Applying multi-agent reinforcement learning to solve sequential moral dilemmas

Incorporation of ethical values in the field of Artificial Intelligence is inevitable. With the rapid development of technologies capable of making autonomous decisions, more attention should be dedicated to the process of learning rules that apply in the real world. Sequential Moral Dilemmas occur when an agent can gain more by pursuing its individual goals rather than acting ethically. This work aims to define such a dilemma with the use of the mathematical framework described in "A Structural Solution to Sequential Moral Dilemmas", and approaches solving it with the help of Multi-Agent Reinforcement Learning. A relevant case study has been devised with two environments prepared to investigate how ethical values could impact the redistribution of resources and ultimately answer whether employing them leads to lowering the level of starvation in the community of artificial agents