AI4Society Seminar - Emily Black - On Generative AI Harms: Evaluating them, and Relevant Law

Talk Abstract: In this talk, I’ll present my recent work on technical pitfalls and legal tensions around the evaluation of GenAI harms. Through four case studies, I’ll show how misalignment between regulatory goals and fairness testing techniques can lead to regulation that admits discriminatory behavior. For example, I’ll show how different forms of GenAI evaluation instability—for example, instability in observed discrimination over multi-turn interactions–exacerbate existing regulatory challenges such as creating reliable evidence of discrimination testing and mitigation, and tensions between developer and deployer responsibilities. In addition, I’ll discuss ongoing work showing how some long-standing consumer protection laws might be a useful tool in preventing harms from GenAI hallucinations. Finally, time permitting, I’ll discuss ongoing work with prior EEOC chairs Charlotte Burrows and Jenny Yang, as well as Pauline Kim (Washington University – St Louis Law School) on the legality of common debiasing techniques in AI and GenAI systems.

Speaker Bio: Emily Black is an Assistant Professor of Computer Science and Engineering at New York University. Her research concerns fairness and accountability in AI systems. In other words, she creates methods to determine whether AI systems will cause harm to the public, studies the equity impacts of AI systems in high-stakes settings, such as the government, and connects her own and related research to the legal and policy worlds to help better regulate AI systems. Professor Black’s work is interdisciplinary as she aims to prevent harm from AI systems used in a variety of contexts: she works with lawyers, accountants, civil society advocates, and others to try to prevent algorithmic harm in practice.