[From Bill Powers (941130.0725 MST)]
Bruce Abbott (941129.1230 EST)--
That was a useful review of the Ecoli4 developments -- Chuck Tucker will
be pleased.
I think we need to review the bidding from another angle, too. There is
nothing basically wrong with what the Law of Effect proposes: Stated
from the PCT perspective, the proposal is that we (all organisms that
learn) learn to control by finding actions that have systematic effects
on perceptions that are important to us. Even Albert Einstein had to
learn which way to twist the cap on his tube of toothpaste to get it
open.
PCT, however, introduces a new consideration: the question of whether
Albert wanted to brush his teeth with toothpaste. If he wanted to use
baking soda, or didn't use anything but water, or didn't want to brush
his teeth at all, the direction of effort required to take the cap off
the toothpaste, while still remaining a fact, would be irrelevant. What
gives the Law of Effect its lawful-seeming force is not any particular
consequence, but the fact the organism intends for certain consequences
to occur.
The Law of Effect and the conception of control by consequences try to
explain the counterclockwise effort that gets the cap off the toothpaste
without mentioning that this consequence is important only because the
organism is actively looking for ways to control it. When this critical
factor is left out, all that remains is the statement that in order for
a certain consequence to occur, some preceding change in the environment
must occur. In order for the cap to unscrew, a counterclockwise torgue
must be applied to it while the opposite torque is applied to the tube.
In order for the puzzle box to open, a string attached to the latch must
be pulled, whether by a cat or by Albert Einstein or by a falling book.
The Law of Effect becomes merely a statement about properties of the
environment: a law of phsyics, or engineering.
The Law of Effect does say one other thing. If the cap is unscrewed, or
the puzzle box is opened, then the organism must have caused the
necessary prior physical state of affairs to occur (assuming that other
causes have been ruled out). But this is not an explanation of learning:
it is merely a statement that learning occurred, a definition of
learning. Where the organism did not previously cause the necessary
prior change to occur, eventually it does cause that prior change to
occur. That's the only legitimate way to describe learning in terms that
make no use of a model of the insides of the organism (as I have always
understood that Behaviorists with a capital B are not supposed to do).
So the Law of Effect as a behavioristic law merely states the obvious:
if B occurs, and if A must happen in order that B occur, then A must
have happened.
This leaves open two important questions. The first is why the organism
selected an A that affects B; this turns immediately into the question
of why the organism wanted B to happen. The second question is HOW the
organism found the A that produces B, given that it wants B to happen.
The answer to the both questions can be given only in the form of a
model of processes inside the behaving organism.
···
-------------------------------------
In Ecoli4, we can work from the final result backward to the process
that brought it about. The final state of affairs must be that when dNut
is positive, the interval between tumbles must be significantly longer
than it is when dNut is negative. This is because we assume that the
organism wants a positive value of dNut, and only this relationship will
produce a positive value of dNut in the given circumstances. How this is
achieved is irrelevant; if this final condition is achieved, and only if
it is achieved, E. coli will make its way up a gradient of an attractant
and the consequence of tumbling, dNut, will be maintained positive on
the average.
Now we must ask how this state of affairs can be achieved. The first
thing that must be found is a way of changing the delay between tumbles.
In one model, a timer would be incremented by variable amounts, and when
it reaches a fixed value a tumble occurs. Now the delay depends directly
on the size of the increment given to the timer. Another way is to
assume that a a random number is repeatedly compared against a
parameter; if less than the parameter, a tumble occurs. Now changing
that parameter will alter the average delay between tumbles inversely to
the magnitude of the parameter. So by either method, we obtain a
parameter that can be used to alter the delay (although one parameter
would have to increase where the other decreases, to achieve the same
effect).
Now we want to change the parameter in a way that will yield the desired
result.
One way is for the organism to alter randomly the relationship between
dNut and the tumble delay (using any of several possible mechanisms),
and cease this random alteration when the desired state of dNut is being
achieved, on the average.
Another way could use a systematic process: start by making the
parameter large if dNut is positive (etc), then systematically decrease
it. Then select the relationship that gives the most positive value of
dNut on the average.
As you propose, the process could be logical and based on properties of
the situation that predict the outcome of increasing or decreasing the
comparison number in probability calculations.
And as I proposed, a second system could be devised that simply adjusts
the parameter on the basis of a criterion that is independent of dNut,
so that the organism tumbles up or down the gradient as required to
satisfy the new criterion (absolute concentration of nutrient, in the
case I proposed, although it could have been anything else).
There are other possibilities. Each one amounts to proposing a specific
model of the organism. I am trying here to put Ecoli4 into the context
of models of the proposed learning process, to show that what you
proposed is one possibility, but by no means either the only one or the
simplest one that would work.
------------------------------------
The three-term contingency that you modeled will work if
1. There is some indicator in the environment that can predict the
direction of change of the system parameters (PTS- or PTS+) that will
produce a favorable result, and
2. The organism can perceive that indicator and is smart enough to see
its implications and take advantage of them in the correct way.
Even given both of those requirements, the fact that the organism learns
to control the consequence does not by itself prove that this particular
learning mechanism was used. As I have tried to show above, there are
other ways of learning the action (or one of the actions if more than
one would work) that will generate the desired consequence. We can't
tell just from seeing that the consequence is achieved which method was
used to achieve it.
In my latest modification of Ecoli4, all I did was remove the indicator
that predicts the best way to change the parameters. Naturally, since
your principle of learning depended on the presence of valid indicators,
your model then failed. I showed part of the remainder of the argument,
which is that if the parameters were properly adjusted _somehow_, the
control process could still work. What remains to be shown is that one
of the other methods of learning (other than my two-level model) would
still work. You can probably see from my descriptions of other methods
that they would work, because the ones I described don't use any
predictors of the best way to adjust the parameters. But I will spend
today working up at least one alternative model of learning that will
work where the Law of Effect model fails.
------------------------------------------------------------------------
Dag Forssell (941129.2100 PST) --
Sure it is hard. A writer wants to be sensitive to the many ways
expressions can be interpreted so the letter does not needlessly hit
the trashcan. I have struggled with how to present the concept of what
PCT is and how to best target prospects. The letter reflects this. My
writings that follow represent my experience properly. I am not
claiming experience where none exists, but refer to others.
I know you're not making false claims -- what you write would be more
impressive if you allowed yourself to do so. Why not make false claims?
Everybody else does it!
The basic problem is that there isn't any one output that will create
the input you want, operating through other people. The output that will
achieve what you want depends on the other person. What you're doing is
fishing, trying to select the bait that will attract a bite. But you
have no way to know what the fish will find alluring, and you're
handicapped by the fact that you have to use real fish food for bait,
not wishing to rely on phoney lures.
What you say about Jim Soldani should give you some real cause to
ponder. Jim is in a company, you say, where he is unable to teach or use
PCT in his job. Suppose he had set himself up as a PCT management
consultant on the basis of his past record and had somehow got into that
company. He would have failed! The fact is that despite knowing
something about PCT, Jim does not know how to turn a company like that
around. Nobody has worked out how to use PCT in such a situation, and
perhaps the true answer is that PCT, as we understand it today, does not
have the answers.
I am not selling a program that _ought_ to work.
But that is just what you ARE selling. Your approach works fine in
demonstrations, and in role-plays where all the people understand PCT.
The abstract principles of PCT are well-established. The laboratory
experiments work. But NOBODY HAS TESTED PCT IN A REAL MANAGEMENT
SITUATION EXCEPT JIM SOLDANI AND HE IS UNABLE TO APPLY IT IN HIS PRESENT
MANAGEMENT SITUATION. That is the cold hard reality. PCT _ought_ to
work, but the track record so far is that it works only under some
conditions -- and maybe something else in those conditions other than
PCT is what made the difference. We have to FIND OUT whether PCT works,
and when it doesn't, WHY NOT.
All you can really honestly offer anyone is a deal that allows you to
find out if PCT works or can be made, with experience, to work. I can't
see any way of doing that except to get interviews with officers of
businesses and lay out exactly what you propose to do, part of which is
enlisting them as partners in the exploration of the possibilities of
this new approach, with no promises. You can tell them what you find
persuasive in the theory and how you would go about putting the theory
to practical tests. But you simply can't sell this as a service, as a
completely-worked-out approach to management psychology, because that's
not what it is -- yet.
What exactly do you propose to put on and who are you going to throw
in? The teacher or the students?
There you have it. Both. And the students have to know what they're
getting into.
A group of students who all learn PCT together from me will make a
water ballet in industry far better than Jim Soldani ever did or I ever
could alone.
Now you're talking. I think you can sell that. You can sell a proposal
that says "Look, I'm an expert about a new theory of human behavior that
is getting a lot of people excited. I want to apply this theory in the
field of management psychology. But I can't do it alone; I've been out
of management too long and haven't been in a position to try it out for
myself. I need the help of a bunch of real managers who are living the
problems every day, who can learn PCT from me, and then turn around and
tell me about all the problems in applying it, and help me to work out
the answers. I need people who will try out these solutions in their own
companies and come back with reports on what worked and what didn't
work."
You can do this in your own living room while you make a living doing
something else. Use the contacts you already have, get them to spread
the word, send out a few invitations. Gather a group of people to meet
regularly, people who will become your friends and allies, teach them
PCT with a firm hand, and get them to sign on to a real project. Your
picture of a "group of students" is right on the mark. That is what you
need: students, not customers. Even better, colleagues. The customers
will come later.
-----------------------------------------------------------------------
Best to all,
Bill P.