Characterizing Fairness in Societal Resource Allocation

Societal biases can lead to disparate impacts on, and treatment of, different demographic groups. This can have substantial effects on the outcomes of public resource allocation scenarios like child welfare, housing allocations for homeless persons, etc. In recent years, there has been an increasing interest in devising algorithms for allocating public resources. However, ensuring the fairness and equitability of these algorithms is challenging since the definition of fairness is highly intersectional, multi-modal, and domain-specific. Moreover, the allocation of these resources is dynamic and timedependent in nature. While there exist several notions of fairness in the Machine Learning (ML) literature, their applicability to Fair Division (FD) of resources is limited. In our research, we aim to bridge the gap between the Fair ML and economic theories of FD for public resource allocation. More specifically, we will study different application areas such as policing, homelessness, and eviction, devise fair algorithms and metrics for these resources, and evaluate their effectiveness in public policymaking.