In just a few years, we have seen artificial intelligence (AI) and machine learning (ML) in healthcare transition from buzzwords to clinical application. AI/ML has permeated not just the frontlines of mammography,1,2 stroke,3 sepsis,4 readmission,5 acute kidney injury,6,7 diabetic retinopathy8,9 and melanoma detection10 but more controversially and less visibly, the murky backlines of health economics like prediction of no-shows11 and healthcare utilization.12

The main theme of existing AI in medicine is diagnosis. Can AI also be used to improve outcomes? In this paper, we explore several such examples where prediction of acute events in prospective trials are used to improve longer term outcomes for patients. We highlight how AI is complementary to traditional statistics. We also discuss the philosophy of AI in clinical studies and how to incorporate this information into our understanding of traditional statistics paradigms that are the foundation of clinical research. We end with a discussion of the future of AI in trials, and explore how AI can move beyond prognostication and decision making into generation of knowledge.