Unpacking the Black Box: Regulating Algorithmic Decisions

Explore a lecture on regulating prediction algorithms in high-stakes decision-making contexts. Delve into the challenges of overseeing complex 'black-box' prediction functions, addressing incentive conflicts between algorithm designers and regulatory principals, and navigating limitations in understanding these models. Examine the inefficiency of restricting algorithms to fully transparent models and evaluate the effectiveness of algorithmic audits. Learn about the importance of targeted audit tools that focus on specific misalignment issues, such as excessive false positives or racial disparities, in achieving optimal solutions. Gain insights from empirical evidence in consumer lending that supports the theoretical findings presented.