Project Grant 2147187

This National Science Foundation Project Grant award of $460,345 supports research into developing normative economic approaches to fairness in artificial intelligence. Funded under the Computer and Information Science and Engineering program (CFDA 47.070), the three-year award to President and Fellows of Harvard College from March 2022 to February 2025 will establish normative economics as a central tool for studying fairness in AI.

The awardee will pursue two research questions. First, reconciling the perspective of normative economics with existing approaches to fairness in AI. Second, rethinking what fairness in AI should be by drawing on normative economics. Theoretical and algorithmic advances will be integrated into refugee resettlement decision systems to examine fairness guarantees and welfare impacts. Research directions include extending observations that parity-based fairness criteria can be Pareto inefficient, as well as conceptualizing classifiers as public resources applying notions of fair division from normative economics. Addressing challenges such as conditions for classifiers in the "core" and algorithms for their computation will generalize findings from training data sets.