Deriving probabilities from causal diagrams — LessWrong