Ethics by Agreement in Multi-agent Software Systems

Most attempts at inserting ethical behaviour into autonomous machines adopt the ‘designer’ approach, i.e., the ethical principles/behaviour to be implemented are known in advance. Typical approaches include rule-based evaluation of moral choices, reinforcement-learning, and logic based approaches. All of these approaches assume a single moral agent interacting with a moral recipient. This paper argues that there will be more frequent cases where the moral responsibility for a situation will lie among multiple actors, and hence a designed approach will not suffice. We posit that an emergence-based approach offers a better alternative to designed approaches. Further we outline one possible mechanism by which such an emergent morality might be added into autonomous agents.