This paper discusses some procedures developed in recent work in machine learning for inferring causal direction from observational data. The role of independence and invariance assumptions is emphasized. Several familiar ...