Plenty of studies have shown that AI chatbots can answer medical exam questions. Far fewer have tested what happens when a chatbot talks to real, unwell patients. A study from Beth Israel Deaconess Medical Center in Boston, published in The Lancet and announced on 9 October, is one of the first to do exactly that.
Researchers invited patients who had booked urgent primary care appointments to chat with an AI system from home before seeing their doctor. The tool, called the Articulate Medical Intelligence Explorer, or AMIE, was developed with Google researchers. Through a secure text chat, it asked about symptoms and medical history, suggested possible diagnoses for discussion with the doctor and produced a summary for clinicians.
What they found
Of 114 patients enrolled, 98 completed both the AI chat and their appointment. Every conversation was watched in real time by a board-certified physician who could step in. None of the 98 conversations needed a safety stop. Supervising doctors spotted one hallucination, meaning made-up information, and added clinical clarification in five cases.
Patients rated the conversations well, especially the system's ability to listen, explain and put them at ease, and their attitudes toward AI improved after using it. Their main concerns were confidentiality and whether the chatbot was honest and trustworthy.
Doctors reviewed the AI's transcript or summary before 44 visits. In 75% of those cases, they said it helped them prepare, and in 57% they said it may have influenced their approach. In one case, a doctor felt it was somewhat harmful, because the AI listed lymphoma as a possibility and may have made the patient anxious.
"AI tools must first be shown to operate safely and acceptably in real care workflows," said Dr. Adam Rodman, director of AI programs at the hospital's Carl J. Shapiro Institute.
The limits matter. This was a small feasibility study at one hospital, designed to test safety rather than whether AI improves health outcomes. Every chat was supervised, so the results may not reflect unsupervised use. The study was funded by Alphabet, Google's parent company, and Rodman was a visiting researcher at Google during part of it.
What it means for patients: AI that interviews you before an appointment may soon be common, and early signs are encouraging. For now, it works best as a helper for your doctor, not a replacement.
Photo: Shixart1985, CC BY 2.0, via Wikimedia Commons
