On the one hand, more and more decisions are being made based on what machines can learn about us: who gets a loan, who gets into college, who gets insurance, etc. On the other hand, people have many reservations about the fairness of algorithms, about algorithmic perpetuation of biases built into historical data, about the mis- or overinterpretation of statistical correlations, and more. This grant funds work by economists Jens Ludwig from the University of Chicago and Sendhil Mullainathan from Harvard to study when, why, and how people should override recommendations based on artificial intelligence. The team will focus on how New York City judges decide to release or hold suspects before trial. Machine generated recommendations—ones that use facts about a suspect to predict whether that subject will commit a crime if released back into the community—are already in use. But judges are also privy to information about a subject that a typical algorithm is not, including a suspect’s courtroom dress, demeanor, accompanying associates, etc. Ludwig and Mullainathan will study whether and how these additional factors affect both judicial predictions of suspect behavior as well as AI predictions of judicial behavior.

This grant funds efforts by Avi Goldfarb, Joshua Gans, and Ajay Agrawal, three leading economists from the University of Toronto, and Catherine Tucker, Sloan Distinguished Professor of Management at MIT, to facilitate rigorous research on the economics of artificial intelligence (AI). Building on a successful conference on the economics of AI held in Toronto in 2017, the team plans to hold a series of three more annual conferences on related topics, commissioning papers for each conference, then publishing and disseminating the collected conference proceedings. Over three years, the team anticipates commissioning more than 50 academic papers. The team will also organize extensive training, support, and other services for graduate students and postdoctoral fellows interested in studying the economics of AI. The plan is to train more than 90 early-career researchers in advanced methodological and analytic techniques.

This grant provides partial support to the National Academy of Sciences for a consensus report on how to use insights from the behavioral, social, and statistical sciences to improve the reliability and reproducibility of research. The project will include five committee meetings featuring various experts across the sciences, five commissioned papers on reproducibility, an expert panel on behavioral economics and the professional incentives facing producers and consumers of research, and a final consensus report. While the majority of the funding will be provided by NSF, this grant will provide supplementary support both for including economic perspectives in the study and for disseminating the final report.