Optimising multi-value alignment: a multi-objective evolutionary strategy for normative multi-agent systems

Maha Riad, Vinicius Renan de Carvalho, Fatemeh Golpayegani

Abstract: Value alignment in normative multi-agent systems (NorMAS) is used to promote a certain value and to ensure consistent behaviour of agents in autonomous intelligent systems with human values. Although various techniques were proposed in the literature for addressing the value alignment challenges in NorMAS, aligning multiple values remains an area requiring further investigation, especially in heterogeneous systems where agents may have different, potentially conflicting, coherent or unrelated norms and values. In such systems, it is crucial to simultaneously compromise and optimise values while synthesising an optimally aligned set of norms. Moreover, to deal with the conflicting or unrelated values and norms, we need to consider the norms and values as independent distinct sets. This research proposes a novel framework, Norms Optimisation and Values Alignment (NOVA), which enables multi-value alignment in heterogeneous NorMAS using parametric norms, multi-objective evolutionary algorithms (MOEAs) and decentralised reasoning. NOVA models the values and norms as independent distinct sets, then formalises the problem as a multi-objective optimisation problem (MOP), and optimises these two sets simultaneously while aligning them. To understand various aspects of this complex problem, several evolutionary algorithms were used to find a set of optimised norm parameters considering two tax scenarios with two and five values. The results show the impact of the selected evolutionary algorithm on the solution, and the importance of understanding the relation between values and norms when prioritising them using different reasoning strategies.

External IDs:dblp:journals/nca/RiadCG25