Explainers
Legibility, Steerability, Cooperativeness: The Three Mathematical Scales of AI Safety
The article introduces Lionel Levine's framework for AI safety, which organizes mathematical approaches into three scales: legibility, steerability, and cooperativeness. This framework aims to clarify the diverse mathematical research landscape and highlight the distinct goals and limitations of methods like formal verification.
Meera Iyer·November 15, 2026