On the Unlikelihood of D-Separation