[From Bill Powers (931215.0900 MST)]
Osmo Eerola (931215.0930) --
As you said, there are probably some difficulties here with the
meanings of words. Whether we have any disagreements or not
depends on what you mean, and what we mean.
Control theory itself is just control theory: applying dynamical
analysis to closed-loop negative feedback systems. That part is
easy. But PCT is about applying this sort of analysis to the
human system to explain human behavior, which means that we have
to identify the parts of the real system in order to know how to
apply control theory.
This problem is made somewhat difficult to communicate to
engineers because control engineers have developed their ways of
thinking about control without considering human systems. It is
not necessary for a control engineer to design a system that
makes sense in terms of human behavior. The engineer can see what
needs to be done, and can design a system to accomplish it; the
system doesn't even need to be a negative feedback system. The
sensors used for control, if any, need not directly sense what
the engineer wants to control. All that matters is that during a
control process, the variable of interest to the engineer be
maintained in the desired state -- desired by the engineer. Most
of the actual control process might be taking place in the
engineer or the user of the system, but that doesn't matter if
the job gets done.
In PCT, we are trying to analyze a system that already exists,
that nobody designed, and that must act like a human being with
no external help. This system does not control in order to
satisfy some external customer, but to satisfy itself. It does
not know anything about the environment except what its sensors
tell it, and what it can deduce from the behavior of its own
sensory signals. Like the engineered system, it has sensors that
report to it the current state of some, but not all, aspects of
the external world. Unlike the engineered system, it has no other
kind of input in addition to sensory inputs; there is no special
input through which an external agency can adjust the reference
signals of the control system. All reference signals are adjusted
by the system itself, from inside, to suit its own purposes. In
fact, the reference signals ARE its purposes.
It may seem to you that you are adjusting reference signals for
another person when you issue orders to a recruit. But that is an
illusion. You have no direct control over anyone else's reference
signals; it's physically impossible. All you can do is emit
sound-waves that enter the recruit's ears to create perceptions.
The recruit must then perceive those sound-waves in the form of
neural signals, and interpret them as a description of something
you want to see the recruit doing. I would make a terrible
recruit, because I don't understand Finnish; the process would
fail right there. My brain couldn't separate the continuous
sounds you make into discrete words, and even if I could, those
words would not indicate any meanings to me -- any perceptions
except the words themselves.
A Finnish recruit can convert the sounds you make into words, and
the words into the perceptions that they are supposed to mean. So
if you are telling the recruit to pull back the bolt on the
rifle, the recruit understands that "bolt" (in Finnish) means a
certain piece of the rifle with a certain appearance, tactile
feel, and sound when it is moved. So the recruit now knows what
perceptions your words are indicating.
However, that does not make the recruit do anything. It simply
describes what is to be done. Now the recruit must figure out how
to create the perception you have just described (you think of it
as giving an order, but you are really just describing something
that you want to see created in your own perceptions). So the
recruit must figure out what "pulling back the bolt" means in
terms of perceptual consequences of motor actions: where the
hands and rifle should be seen and felt, what type of motion
should be experienced, and what sequence of events should appear.
You have no control over any of those processes. Only the recruit
can make them appear. So the recruit must establish reference
signals that specify hand positions, rifle position, movements,
and sounds, and must have in mind a mental picture of how the
bolt is to look and feel when it is "pulled back" as you
requested. And of course the recruit must be organized to act to
make the perceptions of what is now being experienced change into
perceptions like the ones that have been specified. The recruit
must contain the necessarily control systems.
Furthermore, the recruit must have some reason for converting
your words into specific reference signals. As you say, you don't
care what those reasons are, but without some reason to obey you,
the recruit will do as you say only by accident. Those reasons
may be totally voluntary; the recruit may enjoy the soldiering
game, and be very willing to do exactly what he is told to do
(and nothing he is not told to do). Another recruit, perhaps one
who is not there willingly, may act to avoid consequences of
disobedience by doing just the bare minimum that will prevent
punishment. But all that matters to you is that the recruit have
SOME sufficient reason for converting your words into specific
reference signals for specific acts of control, because you are
controlling your own perceptions, not the recruit's. You want to
see the recruits act in a certain way, perhaps because that is
your duty, perhaps because you enjoy the sensation of being in
control. But your reasons don't matter to the recruits, either.
Your interaction with them is at a low level of mental
organization: do this, do that, yes sir, no sir.
As you can see, PCT is concerned with the details of behavior and
how it works. A control-system model gives us a way of
understanding behavior that is quite unlike the way conventional
sciences of behavior understand it. We approach behavior as if we
are talking about a real physical system, in which nothing
happens unless there is equipment present to make it happen. The
normal common-sense approach, in which we just look at external
appearances and try to figure out what event leads to what other
event, is insufficient and mostly misleading. We are trying to
find how how the human system really works, in the same way we
would approach any physical system in order to understand its
internal mechanisms.
As you say, we have little detailed experimental evidence,
particularly about complex behaviors. But there is informal
experimental evidence that can tell us about specific behaviors
even though we can't fill in all the details. In any example of
behavior, we can look for the basic relationships involved in
closed-loop control: sensing the variable to be controlled,
comparing it with a desired state, and acting on the basis of the
difference. Sometimes we can set up quantitative experiments and
measure the apparent parameters of control like loop gain,
reference setting, and, roughly, the system dynamics. In other
cases, we can see the basic relationships, but at present have no
means of instrumenting a formal quantitative experiment. As in
any new scientific approach, we have to start with what we know
how to do and build on it. We speculate about parts of the system
we don't yet understand completely, and experiment with parts of
the system that are accessible to our methods.
Beyond the formalities of scientific methods, there is the matter
of the underlying conception of behavior. To conceive of behavior
as a process of closed-loop control is very different from
conceiving of it as a process of stimuli causing responses, or
plans creating emitted outputs. None of the people who have
believed in other models have shown those models to work for all
of behavior, either. They have all done as we do, which is to
apply the model formally where possible, and speculate about what
would happen if the same model applied to the entire system. In
those cases where we have been able to apply the model to real
human behavior, we have obtained results which are, I think we
have shown, far more predictive and reliable than those obtained
with other basic theories. And we deal successfully with
phenomena of behavior that other theories don't even recognize.
If control theory proves to be the key to understanding human
systems, it will utterly change the way scientists think about
behavioral organization and mechanisms of behavior. As an
organizing principle, control theory is entirely new to the life
sciences, even in cases where a few people have seen and
understood a few basic control processes. Most applications of
control theory in relation to organisms have treated the control
systems that have been found as oddities, exceptions from the
underlying blind physical and chemical processes that are assumed
to be basic. Seeing closed-loop control as the basic underlying
process will completely change the way we think about not only
individual and social behavior, but biochemistry and genetics.
That is why we sometimes claim that PCT amounts to a new science:
it is a new science of life that we mean.
···
--------------------------------------------------------------
Best,
Bill P.