Towards a Logic of Graded Normativity and Norm Adherence

Abstract

A key focus of contemporary agent-oriented research and engineering
is on open multiagent systems composed of truly autonomous, interacting agents. This poses new challenges, as entities in open systems are usually more or less mentally opaque (e.g., possibly insincere), and can enter and leave the system at will. Thus interactions among such black- or gray-box entities usually imply more or less severe contingencies in behavior: Among other issues, in principle, the adherence of agents to norms cannot be guaranteed in such systems. As a response to this issue, this paper proposes a logic-based approach based on the notion of (possibly probabilistic) behavioral expectations, which are stylized either as adaptive (i.e., predictive) or normative (i.e., prescriptive). Some features of this approach are the enabling of "soft norms" which are automatically weakened to some degree if contradicted at runtime, and the possibility to quantify norm adherence using the measurement of norm deviance.