B1428
Title: Equity by design: Crafting algorithms for fair decision-making
Authors: Madison Coots - Harvard University (United States) [presenting]
Abstract: In today's world, algorithms wield increasingly significant influence over critical aspects of society across a number of contexts, including healthcare, lending, and criminal justice. Increasingly, decision-makers in these areas are turning towards algorithms to improve outcomes and promote equity. The use of algorithmic decision-making holds promise for the enactment of policies that make optimal use of limited resources and distribute them across a population more equitably. However, in applied policy settings, algorithms also have the potential to yield unexpected results, and their design should be driven by considering the outcomes they are likely to produce. The nuances of algorithmic fairness, as well as the potential benefits, are explored to be reaped from thoughtful algorithmic design. With an example grounded in the criminal justice context, well describe an algorithmic framework for the fair allocation of resources that directly anticipates the consequences of the allocation decisions and efficiently maximizes decision-maker utility. A consequentialist approach is also considered in the analysis of the use of race and ethnicity in diabetes risk estimation for the mitigation of disparities in diabetes diagnoses. These examples will underscore the need for foregrounding outcomes in the design of fair algorithms and reveal the potential for algorithms to be used for the advancement of equity across diverse domains.
