Ex Post Policing and Ex Ante Regulation
Oren Bar‐Gill, Cass Robert Sunstein · 2025
Abstract Chapter 10 considers policies that target the design and implementation of AI-powered algorithms—either through ex post policing of algorithmic outcomes or through ex ante regulation of the algorithm’s code. This chapter develops an output-based approach that follows closely in the footsteps of disparate impact doctrine. That doctrine allows courts to focus on outcomes—worse outcomes for women or minorities—without conditioning liability on a showing of discriminatory intent (as a decision-making input). In the algorithmic context, the idea is to evaluate the algorithm’s decisions, or outcomes, and target scrutiny toward cases where consumers with a protected characteristic are treated differently. The disparate-impact approach has been developed in the context of discrimination based on race or sex as protected characteristics. This chapter proposes to extend the current approach to discrimination based on imperfect information or imperfect rationality as protected characteristics.