I sometimes like to think that Bayesian causal networks should be drawn directly onto spacetime diagrams, with their edges timelike curves. This automatically enforces some causal structure. But if you care about things whose future lightcone doesn’t meet yours then this becomes impossible, your utility function is (in once sense of the word) causally effected by something that can’t (in the other sense of the word) causally effect it.
I sometimes like to think that Bayesian causal networks should be drawn directly onto spacetime diagrams, with their edges timelike curves. This automatically enforces some causal structure. But if you care about things whose future lightcone doesn’t meet yours then this becomes impossible, your utility function is (in once sense of the word) causally effected by something that can’t (in the other sense of the word) causally effect it.