ABSTRACT

Agents in an uneconomic world: what integral calculus can tell designers of multi-agent systems

John Campbell, Professor of Computer Science, UCL

Now that systems of multiple autonomous agents are in common use, issues of coordination and cooperation between agents are receiving much attention in both basic and commercial research. The emphasis is on economic criteria: quantifiable payoffs, optimisation of utilities, bargaining about prices, and related ideas, as means of ensuring or enforcing effective behaviour. But there are problems for which it is difficult, or simple but unrealistic, to establish such criteria.

The seminar will use one particular test application - exercises in integration in calculus - where the economic criteria are not particularly helpful, to suggest directions in which the structures and properties of agents and agent systems can be developed when the standard (in terms of agent fashions in 1999) assumptions about how agents should regulate their interactions are not sufficient. Topics that will be covered include modelling, use of case-like information, boredom, reasons why agents should consider communicating even when there is no immediate economic payoff for doing so, and what can be done about coordinating groups of agents where each agent has only one narrow competence and no conception of good manners (the "to the citizen with a hammer, everything looks like a nail" problem).

Computer Science Home Page

Maintained by rbennett@cs.ucl.ac.uk