Particle.news
Download on the App Store

AI Matches Primary Care Doctors in Simulated Video Visits

A Gemini-based system combined live audio and video with low-latency clinical dialogue to reach expert-level scores in a randomized simulation, signaling a step toward richer telehealth tools but requiring real-world validation.

Overview

  • A new arXiv preprint reports that AMIE (Video), a Gemini-based multi-agent system, was rated by clinical evaluators as matching or exceeding primary care physicians on history-taking, diagnosis, management, and physical observation in a randomized OSCE.
  • The study compared AMIE (Video), a text-only AMIE (Text), and clinicians across 100 standardized scenarios with 30 primary care doctors and 15 patient actors in a simulated video-consultation exam.
  • Patient actors said they preferred AMIE's clinical explanations and chose the video interface over text for feeling understood, while they still favored human clinicians for rapport and partnership.
  • Authors and coverage stress key limits: the work is a non–peer-reviewed preprint based on simulations and actors, the system struggles with fine anatomical detail, subtle emotional cues, and rapid movements, and further supervised real-world testing is needed before clinical use.
  • The team also released a taxonomy and automated metrics for audio-visual clinical cues to standardize evaluation of multimodal telehealth AI and to guide future development and safety assessments.