A conversational diagnostic system built by Google has taken a step out of the lab and toward the exam room. The Articulate Medical Intelligence Explorer, or AMIE, was put in front of real patients in an urgent primary care clinic run by Beth Israel Deaconess Medical Center, and the prospective study appeared in The Lancet on October 8.
The setup was kept deliberately narrow. People booked for new, non-emergency complaints could message the chatbot from home for up to five days before their appointment. AMIE would gather a person’s symptoms and medical history, offer some possible diagnoses for the patient to raise with a physician, and hand back a written summary for the clinician to look over ahead of the visit.
Every conversation was watched live, over video, by a physician prepared to step in under four pre-specified safety rules. Of the 100 patients who finished the pre-visit chat, 98 kept their appointment, and across all 98 encounters no conversation had to be halted. The supervising doctors flagged one hallucination and added clinical clarification in five cases, BIDMC reported.
When a doctor reviewed the transcript in advance, it helped them prepare 75 percent of the time and may have changed how they handled the visit in 57 percent of cases. AMIE’s top seven diagnostic suggestions contained the final diagnosis in 90 percent of cases, and its first suggestion was correct 56 percent of the time.
The study is small and confined to one center, and Google itself says larger trials are needed. Still, it marks the company’s first paper in The Lancet’s main journal and nudges patient-facing medical AI from simulation into supervised reality.
