go on from
there. As a result, the system designs don't have any obvious
organizing principles behind them; the engineers tackle a control
problem by finding some arrangement that looks workable, and then
tack on modifications until it works.
I've been looking at a 1991 journal in which there's an article
on a control system for using two helicopters to lift a load. The
block diagram of this system is derived from mathematical
expressions for a six-degree-of-freedom control problem. With a
lot of guessing, I've concluded that the only way the two
helicopters are kept from colliding is that the engineer
designing the system built in a required angle of tilt of each
helicopter away from the vertical passing through the suspended
load.
Clearly, if this angle of tilt is properly maintained (and
there's a simulation plot showing that it is), the helicopters
will not collide. But who decided that this angle was needed, and
who adjusted the parameters to assure it was maintained? Not the
control system, but the designing engineer. There is nothing in
the control system itself that cares a whit about collisions. As
far as I can see, the distance between the helicopters is not
even sensed, and I can find no control loop that compares the
sensed separation with a desired separation and adjusts the angle
of tilt to maintain the desired separation. The design engineer
knows that if a certain angle of tilt is maintained, the
separation will be maintained and there will be no collision --
but the control system doesn't know that.
I think this is an example of a very general approach to the
design of artificial control systems. It works, of course, but it
works only because there is an engineer who can perceive the
situation in ways that the control systems themselves can't
perceive it, and who adjusts the design so as to control things
that the control system itself can't control. There is a subtle
control loop that never appears in the final system, one that
passes through the engineers' senses, into their brains, and out
again to the drawing board and the simulators that show the
engineer what the design does. The engineer is part of the
control system, and is essential for making it work.
This unconscious participation of the engineers in the operation
of control systems leads to great confusion about what control
means in living control systems. An engineer can design a
compensating system, for example, in which disturbances that act
on the controlled variable are sensed, and a compensating output
is blindly and automatically applied to cancel the direct effect
of the disturbance. This kind of system has no feedback in it,
yet engineers commonly think of this as a form of control system.
What they forget is that in order to make a system like this
work, some engineer has to monitor the controlled variable and
adjust the compensation parameters until the controlled variable
remains close to the desired state for all expected disturbances.
If any property of the environment changes by even a small
amount, the engineer has to get back inside the system with his
screwdriver and, watching the controlled variable, adjust for
minimum effect of the disturbance.
The system itself, which does not sense the controlled variable,
could not by itself act to bring the controlled variable to any
particular state, or keep it there. It could not do that unless
the engineer, who does know of the state of the controlled
variable and does have in mind a particular desired state for it,
adjusted the compensation to be just right. So the engineer is
really using this compensating system as a complex output
function in a control loop of his or her own. When we remember to
include the engineer, we have a negative feedback control system
organized like any other true control system.
In a living organism this external designer loop doesn't exist.
There is no helpful engineer monitoring the consequences of the
organism's action and adjusting open-loop responses to create
some particular outcome that is stable against disturbances. If
any consequence of motor outputs is to be made repeatable and
stable against disturbances, the organism itself has to have the
ability to do this, with no external help.
That's the challenge in reverse-engineering living control
systems. We may see what has to be made controllable in the
organism's environment for it to prosper, but the fact that WE
can see it doesn't help the organism. A plausible model of an
organism has to explain how the organism itself comes to know
what is controllable, and what state of the controlled variable
should be maintained -- all by itself.
If I were designing that twin-helicopter lift control system the
way organisms are designed, the first thing I would do would be
to supply some sort of range-finder that would sense the
separation of the helicopters, and build a control loop that
would tilt them away from each other if they got too close. Then
I would give them a way of sensing their relative altitude and
orientation, and a control system that would make each one try to
stay level with and parallel to the other. Then I'd built a
superordinate system that would adjust the reference signals for
the lower-level systems so the pair of helicopters could be
controlled as a single unit to do the lifting. In other words,
everything I want to happen would be turned into something that
the control systems themselves want to happen. I would keep
myself and my own knowledge out of the final operating system as
far as possible.
An engineer really doesn't have any motive for approaching a
control problem this way. The customary way seems to work -- why
change it? For all I know, the customary way produces economical
and efficient designs. But an engineer trying to analyze a living
control system might take the same approach, failing to realize
how much of the engineer is being put into the model of the
living system, unawaredly.
When I speak of the "externalized" view of a living control
system, this is what I mean. Aldus speaks of "tasks" that an
organism carries out. But these tasks are defined from the
external engineer's point of view, the view of a planner or
supervisor who directs other organisms and is concerned with
external effects of their actions, the viewpoint of a designer
trying to get a system to do something. The organism -- say, a
dog -- doesn't know that in circling around in one spot in a
field it is performing the task of making its bed. It's simply
controlling perceptions in a way that seems satisfying. When you
eat, you don't perform the task of getting food into your mouth
by using a spoon. You just control the spoon, and with it, the
food, putting it where you want it. The organism acts to make its
experiences be what it wants them to be. Only an onlooker would
break down what it's doing into "tasks."
ยทยทยท
--------------------------------
I wonder if you and I are really all that far apart on the
value of building artifacts that model interesting notions of
control.
We aren't far apart at all. In the CSG many of us make or use
computer simulations of control systems as models of behavior.
The working model shows you what your system diagram really
means: it runs according to the properties you gave it, not
according to what you thought or claimed it would do. If you have
IBM-compatible micros, I can send you some demo programs.
So, can you tell me whether you or anyone else in the PCT
community has applied this theory to group/organizational/
social systems?
We have had for many years an active contingent of sociologists
exploring the applications of PCT, and several people interested
in organizational problems. Others are applying it in related
fields such as education and linguistics. Perhaps Greg Williams
could work up a brief reading list for you (Greg?), or some of
the interested people could send you copies of papers directly.
One program I can send you simulates a crowd of people seeking
various goals; it was developed for Clark McPhail of the U of IL
sociology department.
---------------------------------------------------------------
Hal Pepinsky (930821) --
I propose that a state of peace is perceived not by anything
one does oneself alone, as by reaching a target, but has twin
referents--to my motives (which goals I am seeking) and the
motives of any other whose behavior is affecting me
(individually or as a group). In each case I infer greater
security (you would call it contentment Bill) when both change
at once, and each is chosen from an array newly suggested by
the goals each perceives from the other at the preceding
moment.
In the hierarchical pct model (HPCT), lower-level perceptions are
varied (through adjusting their respective goal-states) as a
means of achieving higher-level goals. What's going on in the
world determines lower-level goal selection, because that changes
what's required in order to achieve or maintain the higher-level
goal. This is true at all levels: there's no one action that will
cause a goal to be reached. Instead, the actions have to be
adjusted continually on the basis of what the world is
independently doing.
A "state of peace" is a high-level goal (I don't like to stick
slavishly to my definitions of levels -- they're just
suggestions). To maintain this state, it's necessary to have
flexible goals for all the things that contribute to your sense
of peace.
As to the motives of others, we can infer them from the way
others act in relation to us. In formal PCT, there's a "test for
the controlled variable," a method for testing hypotheses about
what another organism wants. People use informal versions of it
all the time. One way to think of it is as discovering conflicts
between what I do and what you want. Once I know what you're
controlling for, I can alter my actions to minimize strain
between us -- assuming that what I want between us is peace and
not a contest.
Singing harmony is a nice example of how we alter a lower-level
goal (pitch) as a means of controlling for a higher-level
perception (harmony). It's especially interesting because in
singing harmony, one can control only one element of the harmony;
the other is under someone else's control. Yet we can easily
control the _relationship_ between two pitches. At a higher level
still, we can control for chord progressions, tensions and
resolutions, by adjusting the goals for the particular harmonies
we're controlling. At least three levels are involved, and
probably more (obviously more if you go down levels). As you
comment, you don't really have to know what the other person
wants; you're just maintaining a relationship of your voice to
the other's voice. But the other can send messages about what
where the melody will go next, as by drawing out a penultimate
note to indicate that this is now the end of the last chorus.
I mean (and you actually say this yourself repeatedly Bill,
whether or not you consider it basic to your theory) letting
oneself discover what one wants in the next moment after one
gets there ...
This is what it feels like, I think, to be unaware of one's own
higher levels of control. When awareness is focussed on a
particular level, the goals seem just to come by themselves. By
re-establishing contact with higher levels, one can learn why the
goals changed as they did: the changes which formerly seemed
arbitrary or spontaneous now make sense in terms of maintaining
some higher-level perception near its goal-state. There's a sort
of enlightenment experience (with a small e) that comes from
exploring levels deliberately. When you get to the top (or anyway
as far as you can go) there don't seem to be any more goals, or
if there are any they're not of any familiar kind, like "wanting
peace." You're just observing. You get there by asking why you
want something (that's one way), and asking again and again until
the answer is that you don't want anything at the moment.
Nonviolence is anything but passive, and in fact demands
confrontation of [or?] conflict rather than trying to channel
it toward some target (victory in war, the criminal in prison).
You'll have to explain that. Sounds like violent nonviolence to
me.
Goal directedness becomes more violent, more productive of
revolt as you put it, Bill, as it becomes more highly
organized, as into one group of 250 million people operating on
one multi-trillion dollar tax base.
You're making generalizations out of specific instances here.
Highly organized control is required to play a violin. It isn't
the organized-ness of goal-seeking that leads to violence, but
incompatibility of goals. Anyway this is getting pretty far from
the grounds where I even think I know what's going on.
-------------------------------------------------------------
Best to all,
Bill P.