Google's AI Doctor Now Sees Patients in Real Time—And Matches Human Physicians in Clinical Trial

Google Research has unveiled a significant evolution of its AMIE medical AI system, now capable of conducting real-time video consultations that match the clinical performance of board-certified primary care physicians. The advancement, detailed in a research blog post published August 11, 2026, represents what the company describes as the first demonstration of AI achieving expert-level performance in synchronous clinical video consultations.

Multi-Agent Architecture Solves Real-Time Challenge

The enhanced AMIE system employs a novel three-agent asynchronous architecture designed to overcome the fundamental tension between conversational responsiveness and clinical reasoning depth. According to the research team led by Anil Palepu and Mike Schaekermann, the system comprises a Dialogue Agent that maintains natural conversational flow, a Reasoning Agent that conducts diagnostic analysis, and a Perception Agent that processes audio-visual clinical cues.

Built on Google's Gemini and Project Astra platforms, this architectural approach allows the system to respond to patients within natural conversational timeframes while simultaneously conducting complex medical reasoning in the background—a technical challenge that has constrained previous real-time medical AI applications.

Matching Physician Performance Across Clinical Domains

In a multi-arm randomized OSCE study involving 100 clinical scenarios and 300 live consultations, AMIE (Video) was evaluated against 30 board-certified primary care physicians by 20 independent physician evaluators. The scenarios spanned five body systems: cardiopulmonary, abdominal, head-eyes-ears-nose-throat, neurological/psychiatric, and musculoskeletal conditions.

The AI system was rated on par with human physicians across history-taking, diagnostic accuracy, management appropriateness, and communication quality. Notably, AMIE (Video) received significantly higher ratings than both PCPs and its text-based predecessor in eliciting physical signs and guiding virtual examination maneuvers—addressing a fundamental limitation of text-based telehealth AI.

The 15 trained patient actors who participated in the study expressed a strong preference for the video interface over text-based chat, providing favorable ratings on empathy, rapport, and confidence in the care received.

Clinical Partnerships Signal Path to Real-World Deployment

Google has established clinical partnerships to advance the system toward practical application. A feasibility study is underway with Beth Israel Deaconess Medical Center, while Included Health is conducting an ongoing nationwide randomized study to evaluate the technology in broader clinical contexts.

The research team acknowledges important limitations: the study was conducted with simulated scenarios using trained actors rather than actual patients, the system occasionally exhibited perceptual errors and technical disruptions, and real-world patient benefit remains unvalidated. These caveats underscore that while the technology demonstrates promising capabilities in controlled settings, clinical deployment will require additional validation to ensure safety and effectiveness in actual healthcare delivery.

The advancement positions Google to compete directly with telehealth platforms while potentially expanding access to clinical expertise in underserved areas, though regulatory pathways and real-world performance validation remain critical hurdles before widespread clinical adoption.

Sources

1. Advancing AMIE towards expert-level audio-visual clinical consultations — Google Research