Google's AMIE diagnostic AI passes real-world test: Lancet study of 114 patients, zero safety stops

vivnat · x · 2026-10-13

Google's conversational diagnostic AI AMIE has a real-world clinical feasibility study published in The Lancet, with coverage in The New York Times.

Key findings from the prospective single-centre, single-arm study:

Patients chatted with AMIE up to 5 days before an urgent primary care appointment, with physician safety supervisors monitoring all interactions; transcripts were shared with PCPs beforehand.
From April to November 2025, 114 patients enrolled and 98 completed both the AMIE interaction and the PCP visit. Zero conversation safety stops were required; supervisors noted one hallucination and added clinical information in five interactions.
Clinical evaluators rated AMIE's conversations favorably in 87–100% of cases (17 criteria); patients rated it 48–96% (16 criteria). Patient attitudes toward AI improved and stayed elevated after seeing their physician.
Of 44 PCPs who reviewed the AMIE transcript beforehand, 33 found it helpful for visit preparation and 25 said it might have changed their behavior.

Author vivnat recounts two inflection points: early-2022 access to Google's LaMDA and PaLM, which sparked the Google Brain moonshot leading to Med-PaLM, and late summer 2023, when early AMIE models trained with a self-play loop led directly to this Lancet study (following the OSCE study published in Nature 2025).

Original post →

Google's AMIE diagnostic AI passes real-world test: Lancet study of 114 patients, zero safety stops

More from Apps