Google’s research teams have moved AMIE, their experimental medical AI system, into a live video setting for the first time. In the demonstration, the model read visual and auditory signals, walked through virtual physical exams and reasoned diagnostically on the fly.
AMIE runs on Gemini and Project Astra through a multi-agent architecture. The system now picks up on a cough, an unsteady gait or visible signs of discomfort that words alone would miss, the team wrote in an Aug 11 blog post.
In a randomized study using simulated consultations with patient actors and practicing primary care physicians, outside evaluators rated AMIE highly on history-taking thoroughness, diagnostic accuracy, management appropriateness and communication quality. The actors themselves said they preferred the video session to texting with the system.
The company emphasizes that AMIE is still a research system and that more work is needed before responsible real-world clinical deployment. The demonstration nonetheless signals where the company believes health AI is heading: an AI that can see, hear and converse like a clinician rather than answer questions in a chat box.
The push follows earlier AMIE milestones, including a Nature study on long-term disease management and a triadic care model in which AI, clinician and patient collaborate. Video consultation is the first time Google has shown expert-level performance in a fully synchronous, face-to-face-equivalent setting.
