Andrea Dean (Georgetown University Law Center) has posted Performance-Based AI Governance: Concept, Critique, and Alternatives on SSRN. Here is the abstract:
This Paper defines performance-based AI governance, a regulatory model that evaluates and permits AI systems based on their technical performance—such as accuracy, fairness, or robustness—rather than their institutional purpose or normative legitimacy. The Paper argues against this model, finding that this regulatory orientation suffers from structural problems: it relies on metrics that rapidly lose relevance, reframes value-laden policy questions as technical ones, and shifts regulatory power to industry actors, eroding democratic oversight in the process. The Paper uses case studies to explore performance-based AI governance: the bans on facial recognition use by law enforcement; the European Union’s AI Act; and Large Language Model benchmarks. Two alternatives are offered: governing on principle (identifying and basing regulation on durable societal principles, values and rights), and considering technical invariants of the AI systems. In this vein, the Paper introduces the concept of AI Independence, based on software independence in the voting realm, as a design constraint for important systems so that they do not rely on AI.
Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum