Courtney M. Cox (Fordham University School of Law) has posted Non-Herculean Data: A Philosophical Intervention in a Technical Debate about Judicial Opinions as Data Sources (Proceedings of the 20th International Conference on Artificial Intelligence and Law (forthcoming 2025)) on SSRN. Here is the abstract:
This Article is a philosophical intervention in technical debates about legal AI, and specifically, about using judicial opinions in training legal AI. The intervention is this: Judges sometimes have what is called "normative uncertainty"—despite knowing all the facts and laws, they remain uncertain about how they ought decide in light of them. When judges respond rationally to this type of uncertainty, the reasons they give in their opinions will not be their actual reasoning for the decision. As a result, opinions are, at best, an incomplete representation of judicial reasoning even in ideal cases. This phenomenon has important implications for the design of legal AI tools: Taking opinions at face value threatens both to systematically underestimate the uncertainty in the system and to misinterpret the law. Although the primary contribution of this piece concerns the implications for opinions-as-data, two additional contributions are made: clarifying a confusion in the AI literature about normative uncertainty and providing a new way to model judicial decisionmaking for use in computational treatments of law.
Recommended.