CAREER: Beyond Fair Algorithms: Mathematical Foundations for Long-Term Fairness with Humans in the Loop

Algorithmic and machine learning tools are heavily involved in high-stakes, life-altering decision-making. Yet, previous research has shown that these algorithmic tools can induce inaccuracies and cause discrepancies in desired outcomes. The traditional approach to this problem is to independently fix each given decision-making tool to make it accurate. While valuable, this view fails to reason about the long-term impact on the complex socio-technical systems involved. This project will provide novel paradigms and foundations to adopt a long-term and in-context view of accuracy within these systems. The research will take the academic study of fairness a step closer to practical issues. This will lead to significant longer-term societal benefits. The project will support and promote education about responsible AI and machine learning and will enhance participation through workshops and mentoring activities. The project combines tools from game theory, optimization, and machine learning to contribute new advancements to our scientific understanding of algorithmic fairness. It does so along three axes: First, it models how humans can strategically respond to high-stake machine learning algorithms and aims to understand the impact of such strategic behavior on algorithmic fairness. Second, it proposes new modeling paradigms and approaches to algorithmic fairness interventions for complex decision-making pipelines comprising many inter-connected stages. Third, it reasons about feedback loops and their long-term effects. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.