Optimizing Society? Ensuring Fairness in Automated Decision-Making (Invited Talk)

Abstract

Algorithms are increasingly used to make high-stakes decisions about people; who goes to jail, what neighborhoods police deploy to, and who should be hired for a job. But if we want these decisions to be fair, this means we must take societal notions of fairness and express them using the language of math. What is a fair decision and how can it be guaranteed?
In this talk, we'll discuss recent work from the new and growing field of Fairness, Accountability, and Transparency. We will examine technical definitions of fairness and non-discrimination that have been proposed and their societal counterparts. We'll also discuss methods for ensuring that algorithms are making decisions as desired, from methods to audit black-box algorithms to white-box interpretability techniques. This important field necessitates societally informed and mathematically rigorous work; we'll discuss open problems in this light.