Veridical Data Science and Alignment in Medical AI - Building Trust and Interpretability in Healthcare

Learn about the critical intersection of AI alignment, trust, and healthcare through a comprehensive lecture that explores real-world medical case studies. Examine how AI algorithms evolve from transparent to more complex implementations, including tree-based trauma diagnosis methods, LLM-powered emergency department systems, and mechanistic circuits for pathology report data extraction. Discover the Veridical Data Science principles of Predictability, Computability, and Stability (PCS) and their role in building trust and interpretability in medical AI systems. Gain insights into how medical professionals can assess AI alignment and understand the future implications for medical foundation models. Explore the challenges and opportunities in integrating AI systems within healthcare, considering the perspectives of various stakeholders including medical personnel, patients, administrators, public health officials, and taxpayers.