Mathematical Notions vs. Human Perception of Fairness: A Descriptive\n Approach to Fairness for Machine Learning

Abstract

Fairness for Machine Learning has received considerable attention, recently.\nVarious mathematical formulations of fairness have been proposed, and it has\nbeen shown that it is impossible to satisfy all of them simultaneously. The\nliterature so far has dealt with these impossibility results by quantifying the\ntradeoffs between different formulations of fairness. Our work takes a\ndifferent perspective on this issue. Rather than requiring all notions of\nfairness to (partially) hold at the same time, we ask which one of them is the\nmost appropriate given the societal domain in which the decision-making model\nis to be deployed. We take a descriptive approach and set out to identify the\nnotion of fairness that best captures \\emph{lay people's perception of\nfairness}. We run adaptive experiments designed to pinpoint the most compatible\nnotion of fairness with each participant's choices through a small number of\ntests. Perhaps surprisingly, we find that the most simplistic mathematical\ndefinition of fairness---namely, demographic parity---most closely matches\npeople's idea of fairness in two distinct application scenarios. This\nconclusion remains intact even when we explicitly tell the participants about\nthe alternative, more complicated definitions of fairness, and we reduce the\ncognitive burden of evaluating those notions for them. Our findings have\nimportant implications for the Fair ML literature and the discourse on\nformalizing algorithmic fairness.\n