Related: "In general a machine learning system is built and trained to optimize a specified target objective: classification accuracy in a spam filter or tumor diagnostic, efficiency in route planning or Amazon box packing. Unlike these precise performance metrics, the criteria of safety, trust, and nondiscrimination often cannot be completely quantified. How then can models be trained towards these auxiliary objectives?"

From "Interpreting AI Is More Than Black And White": https://www.forbes.com/sites/alexanderlavin/2019/06/17/beyon...