Google Research and Google DeepMind have released a study showing AMIE, a medical AI system, performing real-time video consultations with human actors. The system interprets visual and auditory cues to guide virtual physical exams and reason diagnostically in real time. Built on Gemini and Project Astra using a multi-agent architecture, AMIE operates alongside primary care physicians during simulated appointments. Clinical evaluators assessed the system favourably across history-taking thoroughness, diagnostic accuracy, management appropriateness and communication quality. Patient actors also preferred the video experience over text-based chat.
This work demonstrates that current models can simulate expert-level clinical interactions effectively. It validates the potential for AI to augment diagnostic processes without replacing human oversight. The results suggest video interfaces may improve patient engagement compared to standard text messaging.
* The system interprets coughs, gait, and visible signs of discomfort
* Evaluators rated AMIE highly on core clinical competencies
* Video consultations were preferred by patient actors over text
