Knightian uncertainty
A situation where the probability distribution itself is unknown, as opposed to calculable risk.
Frank Knight published Risk, Uncertainty and Profit in 1921 and separated two things everyday language runs together. Risk has known or estimable probabilities: it can be insured and priced. Uncertainty has none. The outcome is unknown, the very distribution of possible outcomes is unknowable, and actuarial calculation loses its grip. All of insurance rests on turning a hazard into a calculable risk, using history and the law of large numbers. A phenomenon with no relevant history, or whose environment changes faster than data can be gathered, falls outside classic actuarial science. The AlgoPolis paper applies the concept to agentic AI, whose failures depend on continuously updated models, unpredictable emergent capabilities and shifting threats. Algorithmic risk slides there from priceable risk toward radical uncertainty, which in that analysis is a structural obstacle to insuring it through traditional mechanisms.
Textbook case: an insurer can price mortality, whose probabilities are stable and documented. It is helpless before the failure of an AI agent whose underlying model changes every month, which is Knightian uncertainty.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.