This talk examines how AI is currently used - and increasingly misused - in medical research. It begins with examples where AI methods have achieved genuine advances and others where they have clearly failed. Between these extremes lies a wide area where the term “AI” is often applied inconsistently or without sufficient methodological justification. I will discuss what is actually meant by AI, machine learning, and related concepts, and clarify where these approaches can be useful and where they are not. The talk focuses on three main forms of misuse: the selling argument - using the label “AI” to make research appear more sophisticated; the misuse of method - applying AI techniques where they offer no real benefit; and simple falsification. To illustrate these issues, I will present a recent example from my own research on a biomarker for chronic pain, showing how such problems can arise and how difficult they are to flag and correct within the scientific and publishing landscape. The talk concludes with reflections on how the research community can promote more transparent, critical, and responsible uses of AI.