Social Laws For Agent Modeling

Amol D. Mali

Multiagent domains emphasize that agents should be able to predict actions of other agents. A popular mechanism to achieve this is that of providing agents with models of other agents. Social laws have been proposed as a model of multiagent interaction with claims that they can get rid of perception, reduce communication cost and planning time. However the utility of social laws for agent modeling has not been explored. In this paper types of social laws are characterized and criteria for evaluating and comparing them have been defined. These criteria also serve as measures of the utility of social laws in modeling agents. It is shown here that to be easier to design and test, social laws should exploit other representations like potential fields to model interactions between agents. Social laws and belief, desire and intention based (BDI) agent modeling paradigm are compared. Use of laws to encode recta-knowledge, characteristics of laws and situations under which they can be used are discussed. Some unaddressed issues have been raised at the end.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.