Distributive Justice and Fairness Metrics in Automated Decision-making: How Much Overlap Is There?

The advent of powerful prediction algorithms led to increased automation of\nhigh-stake decisions regarding the allocation of scarce resources such as\ngovernment spending and welfare support. This automation bears the risk of\nperpetuating unwanted discrimination against vulnerable and historically\ndisadvantaged groups. Research on algorithmic discrimination in computer\nscience and other disciplines developed a plethora of fairness metrics to\ndetect and correct discriminatory algorithms. Drawing on robust sociological\nand philosophical discourse on distributive justice, we identify the\nlimitations and problematic implications of prominent fairness metrics. We show\nthat metrics implementing equality of opportunity only apply when resource\nallocations are based on deservingness, but fail when allocations should\nreflect concerns about egalitarianism, sufficiency, and priority. We argue that\nby cleanly distinguishing between prediction tasks and decision tasks, research\non fair machine learning could take better advantage of the rich literature on\ndistributive justice.\n