The Global Healthcare Workforce Crisis
Healthcare systems worldwide face a mounting crisis. They strive for better patient outcomes, lower costs, and improved experiences for both patients and clinicians. Yet progress stalls against a widening gap in clinical expertise. The World Health Organization projects a shortfall of more than 10 million health workers by 2030.
AI has long been seen as a solution to bridge this gap. However, it has not yet fully met the real-world needs of clinicians and patients. To address this challenge directly, Google DeepMind announced its AI co-clinician research initiative on April 30, 2026. The goal is clear: explore how AI can amplify doctors’ expertise and deliver higher-quality care at scale.
What Is the AI Co-Clinician?
Defining the Triadic Care Model
Google DeepMind’s journey in medical AI spans several years. Early work focused on mastering examination-style medical knowledge tests through MedPaLM. Later, the AMIE system matched physician performance in text-based simulated consultations, including real-world feasibility trial settings.
Now, DeepMind hypothesizes a new model for care delivery: triadic care. In this model, AI agents support patients through their care journeys under the clinical authority of their physician. Medicine has always been a team effort. AI agents, therefore, extend clinicians’ reach while keeping judgment and control firmly with human experts.
The AI co-clinician functions as a collaborative member of the care team. It interacts with patients under expert clinical supervision and operates in both clinician-facing and patient-facing settings. Addressing both perspectives is essential for improving quality, cost, availability, and experience across care delivery.
How AI Augments Clinicians
Evidence Synthesis and Clinical Accuracy
For any physician, a tool is only useful if it is trustworthy. DeepMind researchers, working alongside academic physicians, adapted the “NOHARM” framework to test the AI for two critical error types: errors of commission (incorrect information) and errors of omission (failure to surface critical information).
In blind, head-to-head evaluations, physicians consistently preferred AI co-clinician over leading evidence synthesis tools. Furthermore, in an objective analysis of 98 realistic primary care queries, the system recorded zero critical errors in 97 cases. Notably, it outperformed two AI systems already widely used by physicians. These results demonstrate that the system sets a new benchmark for clinical evidence quality.
Medication Knowledge and Reasoning
Answering queries about medications requires precision that most AI systems struggle to deliver. To fill this research gap, DeepMind evaluated AI co-clinician on the OpenFDA set of RxQA questions — a demanding benchmark for complex medication knowledge and reasoning.
The AI co-clinician surpassed other frontier AI systems, especially when clinicians posed questions in the open-ended format used in real care settings. This matters because real clinical questions rarely come as multiple-choice options. As a result, the findings highlight AI’s growing potential to support data-intensive care planning and management.
Real-Time Multimodal Capabilities in Telemedicine
Beyond Text: Eyes, Ears, and a Voice
Expert clinical assessment relies on subtle visual and auditory cues — a patient’s gait, respiratory patterns, or skin changes. Prior AI systems, limited to text, could not capture these signals. Consequently, their clinical value remained constrained.
Google DeepMind took a different approach. Building on Gemini and Project Astra, researchers tested AI co-clinician’s ability to use live audio and video in simulated telemedical calls. The goal was to evaluate whether AI could support better diagnosis and management under expert supervision in multimodal settings.
Simulation Study Results
Working with academic physicians from Harvard and Stanford, the team designed a randomized simulation study. It included 20 synthetic clinical scenarios and 10 physician “patient-actors.” In total, the study assessed over 140 aspects of consultation skill.
Key findings include:
- Expert physicians outperformed the AI system overall, particularly in identifying “red flags” and guiding critical physical examinations.
- However, AI co-clinician performed at or above the level of primary care physicians in 68 of the 140 assessed areas.
- Moreover, the AI demonstrated new capabilities beyond text-only systems, such as correcting a patient’s inhaler technique and guiding shoulder maneuvers to identify a rotator cuff injury.
These results confirm that AI co-clinician is best positioned today as a supportive tool for practitioners, not a replacement for clinical judgment.
Safety Safeguards for Clinical-Grade AI
A Dual-Agent Architecture
Deploying AI in clinical environments demands uncompromising safety standards. To meet this need, DeepMind implemented a dual-agent architecture for patient-facing telemedical conversations. A dedicated “Planner” module continuously monitors conversations. It verifies that the “Talker” agent stays within safe clinical boundaries at all times.
Additionally, AI co-clinician prioritizes clinical-grade evidence for all clinician-facing queries. The system performs verification and citation checking during retrieval. Physicians constructed all evaluations to mirror their actual real-world evidence needs — ensuring the testing reflected genuine clinical complexity
Global Research Collaborations
Phased Evaluation Across Diverse Healthcare Settings
To further develop and rigorously assess AI co-clinician, DeepMind pursues a phased approach with academic and research collaborators across globally diverse healthcare settings. Current collaborators span the United States, India, Australia, New Zealand, Singapore, and the UAE.
As the initiative advances through evaluation phases, DeepMind will expand its research partnerships with mission-aligned healthcare organizations and academic medical centers. The overarching goal is to ensure that medical AI develops and deploys responsibly, in line with applicable standards, and in support of better health outcomes worldwide.
Note: Research collaborations are not, at this stage, intended for diagnosing, treating, or preventing disease, nor for providing medical advice.
The Road Ahead
AI as a Teammate, Not a Replacement
The AI co-clinician initiative represents a significant step forward in responsible medical AI. Rather than replacing physicians, the system aims to extend their reach, reduce administrative burden, and bring higher-quality care to more people. With each phase of research, DeepMind moves closer to a future where AI serves as a trusted, capable, and safe member of every care team.
