Can Statistical Infrastructure Help Govern AI Compute?

Brief: Many proposals for compute governance depend on knowing where advanced chips are produced, sold and deployed. However, existing trade statistics are unable to provide a consistent account. Relevant product codes often combine advanced AI accelerators with much broader categories. Existing work has estimated chip smuggling, mapped semiconductor trade and used mirror statistics to study sanctions circumvention. This project would ask whether the classification systems themselves are fit to support compute-governance measures.

The project would first map current statistical codes against the AI-compute supply chain and recent export-control categories. Secondly, it would assess the implementation strategy for updated codes. Finally, the project would assess which problems arise from weak data collection, which arise from insufficient code granularity, and which cannot be resolved through statistical systems alone.

Outputs: A map of classification gaps across the compute supply chain, and a paper assessing whether existing statistical infrastructure can support compute accounting/verification. Stretch goal: a policy memo.

This project targets the accounting layer in compute governance. If statistical classification systems can be made granular enough to track AI accelerators, they provide low-cost and institutionalised infrastructure for compute accounting. Such a system would support export-control enforcement and verification of international agreements. If not possible, then establishing this redirects effort toward purpose-built alternatives. Either answer improves the field's ability to design enforceable compute governance.

Mentees will lead on desk research, writing/drafting, and expert interviews where appropriate. Mentees will have substantial autonomy to pursue additional research directions within the project's scope. We'll hold a weekly meeting covering drafts, research taste, and mentee development (career planning, network building), with brief async updates between meetings.

My aim is to facilitate a project and devision of labour similar to what you may experience in an AI Safety fellowship such as Pivotal, LASR, or ERA.

Required:

• Strong analytical writing
• Comfortable with self-directed desk research
• Able to commit ~10 hours/week for the full programme period, including a ~weekly call.
• Broad familiarity with AI governance, with strong insights into why compute is a governance lever

Ideal:

• Proficient in Python or R: able to independently clean, merge, and analyse messy tabular datasets (e.g. UN Comtrade, Eurostat). No ML experience needed.
• Prior exposure to at least one of: trade/customs data, official statistics, export controls, or supply-chain analysis
• Willingness to read dry primary material closely (classification manuals, CN explanatory notes, WCO documentation). I expect this project to reward patience with bureaucratic detail.

I expect to be interested in mentees from a wide-variety of backgrounds.