Decision Theory without Logical Omniscience: Toward an Axiomatic Framework for Bounded Rationality (Q4262876)

An agent is said to be logically omniscient if he knows all logical implications of his knowledge. The paper starts with the following observation: ``To use states of the world forces us to make one of two assumptions. Either we must assume that the agent knows all logical implications of his knowledge or we must assume that some of the states of the world are logically inconsistent''. The author proposes an axiomatic approach that endogenizes the state space by making the impossible worlds (i.e. states that are logical impossible but are used to model what the agent considers possible) part of the representation of the agent's preferences. Normally in decision models the primitive ``is a single ex ante or `informationless' preference relation. From this, one derives preferences conditional on information under some assumptions on how the agent responds to information.'' In the author's approach ``the various preferences are the primitives and it is a characterization of the relationship between them that is to be derived''. The main results are necessary and sufficient conditions for deriving a subjective state space from the agent's preferences as well as for representing the agent's preferences via expected utility on this state space.