An Indirect-Evolution Approach to Newcomb's Problem

Players from two populations, predictors and predictees, are randomly matched in a game--theoretic version of Newcomb's Problem. Predictors are able to predict the predictees' choices by observing their type. There are two types of predictees, those who take their predictability into account by using the Backtracking Principle when calculating expected utilities, and those who ignore their predictability by using the Disconnection Principle. Backtrackers are one--boxers, the others are two--boxers. Given predictability, evolution favors the Backtracking Principle. An explicit causal analysis proves that this result does not rest on unusual causal assumptions.