Peer Pressure as a Driver of Adaptation in Agent Societies

Abstract

Abstract. We consider a resource access control scenario in an open multi-agent system. We specify a mutable set of rules to determine how resource allocation is decided, and minimally assume agent behaviour with respect to these rules is either selfish or responsible. We then study how a combination of learning, reputation, and voting can be used, in the absence of any centralised enforcement mechanism, to ensure that it is more preferable to conform to a system norm than defect against it. This result indicates how it is possible to leverage local adaptation with respect to the Rules of Social-Exchange, Choice, and Order to promote a 'global' system property.

Software Support for Organised Adaptation

“…The idea was that co-operative agents should manage the system by voting 'fairly'. Initial experiments showed that 'responsible' agents performed better than selfish or cautious ones (indeed approximated the outcomes achieved with a 'benevolent dictator') [5], and that social networking (gossiping) algorithms can be used on an individual and group basis to protect the system from self-interested behaviour [6].…”