No, it doesnt work that way. By projecting real, complex decisions down to some low dimensional set of criteria, you automatically throw away information and ge

No, it doesnt work that way. By projecting real, complex decisions down to some low dimensional set of criteria, you automatically throw away information and generalize. It's only fair in the sense that the criteria is the same, but it's necessarily based on some kind of average outcome and so effectively sacrifices outliers. Its is still extremely unfair to inputs (who are actual people in the cases we mean) that exhibit characteristics that reliably fit the prediction on average, but that don't conform to the prediction.