Abstract

A prominent topic in contemporary AI ethics is the concern that advanced machine learning systems may become unexplainable ‘black boxes’. This has led policymakers to propose regulating the use of machine learning in high-risk contexts, such as medicine. Machine learning researchers have sought to address this concern by developing methods and tools for making AI systems more ‘explainable’ or ‘interpretable’, a field known as Explainable AI (XAI). However, it is often unclear what kinds of explanations are needed for a given context. In this talk, I outline a philosophical framework for thinking about these questions and apply it to a medical case study. I argue that different kinds of explanations that are needed for different audiences, such as medical researchers, healthcare policymakers and clinicians.