[From Bill Powers (991014.0959 MDT)]
Dan Palmer (991013.1200)--
2) Before going any further, I am convinced that is is essential to clarify
what we mean (or what we specify) by our action language. That is, what
units
do we point to with statements like "Bob driving at 100?" One interpretation
is that we point to Bob. This is the interpretation adopted by both
behaviorism and cognitivism, though differing in the part of Bob they zoom in
on (either his bodily responses or the contents of his head/mind). I think
this is fundamentally the wrong interpretation. Interpreted literally,
action
language refers to the results of bodily efforts (Guthrie) or the things that
people get done (Lee, e.g., 1995).
This is part of the story but not enough of it. I agree that we must speak
mainly of what behavior accomplishes rather than the means by which
behavior is produced, but this leaves open the question of how to define
what is accomplished. Any given action has many effects, but only a few or
even only one count as what "gets done." When I pull open a drawer to get
some socks out, the drawer makes a noise, but if it made a different amount
of noise I would make no attempt to silence it or make it louder, so that
is not what "gets done." The drawer extrudes six inches, but if it happened
to be eight inches I would do nothing to correct the gap, so that isn't
what "gets done." When I pull out my brown socks I move a pair of blue
socks, but that isn''t what "gets done," for I wouldn't reach in to move
the blue socks if they didn't happen to be disturbed. And if I'm doing all
this in the dark, getting a pair of brown as opposed to blue socks isn't
what gets done, either, because I have no way to tell which color I'm
getting. All of these other effects are simply side-effects; if they came
out differently I would do nothing to change them. The only result that
would call for corrective action would be if my hand came out of the drawer
with no socks in it.
There are two viewpoints from which to evaluate what "gets done" -- that of
the observer, and that of the doer. Only the doer really knows what a
physical action is intended to accomplish and which of its consequences are
accidental or incidental. This is because only the doer knows _what
perception_ is intended to be affected by the action. The observer may try
to guess which consequence is the intended one (using his own perceptions),
but without some systematic way of testing guesses, guessing correctly is a
matter of luck. The observer projects his own desires and intentions onto
the doer; without a systematic approach, what else can the observer do? Of
course PCT offers the Test for the Controlled Variable, which is a
systematic way of testing guesses about what "gets done."
Further,
action language often specifies the changes that a person might be said to
control (we are only ever interested in a small subset of the millions of
effects that people have everyday).
Speaking of changes, if you mean this literally, is appropriate only if
change is the intended effect. For example, I can move a coin on a tabletop
an inch to the left of its present position. Since the absolute ending
position relative to other objects is not specified, I can speak of
controlling a _change_ in position (as I see it), where the change is
independent of the location in space of the starting position.
However, what I control may be a position relative to something else: move
the coin until it covers a spot on the tablecloth. Since no starting
position is specified, the amount of change of position is unspecified. But
I can still achieve the end-relationship quite easily, using position
control rather than change control.
There are 11 kinds of controlled variables that I think I have identified;
change is at the fourth level (transitions). So control of changes is far
from the whole story.
PCT
emphasizes that organisms are _not_ passive recipients of stimuli and
emitters
of responses, that organisms causally intervene in the world, and that it is
organisms that control (i.e., effect those changes) rather than be
controlled. And indeed, in its experimental demonstrations (such as
tracking)
PCT is explicitly concerned with effects.
Agreed.
However, despite these facts, it
seems to me that PCT stops short of severing all affiliations with the
former,
more conventional interpretation, in which Bob gets more attention than he
deserves.
I would say that it is not Bob, but the observer who gets more attention
than he deserves. It is very difficult for the observer to pick out of all
the effects produced by Bob's movements just those that Bob intends to
produce -- that is, those effects which Bob is perceiving, and for which
perceptions Bob is specifying reference levels.
4) The things done I am talking about are properly conceptualized as changes
in the state of an object or medium.
Why are they not the end-points of the change rather than the change
itself? When end-point control is involved, the path may be disturbed yet
the same end-point will be achieved by altered means. In this mode of
control, neither the actions nor the changes are predictable, yet the
end-point remains predictable (especially by the person carrying out the act).
Take the depression of a mouse button.
That change in the state of the button from up to down occurs when a whole
bunch of constituents come together in a certain way. The constituents
include a biological organism (which in turn includes a brain, hands,
fingers,
etc. in short, a whole integrated organism), the button, the light by
means of
which the button is seen, a history of relations between button depressions
and other things done (such as a change in the cursor location), gravity, and
so on. Each of these constituents is part of the subject matter of another
science/s. Also, each of these constituents is a necessary but not
sufficient
condition for the occurrence of a button depression. So when Bill asks "how
is the person in this situation organized" he is focusing on one of many
constituents, and one that happens to be physiological. And the only
reason I
can see for focusing on this particular constituent to the neglect of others
is that there is some sort of residual attachment to the organocentric
interpretation of action language as described above.
You seem to be using "organicentric" as a pejorative term, here. If I refer
to the operation of a car by describing how its engine, gears, and wheels
work, would you dismiss this as "mechanocentric?" Would a description of
the thyroid gland's operation in a loop involving thyroid-stimulating
hormone and circulating thyroxin be dismissed as "chemocentric?" In the
pressing of a mouse-button, the important components involve, as you say,
the physical properties of the mouse (according to our physics-models), the
neural properties of the brain and sensors (according to neurological
models), and the physiological properties of the muscles (according to the
models of physiology). Is it organocentric to include the brain and muscles
in the description of mouse-pressing? Would it not be physico-centric to
leave them out?
Perhaps you have an idea that it makes no difference where we put the locus
of control. This is another of those myths that have been propagated
through the decades since mid-century. There is a great deal of difference
between the control system itself and the environmental part of the control
loop. The control system controls; the environment does not, unless it
contains another control system. In one of my Living Control Systems books,
the second, I think, I wrote a little piece called "The assymetry of
control." In it I showed why the organism can be said to control its input
from the environment, while the environment cannot be said to control its
input from the organism. The difference is in the error amplification,
which in the organism can amount to many hundreds, while the equivalent
part of the environment typically has an amplification of less than 1 --
it's normally an attenuator. This assymetry is what allows the organism to
vary its behavior to force the environment's state to be what the organism
wants, while the environment, although capable of disturbing the organism's
state, can't determine it in the same way.
5) Now I can state the problem bluntly. If PCT, like all psychology,
properly
begins with action language, and interprets this action language in terms of
things done (as it seems empirically to do), then why does it couch its
explanation in terms of brain or person organization rather than in terms of
the organization of things done?
Because what is "done" is determined strictly by what the organism
perceives, and by the organism's inner specification for the state in which
that perception will be maintained by variations in the organism's actions
on the environment. This is the key concept within control theory that
makes PCT different from simple empiricism. The empiricist assumes that
what an organism is "doing" is obvious -- it is what the empiricist notices
and considers salient. But PCT forces us to acknowledge that what an
organism is "doing" is far from obvious, for all we can see from the
outside is how the organism physically affects the environment, both
directly and indirectly. And without control theory to organize our
investigations, we have no way to find out which of the effects _we_
consider important is also important to the organism.
If an example of "controlling," such as "Bob
driving at 100" or "keeping the square in the middle of the screen" literally
consists (in psychological perspective) of many hundreds (often thousands and
more) of ongoing things done falling in complex relations, why not do them
justice at their own level rather than interpreting them as evidence for
hypothetical (or real) brain processes?
Because without postulating brain processes, there is no other way to
explain how all these environmental interactions produce the result we see.
A proper model must do justice to the physical effects, but that is not
enough to explain the observations.
Just consider Bob's speedometer. That speedometer reading is influenced by
the pressure of Bob's foot on the accelerator pedal, but it is also
influenced by the quality of the gasoline entering the engine, by the air
mixture, by the level of the engine oil and the state of wheel lubrication,
by bumpiness in the road, by open windows in the car that increase its
aerodynamic drag, by the head- or tail-wind component of any wind that may
be blowing or gusting, by the softness of the roadbed at each moment, by
the tire pressure, and by up- or down-slopes in the roadbed.
So clearly, the final speedometer reading is the resultant of many
converging influences, most of which are variable or unpredictable. Yet we
observe that the speedometer needle remains very close to a reading of 100
KPH. If we analyze this situation carefully, with numbers, we discover an
astonishing fact. No matter what combination of other influences may exist,
or how they may be varying over the short or long term, or even whether
they originate outside or inside the car, Bob's foot will vary its pressure
on the accelerator pedal just enough and in the right direction to cancel
out the sum of all other effects on the speedometer reading (even a loss of
calibration in the speedometer!), and keep the reading at 100 KPH.
Furthermore, on request, Bob can start maintaining the reading at 80 KPH or
any other reasonable number we suggest.
When we ask how Bob can possibly produce just the foot-pressure that is
needed, when there are so many other variable forces acting, mostly unknown
to Bob, we pose a problem that can ONLY be answered by the theory of
negative feedback control. While there are many ways to embody a given
negative feedback control system, we can at least speak of a canonical
system to which all other systems that behave the same way are equivalent.
We know that the state of the variable being controlled has to be sensed
somehow; we know that the reference state has to be specified somehow; we
know that the action must vary in a systematic relationship to the
difference between the sensed state and the specified reference state of
the controlled variable. Given those requirements, it is easy (if you
understand control systems) to construct a picture of a simple
neuromuscular control loop in which the required relationships are
specified as functional blocks, if not actual neural connections. And in
some cases, we can establish the parallels between the canonical system's
interconnected functional blocks and the anatomy of the nervous system and
musculature.
Without a model like this, we can only accept the empirical findings and
marvel at them.
Bill Powers wrote:
"You can show, for example, that if you prevent the driver from visually
perceiving the position of the speedometer needle, that position ceases to be
controlled. The driver's sensory processes are clearly involved."
I write:
Or, how about this: if you prevent the driver from getting some things done
(those involved in "looking at the speedo"), the driver can no longer get
other things done (those involved in "controlling the needle's position").
True, but why be so vague when we can be exact about what is required and
why? If the driver is to keep the car on the road, for example, he must
specifically get sensory information that pertains to the relationship
between the car and the road. It will do no good to get this information
written on a piece of paper he has in his pocket, or to give it to someone
in the car other than the driver. Furthermore (back to the speedometer),
it's not enough that the driver look at the speedometer. The driver must
specifically be perceiving the position of the needle relative to the speed
scale; no other perception involving the speedometer will enable him to
keep the speed constant. The driver must specifically sense the variable
that is to be controlled, or some variable that directly depends on it and
only on it. All else about the speedometer is irrelevant.
I only disagree (or at
least am curious to clarify) about where the referents of those functions are
supposed to reside (and I think this _is_ an important issue).
They reside in the physical form of neural signals in the brain of the
actor -- that is the basic PCT model. HPCT further proposes that they are
the outputs of higher control organizations in the brain which control more
abstract and aggregate variables. The highest level must of course be dealt
with in a different way.
Best,
Bill P.