We're Overestimating Medical AI — and Underestimating the Harm (Jessica Morley, Yale)

Healthcare is rushing to adopt AI not because it solves real clinical problems, but because it's the easiest technology to build with available data—what Jessica Morley, an AI ethicist at Yale, calls 'a hammer in search of nails.' She warns that we're systematically overestimating AI's capabilities while underestimating its harms, especially in ways that amplify bias, erode patient autonomy, and create inaccurate medical records. The real danger isn't AI failing—it's AI succeeding too well at being persuasive, leading patients and clinicians to trust flawed, hallucinated diagnoses and recommendations without question. Morley argues that ethical principles like beneficence and justice were designed for one-on-one patient care, not the mass, algorithmic decision-making that defines today’s generative AI. Without new frameworks for population-level harm, we risk scaling systemic inequities. Yet she remains cautiously optimistic, offering concrete, actionable guardrails: never trust AI at face value, always ask 'why?', use adversarial prompting to challenge its answers, and create personalized 'harnesses' that reflect your own medical history and values. The most powerful tool isn't the AI—it's your ability to question it.