Existing regulations on powerful AI capabilities are narrowly focused on “foundation” or “frontier” models, but these terms are vague and inconsistently defined, leading to an unstable foundation for governance. Critically, policy debates often overlook the data used with these models, despite the clear link between data and model performance.

In his recent paper, “Data-Centric AI Governance: Addressing the Limitations of Model-Focused Policies,” Ritwik Gupta and his colleagues highlighted the importance of considering dataset size and content as key factors in assessing the risks posed by AI models, both now and in the future. They also emphasise the dangers of reactive over-regulation and propose a path towards a more careful, quantitative evaluation of AI capabilities, which could lead to a simplified yet effective regulatory environment.

This event is organised by the Oxford Martin AI Governance Initiative.