What the research found
In a prospective single-centre feasibility study at Beth Israel Deaconess Medical Center, patients booked for urgent primary care used a conversational diagnostic AI (AMIE, developed by Google) by text chat before their appointment, while a physician supervisor monitored every conversation against predefined safety criteria. Of 114 patients enrolled, 98 completed both the chat and the visit; no conversation needed a safety stop, supervisors noted one hallucination and added clinical information in five, and clinical evaluators rated conversations favourably in 87–100% of cases across 17 criteria. The study was funded by Alphabet and measured feasibility rather than health outcomes; an accompanying Lancet comment notes that clinical evidence for such tools is still mostly synthetic.
This is what responsible evidence looks like, with real patients, a clinician watching every exchange and safety as the primary outcome, and it is the bar that unsupervised deployments have not yet met.