Aarhus University Seal / Aarhus Universitets segl

Lessons from nuclear arms control for the responsible governance of AI development

Matthijs Michiel Maas, University of Copenhagen, Denmark

New technologies, particularly those which are perceived to offer a significant competitive edge, frequently disrupt previously stable international governance arrangements or power distributions. Artificial Intelligence is one such critical technology. How can one design stable multilateral governance arrangements, which can minimize the risks, when these technologies are seen to offer strong unilateral strategic advantages to parties. Drawing a parallel with the historical experience of nuclear arms control, suggests that (1) arms races are not inevitable, but can be slowed or even averted; (2) small communities of experts, appropriately organized and mobilized, can have a disproportionate policy-framing impact; (3) AI systems, like other tightly coupled, opaque systems, will always generate ‘normal accidents’, ensuring that perfect safety will remain categorically out of reach.