Credible, Verifiable, Trustworthy Health AI: from hypothesis generation to clinical validation

Keynote presented at the AI in Precision Medicine Symposium, University of Alabama at Birmingham, 28 October 2026. AI increasingly helps us generate predictions, propose explanations, and decide what to investigate next. This talk connects those capabilities to a question that matters for precision medicine: what would justify relying on an AI system for a particular patient? The recurring distinction throughout is between generating an interesting possibility and establishing what that possibility can support. Scientific abundance motivates a cycle of prediction, justification, empirical testing and intended-use validation. Mathematical proof checking illustrates what a precise verification environment looks like. TESSERA connects graph-based mechanistic hypotheses to questions of evidence and experimental testing. AI co-scientists and cognitive robotics extend the discovery loop into the laboratory. Clinical studies and the REALM project show why evaluation must match the claim, the user, the setting and the benefit. CHARM and PCX provide the concrete UAB and Maastricht context before the closing synthesis. AI contribution statement. An AI assistant supported source checking, synthesis, slide editing, speaker-note drafting and visual review. Existing scientific figures, platform screenshots, logos and the embedded TESSERA animation derive from the supplied presentation. The assistant did not independently validate the scientific results, execute formal proofs, or establish clinical effectiveness. Michel Dumontier is responsible for the final presentation and interpretation.