can system dynamics learn

["ALLOCAR SRASBOURG"]

Hi everybody


I have questions but no answers.

First about the question and the object of the learning:

What is the object of the learning?
Is the model learning about the way it as a model is working
or is it learning something about the subject it is supposed to
work on?

Is the learning inside or outside the boundaries of the SD model?

If it is inside, is it inside the strict conception of the model or
can it discover something outside of that conception? This looks
difficult to imagine because even inside the boundaries of the model
the learning will be limited by the very conception of the model.
Is the model able to extend its original boundaries?
This looks like science fiction.

Is the model able to learn by itself ot does it need the help of a human?
In the case of inside learning or outside learning or both?
Who is then learning the model or the human?

Is there not a dynamic system, LIFE an exemple of a system having
learned outside or its original boundaries?
But was it learning or the selection of the most fit through hazard?

Is then pure learning outside or its original boundaries, only the
effect of chance?

Would a well conceived dynamic system that has boundaries
large enough for the subject studied, FREE to make stochastic experiences
even out of the bound ot its conception, having plenty of time and a very small
chance but still positive to find something a possible candidate?
Would "he"?

J.J. Laublé.



From: "ALLOCAR SRASBOURG" <allocar-strasbourg@wanadoo.fr>

["John J. Voyer"]

This has been a very interesting dialogue so far. I wanted to
add one observation.

No one has explicitly mentioned the distinction between single-
loop and double-loop learning. The former is when an
organization learns better how to hit its targets and goals.
This, implicitly, is the form of learning that has dominated this
dialogue so far. It also is the dominant mental model
underlying George Richardsons paper, which I just finished
reading.

Double-loop learning is when the organization questions its
targets, goals and underlying assumptions. I commend George
for touching on this in the latter portions of his presentation.
Even though he doesnt use the term "double-loop learning,"
this is what hes referring to when he asks, "Can a model
explore multiple policies and select on its own the most
advantageous?"

George mentions that this can be modelled, but would be
"daunting." Perhaps this is why so many organizations dont
actually do double-loop learning. Its probably pretty daunting
in real life as well as in models.

John Voyer, Ph.D.
Professor of Business Administration
School of Business
University of Southern Maine
96 Falmouth Street
P.O. Box 9300
Portland, Maine 04104-9300

voyer@usm.maine.edu
phone: 207-780-4597
fax: 207-780-4662