Special Seminar in CMS and HSS

Automated decision-making tools are increasingly in charge of making high-stakes decisions for people—in areas such as education, credit lending, criminal justice, and beyond. These tools can exhibit and exacerbate certain undesirable biases and disparately harm already-disadvantaged and marginalized groups and individuals. In this talk, I will illustrate how we can bring together tools and methods from computer science, economics, and political philosophy to define, measure, and mitigate algorithmic unfairness in a principled manner. In particular, I will address two key questions:

• Given the appropriate notion of harm/benefit, how should we measure and bound unfairness? Existing notions of fairness focus on defining conditions of fairness, but they do not offer a proper measure of unfairness. In practice, however, designers often need to select the least unfair model among a feasible set of unfair alternatives. I present (income) inequality indices from economics as a unifying framework for measuring unfairness--both at the individual- and group-level. I propose the use of cardinal social welfare functions as an alternative measure of fairness behind a veil of ignorance and a computationally tractable method for bounding inequality.
• Given a specific decision-making context, how should we define fairness as the equality of some notion of harm/benefit across socially salient groups? First, I will offer a framework to think about this question normatively. I map the recently proposed notions of group-fairness to models of equality of opportunity. This mapping provides a unifying framework for understanding these notions, and importantly, allows us to spell out the moral assumptions underlying each one of them. Second, I give a descriptive answer to the question of "fairness as equality of what?". I mention a series of adaptive human-subject experiments we recently conducted to understand which existing notion best captures laypeople's perception of fairness.