# Google Reports Its Medical AI Matched Doctors in Simulated Video Consultations

In a randomized examination with 30 primary care physicians, 15 patient actors and 100 scenarios, evaluators rated the video version of AMIE at or above physicians on diagnosis and management.

- Published: 2026-08-12T06:11:01.127Z
- Canonical: https://polylog.news/ai/2026-08-12/google-reports-its-medical-ai-matched-doctors-in-simulated-v
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Google AI Blog](https://blog.google/innovation-and-ai/models-and-research/google-research/amie-video-consultations/), [Google Research](https://research.google/blog/advancing-amie-towards-expert-level-audio-visual-clinical-consultations/), [arXiv](https://arxiv.org/abs/2608.09861)

Google Research published results for a video-capable version of AMIE (Articulate Medical Intelligence Explorer), its experimental diagnostic system, in what it describes as [the first demonstration of a real-time clinical video consultation conducted by an artificial intelligence (AI) system](https://blog.google/innovation-and-ai/models-and-research/google-research/amie-video-consultations/).

The study used a randomized Objective Structured Clinical Examination, the standard format for assessing clinical skill. Thirty primary care physicians, fifteen professional patient actors and one hundred scenarios were used to compare AMIE over video, AMIE over text, and physicians over video. Independent clinical evaluators [rated the video system on par with or better than the physicians](https://research.google/blog/advancing-amie-towards-expert-level-audio-visual-clinical-consultations/) on history-taking, diagnosis, management and physical observation. Patient actors preferred the system's explanations but preferred physicians for rapport.

Architecturally, [the paper describes](https://arxiv.org/abs/2608.09861) a Gemini-based asynchronous multi-agent system with separate Talker, Planner and Perception components, built on the low-latency stack from Project Astra, so that visual and auditory input is processed while dialogue continues.

The authors state the limits themselves. Every encounter involved actors performing scripted conditions, not patients with real illness, and the system remains weak on fine anatomical precision and rapid movement. Actor-based examinations reward structured questioning, which is exactly what a language model does well.

## What this means

Real-time perception combined with dialogue at consultation speed is the technical barrier that separated medical chat systems from telehealth, and Google now reports clearing it in simulation. Telehealth vendors and primary care staffing models are the exposed parties, because triage and history-taking are the highest-volume and lowest-margin part of that work. The decisive question is whether the result survives contact with real patients, whose presentations are ambiguous and where the cost of a missed diagnosis falls on the operator. If the actor-based advantage disappears in clinic, that would mark this as an artifact of the evaluation method rather than a genuine capability shift.

## What to watch

- Whether Google moves AMIE into a prospective study with real patients and reports outcomes, the only evidence that would support deployment.
- How regulators in the United States and Europe classify a conversational system that conducts examinations, since device approval pathways were not written for this.
- Whether competing labs publish comparable examination results, which would show the advance comes from the multimodal architecture rather than from Google's medical data access.