Justin Key Canfil (Carnegie Mellon University) has posted Learning to Disagree: Descriptive and Prescriptive Convergence in AI Governance on SSRN. Here is the abstract:
How do international institutions govern technologies whose nature remains unsettled? Existing theories of deliberative fora either treat international institutions as epiphenomenal to material interests or as engines of socialization that generate norm convergence through persuasion. Both, however, assume convergence is unidimensional. We distinguish between descriptive convergence-agreement on what a technology is and what it means-and prescriptive convergence, agreement on how it should be governed. Using an original corpus of all diplomatic commentary on military artificial intelligence (AI) at the Convention on Certain Conventional Weapons (CCW), we show that states have increasingly converged on definitions and conceptual vocabularies even as disagreements over regulation have intensified. Enduring divisions are structured less by persuasion than by durable strategic and technological qualities, which have come into sharper focus as discussions have matured. We conclude that “weak” multilateral institutions like the CCW are neither engines of norm change nor empty venues, but rather vehicles for collective sense-making in technically uncertain environments. Our findings refine existing theories of institutional effectiveness by showing that agreement about problems and agreement about solutions are analytically distinct outcomes that institutions may facilitate independently.
Highly Recommended!