Offline norm evolution

Abstract

An approach to the pre-runtime design of normative systems for a class of problem-solving norm-regulated multi-agent systems is suggested. The basic idea is to employ evolutionary mechanisms to evolve efficient normative systems for so-called norm-regulated Dalmases, as part of the design process. The Dalmas architecture uses an algebraic approach to normative systems, in which normative consequences are based on an extended set of one-agent types of normative positions, which is given a semantics in terms of prohibition of certain types of state transitions. To illustrate the approach, a genetic algorithm is used to evolve norms for an example system. Furthermore, some approaches to reducing the algorithm’s search space, including to employ a notion of ‘operational equivalence’ of norms, are discussed. It is demonstrated that an evolutionary algorithm may be a useful tool when designing norms for problem-solving multi-agent systems.