From typed history to seeing and hearing
Google announced an August 11 study extending AMIE into real-time clinical video consultations. The Gemini- and Project Astra-based multi-agent system combines low-latency dialogue, clinical reasoning and live audio-visual perception to observe nonverbal cues and guide patient actors through virtual examination maneuvers.
The study covered 100 scenarios, 15 patient actors and 300 standardized consultations across AMIE Video, AMIE Text and physician video arms. A detailed Google Research account says the work involved 30 primary-care physicians and 20 independent physician evaluators. Evaluators rated AMIE Video on par with or above physicians in history taking, diagnosis, management and communication. Patient actors preferred AMIE for assessment and explanation, but physicians for rapport and partnership.
A research system, not a clinical product
These were simulated OSCE encounters with trained actors—not real-world patient care, long-term outcomes, emergencies or broad health systems. The authors report limits in fine anatomical precision, subtle affect and high-frequency movement. Google explicitly says more research is required before real-world deployment. Clinicians must remain responsible for supervision and decisions, and users should not treat this research system as a diagnostic service.