Google DeepMind Reveals AI Co-Clinician for Real-Time Clinical Decision Support
Google DeepMind announced an AI co-clinician system that uses live video and audio to analyze physical symptoms—including gait, breathing, and rash appearance—in real time. In a simulation study with patient-actors alongside Harvard and Stanford physicians, the system using the NOHARM safety framework achieved zero critical errors in 97 of 98 primary care queries and matched or outperformed physicians in 68 of 140 assessed areas including triage. A dual-agent architecture uses a Planner agent to continuously verify that the patient-facing Talker agent stays within safe clinical boundaries.
Why It Matters
The dual-agent safety architecture and the clinical partnership model represent a significant step toward regulated AI deployment in healthcare—the first major lab demonstration of real-time multimodal clinical support at this performance level.