Beyond a piecemeal approach: prospects for a framework convention on AI

Abstract

Solving many of the challenges presented by artificial intelligence (AI) requires international coordination and cooperation. In response, the past years have seen multiple global initiatives to govern AI. However, very few proposals have discussed treaty models or design for AI governance and have therefore neglected the study of framework conventions–generally multilateral law-making treaties that establish a two-step regulatory process through which initially underspecified obligations and implementation mechanisms are subsequently specified via protocols. This chapter asks whether or how a Framework Convention on AI (FCAI) might serve as a regulatory tool for global AI governance, in contrast with the more traditional piecemeal approach based on individual treaties that govern isolated issues and have no subsequent regime. To answer these questions, the chapter first briefly sets out the recent context of global AI governance, and the governance gaps that remain to be filled. It then explores the elements, definition, and general role of framework conventions as an international regulatory instrument. On this basis, the chapter considers the structural trade-offs and challenges that an FCAI would face, before discussing key ways in which it could be designed to address these concerns. We argue that, while imperfect, an FCAI may be the most tractable and appropriate solution for the international governance of AI if it follows a hybrid model that combines a wide scope with specific obligations and implementation mechanisms concerning issues on which states already converge.