Multi-agent systems study environments that consist of a number of agents who interact with each other to achieve individual and system goals. These agents could be software agents that reason about and execute programmed actions; they can also be humans with their own goals and motivations. These agents negotiate, cooperate, and compete with each other, to achieve their own goals. A normative multi-agent system consists of a set of rules/norms and values that guide, control, and regulate the expected behaviour of agents. Recent years have witnessed the dynamic development of normative multi-agent systems, from defining and computing the norms to norm selection and norm emergence.
However, the research on normative multi-agent systems has not yet benefited from another well-established area: game theory. Designing norms that satisfy environmental requirements is, in spirit, similar to mechanism design. Game theory and mechanism design comprise rich studies in incentives, fairness, equilibria, and welfare maximization. Despite their similarities, normative systems have not yet taken advantage of these studies. This project aims to investigate normative multi-agent systems through a game theoretic lens by adapting game theory methodologies to a normative setting. These include, but are not limited to, the way agents are modelled, the representation of a normative multi-agent system, the format of agent interactions, and the evolution of the systems. In particular, the project will examine the extent to which the agents’ risk attitudes affects their decision-making processes and the effects of norm computation and selection on the fairness of the systems. The project will implement and test the designs and analysis in application areas such as transportation, logistics, and energy consumption, with the novel addition of societal and environmental aspects, informed by the rich features of Sustainability Development Goals, while accounting for and embedding responsible innovation characteristics, to guide the creation of systems that incorporate their total cost and minimize externalities.
The outcome of the PhD project will be in the form of accountable and explainable artificial intelligence tools for agent modelling and autonomous system evaluation, in alignment with the ideas set out at the end of the previous paragraph.
Successful applicants will have, or expect to receive, a master's degree or first or upper-second bachelor's degree in computer science or other related subjects, and an interest in game theory and multi-agent systems.
Formal applications should be accompanied by a research proposal and made via the University of Bath’s online application form. Further information about the application process can be found here. Enquiries about the research should be directed to Dr De Vos.
Start date: Between 8 January and 30 September 2024.