Is there an issue about different types of system here ?

Manufacturing systems are by design, relatively stable and repetitive
(Types 2 or 3 according to Wolfram) Therefore, it has been worth
building specific and detailed models - because those models are often
predictive over a range of different initial conditions.

However, most social systems are complex and non-repetitive (Type 4).
Therefore a highly detailed model, based on empirical data, may well be
invalid under any slight variation.

I suspect we have to use models of workforce behaviour, rather
differently than those of engineered systems.

Cheers,
Jeremy

>>> Peer-Olaf Siebers <[log in to unmask]> 10/02/02 04:42pm
>>>
Dear All,

I would like to discuss a general question about the philosophy of
Multi
Agent System models.

First a short introduction to my research (a working paper with more
details can be found at
http://www.cranfield.ac.uk/~ws056755/research/).
The ones who know me already can skip the next two paragraphs.

I am working on a group project looking at "Human Performance Modelling
as
an Aid in the Process of Manufacturing System Design" at Cranfield
University (UK). My PhD research contribution to the project is to look
at
ways to model interdependencies between workers and their environment
(which includes other workers, cultural- and physical environment).

Discrete Event Simulation models allow engineers to understand and
predict
the behaviour of manufacturing systems. A standard way of taking
workers
into account is to model them as resources not considering any
person-,
organisational- or environmental influences. But worker behaviour
differs
notably between people, tasks and systems. There is a need to model
the
behaviour of individual workers and workgroups during the
manufacturing
system design process, especially in labour intensive systems like
assembly
lines. This would lead to more realistic line performance predictions
and
could also be used as an aid to design better working environments
that
will improve worker performance. As a technique Multi Agent Systems
(MAS)
as used in Computational Organisation Theory seem to be just perfect.
The
output of such a simulation tool, which can be described as a kind of
artificial white-room, will support the development of functions and
distributions that describe variations in state and performance of a
workforce at individual and group level. It can be used inside new or
existing manufacturing simulation models.

There are lots of frameworks out there. The big problem is to find
some
source code of real world examples (something less abstract and more
related to human performance than heat bugs that live in a 10x10
maze).
Another point that worries me a bit is that I haven't heard of any
real
success stories about using MAS to model human behaviour within
complex
systems (except from the MOD in the area of decision making). It seems
like
research in this area never gets to a stage of maturity. After
developing a
framework, building some very basic example models, and running a few
tests, which show that theoretically the framework could work,
everybody
seems to be happy. In the concluding remarks of research papers about
these
frameworks it is very often stated that it is just a matter of getting
empirical data and the framework would be perfectly suited to model
human
behaviour.

But no one is actually doing this final step. Or is my impression
wrong?
Are there any applications (that use MAS and empirical data) that are
actually capable of predicting human behaviour?

Research projects in the department that are related to my project are
expected to produce a wealth of empirical data that link environment
and
people characteristics with performance. To start the practical part of
my
research my plan is to build a basic MAS model using Java (which I am
currently teaching myself) and the JAFMAS libraries (Java-Based
Framework
for Multi-Agent Systems: http://www.ececs.uc.edu/~abaker/JAFMAS/) with
some
basic social intelligent adaptive agents. When I have designed the
system I
would like to run some tests to get a better understanding about the
technique itself and a feeling for the data requirements (so I get the
right data in the right form from my colleagues) and the results I can
expect from such a system (so I can define the user group and tell the
users what they can expect from the tool). As a next step I would like
to
extend the agent structure to fit the requirements of a worker
representative.

To be honest it is quite hard to get started - maybe I am doing
something
wrong. Can you think of an easier way to undertake this first step. Is
there anything existing that I could play with in order to gain this
understanding or do you know of some detailed descriptions on how to
build
such a system?


Thanks for your advice,

PEER-OLAF SIEBERS


Legal disclaimer
--------------------------

The information transmitted is the property of the University of Paisley and is intended only for the person or entity
to which it is addressed and may contain confidential and/or privileged material.  Statements and opinions expressed in this
e-mail may not represent those of the company.  Any review, retransmission, dissemination and other use of, or taking
of any action in reliance upon, this information by persons or entities other than the intended recipient is prohibited.
If you received this in error, please contact the sender immediately and delete the material from any computer.

--------------------------