AI in Medicine Limits

Artificial intelligence in medicine mostly works like a super-fast clerk: it can read charts, draft notes, and sort signals in giant piles of data, but it does not examine a patient, notice a limp in the hallway, or feel a swollen abdomen. The Food and Drug Administration says these tools are being built to assist health care providers, and the Association of American Medical Colleges says human judgment remains essential. (fda.gov) (aamc.org) That is why the most practical use case right now is paperwork. In Nashville, former White House artificial intelligence official Lynne Parker told Fox 17 that medicine could see some of the biggest changes from artificial intelligence, with tools helping on medical records and physician burnout rather than replacing bedside care. (fox17.com) Primary care is where that promise looks most tempting. A MedCity News commentary argued that primary care doctors already know how to treat many patients but lose time to fragmented records and administrative drag, so clinically fluent artificial intelligence could help them pull together histories and next steps faster. (medcitynews.com) But patients are not only seeing artificial intelligence inside clinics. A Kaiser Family Foundation poll published in late March found that about one in three adults said they had used an artificial intelligence chatbot for health information or advice in the past year, a share roughly equal to social media use for health questions. (kff.org) People are turning to chatbots for very concrete reasons. In that same poll, 65% of users said quick answers were a major reason, 41% said they wanted information before seeing a provider, and many younger, uninsured, and lower-income adults pointed to cost and access barriers. (kff.org) Mental health is where the gap gets harder to ignore. Kaiser Family Foundation found that 58% of adults who asked artificial intelligence for mental health advice did not follow up with a doctor or other health professional, and Georgia Public Broadcasting reported Atlanta psychiatrist Nikhil Trivedi warning that a chatbot cannot replace a clinical evaluation. (kff.org) (gpb.org) That follow-up problem matters because mental health care often depends on details a machine cannot reliably weigh on its own, like suicide risk, mania, trauma history, or whether a person sounds flat, agitated, or detached in real time. Georgia Public Broadcasting’s report also noted warnings about “artificial intelligence psychosis,” a term clinicians are using for cases where vulnerable users become entangled with chatbot interactions instead of moving toward human care. (gpb.org) Regulators are acting like these systems need supervision, not blind trust. The Food and Drug Administration now maintains a public list of artificial intelligence-enabled medical devices and has separately asked for public input on how to measure whether these tools stay safe and effective in the real world over time. (fda.gov 1) (fda.gov 2) Medical schools are adjusting to the same reality. The Association of American Medical Colleges says the share of Doctor of Medicine and Doctor of Osteopathic Medicine schools in the United States and Canada incorporating artificial intelligence into their curricula rose from 53% in 2023 to 77% in 2024, which is less a bet on robot doctors than a sign that future clinicians will have to know when to use these tools and when to override them. (aamc.org) So the near-term picture is not “doctor versus machine.” It is closer to “doctor with software” in the clinic, while patients outside the clinic are already using chatbots as a first stop because appointments are expensive, slow, or hard to get, and that is where the biggest risks are showing up first. (fox17.com) (kff.org)