[edit]
The Human Values Project
Proceedings of the 4th Machine Learning for Health Symposium, PMLR 259:14-18, 2025.
Abstract
Alignment of AI models to ensure that they are safe and useful decision-aids or decision-makers in human society is close to the top of the technical concerns of many if not most major AI deployment efforts. Here I explore a class of categorical decisions, triage, for which AI models are already being used in medicine. I use this to motivate the urgent need for an international Human Values Project (HVP) that will be at least as hungry for empirical data as any existing international project. A major component of the HVP will be the Clinical Decision Dynamics Study that captures descriptive and normative decisions across a wide range of clinical context and from highly diverse, lay and professional perspectives. Along the way, there are important preliminary studies that the AI (and medicine) community should embrace.