Fairness mechanisms in AI and multi-agent systems

Interactions constitute the foundations of temporary or stable relationships between agents so a system of interactions can be defined as a set of agents that create such temporary and stable relationships between them. The interactions thus have consequences for the characteristics of the overall system itself. This helps both to understand which interactions should be considered in a system and to simplify the complexity of what we want to analyse. This project deals with the study of fairness in systems of multiple interacting agents, where the interactions create social relations, of power, of value, of wealth, of reputation, of dependence, etc.

Looking at the real world, we say that in the interactions between agents, unfair relationships can arise, with unfair consequences for the single agents and for the overall system. For example, while our society seeks to guarantee a minimum welfare for all citizens, it's not clear how such a society can redistribute resources in a fair way for both individual citizens and the overall economic system. This is a problem arising from the relations between these individuals and the systemic dynamics. It also relates to several aspects of fairness concerning resource allocation and morality. The project is multi-disciplinary including elements of both computer science and social sciences. The following initial research questions arise from these considerations.

- What is required of a decision-making agent to be understood as fair?

- What is required of a group of agents for them to be fair?

- How do we distinguish between and compare fairness measures to determine which measure is most appropriate for a particular interactive system?

- How can we use these concepts and mechanisms to ensure that systems in general (whether machine or human) are fair, and what mitigations might be constructed if not.

In the context of this PhD, we will pursue these questions through the following steps:

- In order to address...