Agent Architectures for Compliance

A Normative Multi-Agent System consists of autonomous agents who must comply with social norms. Different kinds of norms make different assumptions about the cognitive architecture of the agents. For example, a principlebased norm assumes that agents can reflect upon the consequences of their actions; a rule-based formulation only assumes that agents can avoid violations. In this paper we present several cognitive agent architectures for self-monitoring and compliance. We show how different assumptions about the cognitive architecture lead to different information needs when assessing compliance. The approach is validated with a case study of horizontal monitoring, an approach to corporate tax auditing recently introduced by the Dutch Customs and Tax Authority.