A Principled Approach To Trustworthy Machine Learning