Fairness considerations arise naturally in many algorithmic problems involving the allocation of resources, opportunities, or outcomes. In fair division, the focus is on allocating resources among agents according to fairness criteria such as envy-based or share-based guarantees. In algorithmic fairness, the focus is on the outcomes of automated decision-making procedures, and on whether these outcomes treat different indi-viduals or groups fairly. Although these areas have developed largely within different communities, they share several underlying themes: the choice of appropriate fairness notions, the design of algorithms that satisfy them, and the role of information and incentives in shaping the final outcome.

There are several natural connections between the two areas. A positive classification outcome can often be viewed as the allocation of a desirable opportunity, especially when such opportunities are scarce, as in school admission, loan approval, or access to public services. Similarly, group fairness has a close concep-tual relation to share-based fairness, while individual fairness is related in spirit to notions that compare the treatment of different agents. Some recent works have already begun to explore such analogies, for example by applying envy-freeness-inspired ideas to classification problems or by incorporating preferences into algorithmic fairness notions. Still, these connections remain only partially understood, and we believe that a more systematic interaction between the two communities can be highly fruitful.

A second central theme is incentives. In both fair division and algorithmic fairness, the individuals affect-ed by the outcome may try to influence it in their favor. In fair division, this leads to the classical tension between fairness and truthfulness. In algorithmic fairness, similar issues arise in strategic classification, performative prediction, and incentive-aware learning, where individuals may change their reported infor-mation or observable features in response to the deployed algorithm. In both settings, imposing fairness together with incentive constraints is a demanding task, often leading to impossibility or hardness results. At the same time, recent progress suggests that positive guarantees may be achievable under suitable re-strictions, in online models, or with the help of predictions.

Given these developments, the main objective of this Dagstuhl Seminar is to bring together researchers from Economics and Computation and from Fairness in Machine Learning in order to discuss common challenges at the intersection of fair division, algorithmic fairness, and mechanism design. The plan for the seminar is to focus on three main categories of research topics: a) the interplay between fair division and algorithmic fairness, including the relation between envy-based, share-based, group-based, and individual-based fairness notions; b) fairness in mechanism design, including fair and truthful mechanisms, online aspects, and the limitations imposed by strategic behavior; and c) data-driven fairness, including learning-augmented fair division, strategic classification, performative prediction, and other incentive-aware learn-ing models.

To conclude, we expect the seminar to create a productive interaction between two active research com-munities that have so far had only limited overlap. We hope that the discussions will help identify shared structures, clarify common trade-offs and impossibility phenomena, and formulate new research directions at the intersection of fairness, incentives, and data-driven decision making. We also hope that the seminar will lead to new collaborations among the participants, and to a clearer research agenda for studying fair algorithmic systems in settings where incentives and information play a central role.

Creative Commons BY 4.0

Giorgos Christodoulou, Alon Eden, Kurt Mehlhorn, and Goran Radanovic