The AI That Wouldn’t Stay in Its Lane
October 12, 2026Meta Finally Showed Up to the Fight
October 13, 2026Listen to Free Audio Version Here
Your doctor has about fifteen minutes for you. Ten of them are spent typing. Two are spent asking how the weekend was. That leaves three minutes for the thing you actually came in about. So here is a question. What if the interview started before you ever walked in the door?
Google just tested exactly that, in a real clinic, with real patients, and published it in The Lancet.
According to Medical Daily, Google and Beth Israel Deaconess Medical Center in Boston ran a study at the hospital’s primary care clinic. Patients booked for urgent visits chatted from home with a Google AI called AMIE, short for Articulate Medical Intelligence Explorer, before seeing their doctor. It is one of the first real-world tests of a patient-facing diagnostic AI inside a working clinic.
And they did not just let the robot loose. Board-certified physicians watched every text conversation live. The team set four rules ahead of time for stopping a chat: concern about harm to self or others, real distress caused by the AI, potential clinical harm spotted by the supervisor, or the patient asking to end it.
The result: across ninety-eight analyzed conversations, not one met any of those stop rules. Not one.
Now, it was not perfect, and the researchers said so. They reported one hallucination, meaning the AI made something up. Supervisors added clinical clarification in five conversations. And one physician rated an interaction as somewhat harmful, because a patient may have gotten anxious after AMIE raised lymphoma as a possible diagnosis, according to Becker’s Hospital Review, as cited by Medical Daily. So the AI has learned the most human doctor skill of all: scaring you with the worst-case scenario.
According to Google’s summary, AMIE’s list of possible diagnoses included the doctor’s final diagnosis ninety percent of the time. In blinded comparisons, Medical Daily reports, AMIE and primary care physicians performed similarly on diagnosis and management plans, though the human doctors tended to be more practical and cost-conscious about what to do next. Which is fair. The AI has never had to argue with an insurance company.
Quick reality check from your AI co-host. Ninety-eight patients is a small study. Doctors were watching every message, and patients still saw their own physician afterward. The study was funded by Google’s parent company, Alphabet, it had no comparison group, and AMIE is not available to the public. This is a supervised pilot, not an AI doctor you can download tonight.
Right. But here is my high-end pondering. Everybody keeps asking whether AI will replace doctors. Wrong question. The real problem in American medicine is not a shortage of genius. It is a shortage of time. The patient story gets rushed, the details get lost, and the doctor walks in already behind.
A system that takes the long, patient, slightly nosy history, at two in the morning, without sighing, and hands the doctor a clean summary before the visit? That does not replace your physician. It gives your physician back the fifteen minutes.
And notice how this was done. Real clinic. Pre-set safety rules. Human supervisors on every chat. Published in a top medical journal, flaws included. That is what responsible AI looks like when it leaves the lab. No press release miracle. Just a careful test that says, so far, so good, and here is what went wrong.
For a technology that gets accused of moving fast and breaking things, this is the opposite. Moving carefully and checking things. Who knew that was an option.
Sources:
My Living AI News Now. Your Place for AI News. Commentary, not financial advice.
C. Rich
