Combining Normative Ethics Principles to Learn Prosocial Behaviour

Prioritising citizens in the design and development of multi- agent systems (MAS) is key to ensuring that MAS are socially benefi- cial. To ensure that artificial agents within MAS act in citizen-centric ways, agents should foster prosociality, defined as behaving in ways that support the well-being of others. Principles from normative ethics—the philosophical study of morality—can be operationalised in the decision- making capacities of agents to discern ethically acceptable actions and promote prosocial behaviour. However, challenges exist in operational- ising principles: (1) individual principles may be unintuitive; (2) while incorporating multiple principles mitigates issues with individual princi- ples, conflicts may arise between them. We present PriENE, a method for combining multiple principles in individual decision-making to encourage agents learning prosocial behaviour. We evaluate PriENE in a simulated berry harvesting scenario. Interestingly, preliminary results show that societies of PriENE agents do better along metrics one might expect individual principles to have an advantage: one might expect egalitari- anism to minimise inequality, but PriENE societies minimise inequality further; one might expect maximin to have highest minimum experience, but PriENE societies raise minimum experience further; one might ex- pect utilitarianism to have highest cumulative experience, but cumulative experience is further increased in PriENE societies.