Moving from correlation to causation is especially important when it comes to understanding the conditions under which a machine learning model may fail, how long we can expect it to continue ...