Veridical Data Science for Healthcare in the Age of AI

Learn about veridical data science principles and their application to healthcare AI through this lecture by Bin Yu from UC Berkeley at the Simons Institute. Discover how human judgment introduces hidden uncertainties throughout the data science lifecycle that extend beyond traditional sampling variability and contribute to AI-related risks. Explore the three fundamental principles of veridical data science—Predictability, Computability, and Stability (PCS)—designed to make uncertainties explicit and assessable while aggregating reality-checked algorithms for improved outcomes. Examine how the PCS framework unifies and extends established best practices in statistics and machine learning through practical healthcare applications, including identifying genetic drivers of heart disease, reducing prostate cancer detection costs, and enhancing uncertainty quantification methods that go beyond standard conformal prediction approaches.