Multiverse-wide cooperation in a nutshell

Abstract

MSR proposes that agents should cooperate in prisoner’s dilemma-like situations, even without communication, based on the concept of superrationality. This concept argues that if agents reason similarly and face similar problems, they are likely to reach the same conclusion, leading to a symmetric outcome. Since cooperation yields the best outcome for both parties among the symmetric outcomes, superrationality suggests that cooperating is the best course of action. While this might seem like a “shady action at a distance,” it’s based on the idea that deterministic reasoners following the same decision algorithm in highly similar situations are unlikely to reach diametrically opposed conclusions. The similarity of the decision situation is crucial; it refers to the decision-relevant variables, such as the payoffs expected from different options. Factors that are unlikely to affect the decision, like the color of a button, are irrelevant. The key is that both agents should anticipate similar payoffs for their choices.