Researchers from MIT and Empirical Health used 3 million person-days of Apple Watch data to train a new AI foundation model that can predict medical conditions from incomplete wearable data. The model, called JETS, builds on Yann LeCun’s JEPA architecture, which learns the meaning of missing data rather than trying to reconstruct it.
Trained mostly without labeled medical histories, JETS showed strong predictive performance — including 86.8% AUROC for high blood pressure and solid results for heart rhythm disorders and chronic fatigue syndrome. The study highlights how advanced “world model”–style AI can unlock valuable health insights from everyday wearable data, even when readings are irregular or incomplete.
Source: 9to5Mac
