Search for a command to run...
Abstract Artificial intelligence (AI) is emerging as one of the most transformative technologies of our time and policymakers are looking for ways to govern its responsible development and deployment. One promising approach to AI governance focuses on leveraging ‘compute’—the computing power, generated by specialized chips, hardware, and infrastructure needed to develop and deploy advanced AI systems. This article explains the role of compute in AI development and deployment, and the specific qualities that make it an attractive governance node for policymakers. Compute plays a critical role across the AI life cycle, underpinning both the development and deployment of advanced models. Two key qualities make compute a promising node for AI governance: Compute is governable due to detectability, excludability, quantifiability, and concentrated supply chain, and it is also effective at targeting regulations, as compute usage also correlates strongly with AI capabilities. Focusing on compute enables three critical governance capacities: visibility into AI development and deployment through monitoring compute resources; compute allocation to direct and influence AI development trajectories; and enforcement mechanisms to respond when rules are violated. Using compute for AI governance is not without its challenges. These include the potential for technological advancements to change the relationship between compute and AI capabilities, difficulty distinguishing AI-specific compute from general-purpose hardware, and potential risks to privacy, innovation, and market competition.