I received the following referee's report last Friday:
> I don't think it sheds much light on multi-agent systems. It says
> something on social simulation but the link back to multi-agent
> systems is very tenuous. Little attempt to link the paper to
> terms, concepts, and literature of multi-agent systems.
>
The report was for the UK workshop on MultiAgent Systems (UKMAS). The
call for papers said that papers which had been presented at ICMAS ECAI
and similar would be especially welcome. So I sent my MABS98 paper,
"Social Simulation and Reality: Three Approaches". (the pre-print
version is at http://www.cpm.mmu.ac.uk/cpmrep35.html). The UKMAS chair,
Michael Fisher and, through Michael, the author of the above report (who
remains anonymous) have agreed to its circulation to the simsoc list in
order to enable me clearlyly to raise the issue of the relationship
between their approach to MAS and the broader social simulation
approach.
It has seemed for some time that certain elements in the computer
science end of the MAS community have wanted to preserve a kind of
purity in the use of phrases like agent architecture or agent systems.
Perhaps this is appropriate. Perhaps the use of the agent concept in
computer science is simply different from the concept as used in social
simulation. But there are two reasons for believing there to be a
substantial degree of elision between the two.
The first reason is that shared techniques exist and are frequently
published in both the conventional MAS and the social simulation
literature. The main technique shared in this way is the representation
of agents as BDI logics. I wonder if, for computer scientists, this is
agent modelling as engineering: in a given set of conditions a software
agent programmed in a particular way will always respond reliably and
predictably. However, authors such as Castelfrancchi, Conte and their
colleagues have used essentially the same form of representation to
investigate the conditions under which social norms will emerge. This
seems to me to be much less like engineering and much more like social
science.
The second reason is that in such a new field of research as MAS, an
exclusionary approach can lead not only to an unnecessarily small base
for further development but also to the application of bad science. For
a description of such an application of bad science and a constructive
alternative approach, see my invited talk to the AgentLink SIG on
intelligent information agents at
http://www.cpm.mmu.ac.uk/~scott/i2a/markets.html
I see the quoted report at the start of this message as a challenge to
demonstrate the power of social simulation not only as a means of
analysing and understanding societies of humans and other primates but
also as a source of ideas, techniques and procedures for use in the
design and implementation of societies of software agents. Both the
computer science and the social simulation contributors to the MAS
literatures would be likely to benefit.
Michael Fisher helpfully wrote:
> I guess it probably reflects a particular view from MAS people -- and
> something that you might need to convince the MAS community is
> incorrect?
>
I don't think we **need** to convince them but I do think that we should
try. With that one amendment, I endorse Michael's sentiment because I am
convinced that both they (the computer science MAS community) and we
(the social simulation MAS community) would benefit from the discussion.
I wonder what other members of the list think.
--
Scott Moss telephone: +44 (0)161 247 3886
Director fax: +44 (0)161 247 6802
Centre for Policy Modelling
Manchester Metropolitan University
Aytoun Building
Manchester M1 3GH
UNITED KINGDOM
http://www.cpm.mmu.ac.uk/~scott
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|
