If Predictors are to be modeled as accuracy-maximizing agents, they could acausally cooperate with each other, so that one Predictor optimizes its accuracy about a world where it's absent through controlling the predictions of another Predictor that is present in that world.