Explore the challenges and opportunities of applying machine learning to healthcare in this 55-minute conference talk by Isaac Kohane from Harvard University. Delve into the limitations of current approaches to medical decision support systems and discover why increased interpretability alone is insufficient. Examine the under-appreciated assumptions in applying machine learning to patient care and understand how these define a crucial research agenda for shared human-ML decision making in medicine. Learn about the impressive successes in medical image analysis and the potential pitfalls of broadly applying these methods across patient encounters. Gain insights into the necessary steps for developing more effective and trustworthy AI systems in healthcare that go beyond interpretability and require active human-machine interactions.