Page 275 - Socially Intelligent Agents Creating Relationships with Computers and Robots

258 Socially Intelligent Agents

much more difficult to design software agents and mechanisms for general
multi lateral negotiation. In general, bilateral negotiation is a special case and
there is no reason to infer anything about negotiations among three or more
parties from results with models of bilateral negotiation.
The decision by each agent concerning which other agents to engage in
negotiation is far from trivial. In the model, agents were concerned with the
trustworthiness, reliability, helpfulness and similarity of other agents. Agents
did not appear to learn which, if any, of these characteristics should be given
priority in selecting negotiating and coalition partners.
In general, it would be hard to justify as good science the repeated revision
of abstract simulation models until we found one that produced convergence in
a negotiating process and then to assert that such a model describes a socially
useful approach to negotiation. Producing such a model is a purely intellectual
exercise. To be useful, it must be validated. To be validated, it must be shown
to be a good descriptor of actual successful multi lateral negotiations. If such a
model can be validated against a range of negotiating processes, we might then
have some confidence in the model as a pattern for good negotiating practice.
It is hard to see any substantive difference between validating abstract models
and building models around descriptions of actual negotiations. Both involve
the development of a general understanding by means of the development of
descriptively accurate simulation models.

References

[1] Chialvo, D.R. and Bak P. Learning from Mistakes, Neuroscience 90:1137-1148, 1999.