Google DeepMind Launches AI Co-Clinician Research Program: Multimodal AI Agent Assisting Doctors and Patients

BlockBeats News, May 1st — DeepMind announced the launch of the AI Co-Clinician research program on April 30, 2026, to explore how a multimodal AI agent can support medical decision-making under physician supervision. The system enables real-time interaction through video and audio to conduct medical history taking, guide physical examinations, provide initial diagnostic reasoning (such as rotator cuff injuries or tendinitis), offer care recommendations, and explicitly identifies itself as a non-physician, with all suggestions intended for use under physician supervision.

Technical features: It employs real-time multimodal processing (analyzing gait, respiration, rashes, etc.), a dual-agent architecture (Planner for ongoing monitoring, Talker to ensure safety boundaries), and the NOHARM security framework.

Test results: Out of 98 primary care inquiries, 97 achieved zero critical errors; in 20 simulation scenarios and across 140 evaluation dimensions, 68 dimensions showed performance matching or surpassing that of physicians (especially in triage), while human physicians still demonstrated a significant advantage in identifying key "red flag" signals and guiding examinations.

This project, conducted in collaboration with Harvard Medical School, Stanford School of Medicine, among others, is currently in the research phase and will progressively expand into clinical trials. DeepMind emphasizes that the AI is intended to enhance clinical judgment as a "collaborative member" rather than replace it.