Agent-directed Runtime Norm Synthesis

To maintain fitness-for-purpose, the set of norms governing a MAS will typically need to evolve to reflect the changing needs of both participants and the environment. We put forward a conceptual framework to address this problem comprising dynamic institutions (sets of norms), that depend upon the formulation of new norms and the revision of existing norms, informed by the experiences of agents participating in the MAS. The objective is to allow participating agents to influence the revision of the norms governing the MAS, thereby taking a first step towards adaptable self-governance of socio-technical systems through explicit norms. This paper proposes a novel framework for revising at runtime the norms of a formally specified institution, directed by the agents in the MAS. The framework employs special-purpose synthesiser agents with partial observability of the state of the MAS to formulate new norms or revise existing ones, in response to requests from agents for changes to the institution. To demonstrate the feasibility of the framework, we capture a set of norms using the instAL institutional specification language and revise those norms using the XHAIL symbolic machine learning system. Building freely on Sergot's room scenario as a case study, we show how to synthesise norms that can resolve runtime institutional conflicts, and so establish the viability of a method for decentralised agent-directed runtime (online) revision of explicit norms.