Knightian uncertainty

Knightian uncertainty refers, following the distinction drawn by the economist Frank Knight in 1921, to a situation where not only the future outcome is unknown, but the very distribution of the probabilities of possible outcomes is unknowable. Knight thus contrasts risk, where probabilities are known or estimable and which can therefore be insured and priced, with uncertainty, where the absence of any probabilistic basis makes actuarial calculation impossible. This distinction is fundamental for insurance, because the whole industry rests on the ability to turn an unknown into a calculable risk, using historical data and the law of large numbers. When a phenomenon falls under pure uncertainty, for lack of relevant history or because its environment changes faster than data can be collected, it escapes classic actuarial science. The AlgoPolis paper draws precisely on this concept to analyze the insurability of agentic AI, whose failures depend on continuously updated models, unpredictable emergent capabilities and dynamic threats. In this framework, algorithmic risk shifts from the domain of priceable risk toward that of radical uncertainty, which constitutes, according to this analysis, a structural obstacle to its insurability through traditional mechanisms.

An insurer can price mortality, whose probabilities are stable and documented, but is left helpless before the failure of an AI agent whose underlying model changes every month, a situation of Knightian uncertainty.

incertitude radicale, incertitude de Knight, knightian uncertainty