Theoretical Omniscience: Old Evidence or New Theory

C. R. Martins, André (2005) Theoretical Omniscience: Old Evidence or New Theory.

Abstract

I will show that, in the Problem of Old Evidence, unless a rational agent has a property I will call theoretical omniscience (a stronger version of logical omniscience), a problem with non-commutativity of the learning theories follows. Therefore, scientists, when trying to behave as close to rationality as possible, should behave in a way close to the counterfactual strategy. The concept of theoretical omniscience will be applied to the problem of Jeffrey conditionalization, as an example, and we will see that a more complete theoretical model can provide a classical conditionalization where you can learn that data was wrong and all you will not unlearn is your memory.