Towards the emergence of norms in normative multiagent systems: Agent-directed norm synthesis using symbolic machine learning

Abstract

The norms in a normative multiagent system (MAS) provide a governance mechanism for the behaviour of participating agents within the system. The set of norms – the institution or normative system – is meant to be stable to facilitate continued gover- nance of these agents over time. To remain fit for purpose, the norms of the institution must evolve to reflect the changing needs of its participants and their environment. Consequently, we put forward a novel framework for the emergence of norms within a normative multiagent system directed by the agents participating in the system. To realise the proposed framework, it is necessary to conceptualise a way of synthesising norms, formulating new norms or revising existing norms, from the experience of par- ticipating agents. While human societies have various mechanisms for revising norms, current normative multiagent systems do not adequately provide similar mechanisms.
Subsequently, we also present a novel computational approach for revising the norms based on requests by participating agents, coined the agent-directed norm synthesis framework. The proposed approach builds on an existing framework, InstAL, for spec- ifying a collection of institutional norms, and a symbolic machine learning system, XHAIL, to revise such normative specifications in response to agent experience data at runtime. It utilises special-purpose synthesiser agents with partial observability to formulate or synthesise new norms or revisions in response to requests by participating agents for changes to the institution. Evaluation of our framework demonstrates that we are able to successfully synthesise norms that can resolve conflict situations reported by the agents.
The success of our agent-directed norm synthesis framework allows for the realisation of decentralised runtime (online) norm synthesis through the use of synthesisers ded- icated to synthesising norms based on participating agents’ requests. The synthesis of the norms in an institution is integral for the realisation of our framework for the emergence of norms within a normative multiagent system directed by the partici- pating agents. In so doing, we pave the way for the observance of a norm emerging in a normative multiagent system where the agents themselves are involved in norm synthesis. Thereby, influencing the norms governing the system in line with two of Ostrom’s eight principles for institutional design: participatory decision-making and defining rules that meet the needs of the local context.

Date of Award	26 Jun 2024
Original language	English
Awarding Institution	University of Bath
Supervisor	Marina De Vos (Supervisor) & Julian Padget (Supervisor)