Algorithmic Fairness is More Than a Math Problem

Host Kirsten Martin is joined by Ben Green, an assistant professor at the Gerald R. Ford School of Public Policy and a postdoctoral scholar in the Michigan Society of Fellows at the University of Michigan. Specializing in the social and political impacts of government algorithms, with a focus on algorithmic fairness, smart cities, and the criminal justice system, Ben is also an affiliate of the Berkman Klein Center for Internet & Society at Harvard University and a fellow of the Center for Democracy & Technology.

He came on the show to talk about his paper “Escaping the Impossibility of Fairness: From Formal to Substantive Algorithmic Fairness,” which recently appeared in Philosophy & Technology.

Ben begins by explaining the aforementioned “impossibility of fairness,” an idea that describes the incompatibility of different mathematical notions of what makes a system fair. By focusing on meeting one of these formal definitions of fairness, an algorithm that is mathematically “fair” can nevertheless yield decisions that re-entrench real-world injustices, including those it may have been designed to counter.

Asking whether the ultimate purpose of an algorithm is to satisfy a mathematical formalism or rather improve society, Ben puts forward an alternative notion of what he calls substantive algorithmic fairness—his detailed diagram of which, labelled Figure 2 in the paper, made a lasting impression on Kirsten. His approach still envisions a role for mathematical conceptions of fairness, but it repositions them as one consideration in a broader process where the primary concern is accounting for and mitigating both upstream inequalities that exist before an algorithm is deployed and downstream harms present afterwards.