Balancing risk-model scope and granularity

Determine how AI risk models should balance comprehensive coverage of the potential risk universe against sufficiently fine-grained representation of causal pathways within individual risk scenarios.

Background

Frontier-AI risk landscapes contain many possible scenarios, while detailed causal models require substantial data and structural assumptions. Broad models may omit important pathways and underestimate aggregate risk, whereas highly granular models may become difficult to parameterize and overly sensitive to weak evidence. The paper identifies the appropriate balance between coverage and detail as a fundamental unresolved design problem.