I agree that decision procedures that track counterfactuals are certainly trying to fit an important desideratum. You should pay in Counterfactual Mugging / Counterfactual Prisoner’s Dilemma.
But, on the other hand, it seems to me also an important part to be able to update, that what actually happened was actually more likely to happen than you previously estimated.
For example, you thought this oracle, Omega, was very good, so you a priory reason that you should act in one way. Then you observe Omega making a mistake. Here you have a fork, change your mind. Or decide that this is what you should do in 1/trillion cases and eat that cost.
It’s unclear to me how exactly to integrate that consideration back alongside timeless picture. Just imagine what observations you could get and decide your actions downstream, also in a priory way? 1) it’s unclear to me how exactly to factor that update in 2) you are bounded
I agree that decision procedures that track counterfactuals are certainly trying to fit an important desideratum. You should pay in Counterfactual Mugging / Counterfactual Prisoner’s Dilemma.
But, on the other hand, it seems to me also an important part to be able to update, that what actually happened was actually more likely to happen than you previously estimated.
For example, you thought this oracle, Omega, was very good, so you a priory reason that you should act in one way. Then you observe Omega making a mistake. Here you have a fork, change your mind. Or decide that this is what you should do in 1/trillion cases and eat that cost.
It’s unclear to me how exactly to integrate that consideration back alongside timeless picture. Just imagine what observations you could get and decide your actions downstream, also in a priory way? 1) it’s unclear to me how exactly to factor that update in 2) you are bounded