AJAR: An Argumentation-based Judging Agents Framework for Ethical Reinforcement Learning

An increasing number of socio-technical systems embedding Artificial Intelligence (AI) technologies are deployed, and questions arise about the possible impact of such systems onto humans. We propose a hybrid multi-agent Reinforcement Learning framework consists of learning agents that learn a task-oriented behaviour defined by a set of symbolic moral judging agents to ensure they respect moral values. This framework is applied on the problem of responsible energy distribution for smart grids. CCS CONCEPTS • Computing methodologies → Multi-agent reinforcement learning.