[From Bill Powers (941129.1005 MST)(]
Dag Forssell (941128 1315)--
It's hard to write a letter to an unknown person that will get that
person interested in your PCT management program. It's even harder when
you can't cite your own experience in applying PCT to real management
situations, and can present only one real instance of its use by another
person, ten years ago or more.
When you look at how Ed Ford has applied PCT in schools, you get a
different picture. Ed has used PCT in dealing with people in many
situations, from individual counselling to prisons to school districts
in half a dozen states. Most of what he teaches he's worked out himself
in the field, with real children and adults.
I think you need to get that kind of experience, and the only way I know
of to get it is the way Ed did: offer your help to other people, free if
necessary, and make it clear that you're developing a new way to handle
management problems but need hands-on experience to turn the theory into
a practical tool. It just seems to me that it's too early to present
this program as a finished and proven package. Salesmanship can't make
up for a lack of research. What worries me is the risk of selling a
program that _ought_ to work, but through lack of experience doesn't
handle the real issues and fizzles out, taking your hopes down with it
and doing the reputation of PCT no good, either.
If I were as young, energetic, and intelligent as you are, I would seek
out local businesses who would be willing to let me put in a program and
work with it for a while, experiencing real-life situations and learning
what the problems are that PCT can handle. When you add up the thousands
of hours you have put into developing seminars and promotional material,
you must ask whether the same time (at the same hourly wage) might not
have been spent better in trying to solve real management problems in an
experimental setting, in a real company -- even if you weren't getting
paid.
It seems to me that there are many people who would like to apply PCT
but have a problem in getting out of the theoretical mode. It's easiest
to test a new life jacket by subjecting it to thought experiments and
laboratory analysis, but the only really effective way is to put it on
and jump into the ocean. That's the only way to find out if the theory
works. And those who have got themselves wet are generally a lot more
convincing when it comes time to sell the life jacket to someone else.
ยทยทยท
-----------------------------------------------------------------------
Bill Leach (941128.1851 EST) --
Is there any evidence in studies of e.coli that there is such a thing
as "too much" nutrient? Not necessarily important but I am curious.
I don't know the literature that well. Koshland, my main source on E.
coli, didn't mention it. But it's reasonable to assume that E. coli
would avoid toxic concentrations even of nutrients.
Is there any evidence that shows that e.coli will avoid contaminants?
Yes. E. coli can seek or avoid something like 27 attractants and
repellents. Who says one-cell organisms are simple?
RE: reorganization
I gather that this would be a leaky integration as opposed to doing
table (memory) based mathematical calculations, yes?
A little more complicated than that. Consider the simple adaptive
control model that Tom Bourbon tested a few years ago. There was a
factor k that multiplied the integral of the error signal to produce the
output. The idea was for the model to adjust k by itself to achieve the
least error signal.
To do this an auxiliary control system was used. It sensed the _square_
of the error signal (to this auxiliary system, the error signal in the
main system was the controlled variable). There was a small number dk
which was added to k on every integration, so k was always changing (but
dk was very small, so k changed very slowly). If the squared error began
to increase, a new value of dk, between positive and negative limits,
would be selected at random. This random selection would continue until
the squared error began to decrease again. Then dk would simply continue
to be added to k on every iteration. The magnitude of dk was also
adjusted to be proportional to the squared error, so the smaller the
remaining error, the smaller the changes in k would be.
This system converged quite quickly to a value of k that kept the error
small. The method works well when there is an optimum value of k, with
the error getting worse for both higher and lower values. Since the
reorganizing method doesn't make any assumptions about the relationship
between changes in k and changes in the error signal, it can adjust k in
whatever direction is required.
RE: simplified model
It seems to me that this would be true only for the environmental
conditions used in your (and Bruce's) models but not so for Rick's,
again yes?
The simplified model will work with Rick's changed conditions, because
it doesn't depend on past values of dNut. All that the simplified model
does is make the two probabilities change in the right direction. The
basic control system is the same as the one Rick uses.
Bruce's model will work under Rick's conditions, too, if the
probabilities are manually set to the required values. All that won't
happen is for the probabilities to be adjusted automatically in the
right directions.
---------------------------------------------------------------------
Best,
Bill P.