Thesis Defense: Sarah Cen. Title: Paths to AI Accountability: Making AI Fit for Humans Through Design, Incentives, and Evidence

Abstract: Artificial Intelligence (AI) has gained increasing sociopolitical significance over the past decade. In response, there are efforts devoted to understanding the implications of AI's progress and developing AI in a way that is "responsible," "ethical," and "safe." Within this broader context, this thesis studies how we can better integrate AI into a fundamentally human society. We focus on three particular avenues for making AI "fit" for humans. First, we examine ways to design responsible AI from the ground-up through a work on algorithmic fairness. Second, we explore the role of incentives in better aligning humans and algorithms through a game-theoretic model of trustworthy AI. Finally, we discuss the power of evidence in AI accountability through the lens of algorithmic auditing.