[From Bill Powers (2003.05.12.0750 MDT)]
Here is a quote from the start of Chapter 3 in Argyris' _Action Science_,
about page 80 of the book:
"The theory of action approach begins with a conception
of human beings as designers of action. To see human behavior
under the aspect of action is to see it as constituted by the
meanings and intentions of agents. Agents design action to
achieve intended consequences, and monitor themselves to learn
if their actions are effective. They make sense of their environment
by constructing meanings to which they attend, and these
constructions in turn guide action. In monitoring the effectiveness
of action, they also monitor the suitability of their construction
of the environment." (p. 80-81)
This is one of the theories with which PCT has had to compete since the
beginning. The organism analyses its environment, and designs or computes
the actions required to achieve intended outcomes. When you ask about the
details of what is needed to be able to do this, you end up with "modern
control theory," and a model with infinite knowlege of the physical world,
the laws of physics and chemistry, and its own anatomy -- not to mention a
Pentium 40 computer and actuators and sensors of infinite precision that
never change their calibration, in an environment containing no unexpected
("unmodeled") disturbances.
It also includes a second model that is actually incompatible with the one
just described: it says that actions are guided by sensory inputs , not
desired consequences. This seems to be an attempt to merge the
stimulus-response view with the cognitive-science view.
Now consider this passage from later in the chapter:
" ...agents learn a repertoire of concepts, schemas, and strategies,
and they learn programs for drawing from their repertoire to
design representations and action for unique situations. We
speak of such design programs as theories of action.
These theories may be thought of as a very large set of
complexly related propositions. The form of a proposition in a
theory of action is, "In situation s, to achieve consequence c,
do action a" (Argyris and Schon, 1974). From the perspective
of the agent who holds the theory, it is a theory of control. It
states what the agent should do to achieve certain results. From
an observer's perspective, to attribute a theory of action to an
agent is to propose a theory of explanation or prediction. In the
language of the previous chapter, it is to make a dispositional attribution.
The example we used is, "John follows the rule, 'If I
am about to deprecate someone, first deprecate myself." " But
from John's perspective, this is a theory of control. We can see
this by making explicit the intended consequence of enacting
the rule, which, let us suppose, is to avoid making the other person
defensive. Hence a proposition of a theory of action can be
understood both as a disposition of an agent and as a theory of
causal responsibility held by an agent."
When I first came across this passage I thought it looked promising, but
the catch is in what Argyris means by the word "control". He does not mean
varying actions to control perceptions, but says " a theory of control ...
states what the agent should do to achieve certain results." That is not
what control means in PCT. He is repeating what he said in the first quote
above: control is a mtter of deciding what output to produce to have an
intended consequence.
If my guesses about the levels of control are correct, we shouldn't be
surprised to find that Argyris' writings include programs, principles, and
system concents: the above snippet clearly shows programs (the rule for
self-deprecation) and principles (don't make others defensive). Everybody
who writes about such things speaks in terms of programs, principles, and
system concepts whether they know it or not (they usually don't), so this
is nothing startling. What distinguishes Action Science from other schemes
is only _which_ programs and principles are proposed to be important.
The basic problem is this. Argyris says, " The form of a proposition in a
theory of action is, 'In situation s, to achieve consequence c,do action a
(Argyris and Schon, 1974)' ". This is a purely logic-level prescription,
formally identical to Skinner's theory of reinforcement. The "situation" is
the discriminative stimulus; the consequence is the reinforcement, and
"action a" is the behavior that is reinforced. I claim, of course, that
while it is possible for people to construct theories of action of this
sort, it is also possible (I would say probable) that the behavior
described in this way is not actually produced that way at all.
This problem is similar to one I have mentioned before concerning
rule-driven behavior. You may be able to interpret behavior as following a
certain symbolicly-stated rule, but symbolically-stated rules may have
nothing to do with that behavior even if you can predict it from the rule.
Consider the raindrop rule: if you have not hit the ground, keep falling.
Otherwise, go "splat." This rule predicts what raindrops will do in the
vast majority of cases, but raindrops do not apply this rule or behave
because of it; they fall and go splat because of gravity and mechanical
interactions. The rule is in the eye of the observer.
In the case of the "theory of action", it may well be that we observe
action a, in situation s, to be the one that produces consequence c, so we
make up a rule to summarize this observation. The flaw in our assumption
will remain undiscovered until something in the environment changes so that
in the same situation. action a produces consequence d . Then we find that
instead of action a and a new consequence d, we get a new action b and
consequence c as before. The action, under the same perceived conditions,
changes so as to make the same consequence as before occur. The rat changes
its action until, after the same discriminative stimulus as before, it
makes the reinforcer appear again.
To discover these somewhat surpring anomalies, one has to deliberately
introduce disturbances that alter the way behavior affects the final
outcome. Then it becomes clear that what is controlled is the outcome, and
that this is not done simply by repeating the behavior that produced it
before. To understand how it _is_ done, you have to get into negative
feedback control systems and understand how they work.
I find books like _Action Science_ very hard to read; I feel lost in a
blizzard of words. Once in a while something interesting shows through the
murk, but most of the time I find myself wondering "How does he know that's
true? Why did he cite that person? Why did he use that word? When is he
going to get to the point?" Little snippets of what this person or that
person said don't tell me anything, particularly when nobody ever defines
any word, and when mentions of some idea are simply dropped, as if they had
served some purpose and now could be forgotten. I forget all too easily.
I guess I'm just not the academic type.
Best,
Bill P.
Best,
Bill P.
"Pay no attention to that man behind the curtain."
The Wizard Of Oz
"Convince a man against his will;
He's of the same opinion still."
Embroidery by Grammy Alice