updateless decision theory

A decision theorydecision theory, n.: The study of how to make the best choice when outcomes are uncertain, by weighing how likely each result is against how much it matters. in which an agent does not condition its choice on what it has observed; instead it selects, from the standpoint of its prior beliefs, the policy — a mapping from possible observations to actions — with the highest expected utility, and then does what that policy prescribes for its actual observations. Treating its decision as the output of an abstract computation that may also run inside predictors, copies or simulations of itself, it reasons about all of these instances together. UDT therefore pays up in 'counterfactual mugging', cooperates with exact copies of itself in a prisoner's dilemmagame theory, n.: The mathematics of strategic decisions, where each player's best move depends on what the others do. It is used to study competition and cooperation among people, firms, nations and AI agents. and takes one box in Newcomb's problem, cases where causal decision theory does worse. Functional decision theory (FDT), proposed by Eliezer Yudkowsky and Nate Soares, generalizes the same move — evaluating an act as the output of the agent's decision function — and counts UDT as an updateless variant. Open problems include defining logical counterfactuals and avoiding 'commitment races' between updateless agents.

An updateless agent pays the predictor, because the policy of paying had the higher expected value before the coin was flipped.