Inference of Graphical Causal Models: Representing the Meaningful Information of Probability Distributions