Engineering trust with clinical-grade AI security
The transition and implementation of AI in clinical environments requires uncompromising architectural and operational security. In our research on simulating telemedicine conversations with a patient, an artificial intelligence co-clinician uses a dual-agent architecture: the “Planner” module constantly monitors the conversation, verifying that the “Talker” agent is within sheltered clinical limits.
Similarly, to meet the needs of physicians, the AI co-clinician prioritizes clinical-grade evidence, performing verification and citation checking to retrieve it. The assessments we present above were developed by medical professionals to reflect a range of their real-world evidence needs, formulating questions based on hypothetical scenarios to rigorously assess the capabilities of artificial intelligence.
Research collaboration to rigorously evaluate co-clinician AI in the real world
To further develop and evaluate AI co-clinicians, we are currently developing a phased approach in collaboration with academic and research collaborators in various healthcare settings around the world, including the US, India, Australia, Up-to-date Zealand, Singapore and the United Arab Emirates.
As we progress through these evaluation phases, we will continue our research in more locations, including mission-driven health care organizations and academic medical centers. Our goal is to ensure that medical AI is developed and deployed responsibly, in accordance with applicable standards, supporting better health around the world.
Thanks
We are grateful to our research partners at Harvard Medical School and Stanford Medicine and the many medical centers and care organizations that have engaged in further evaluations of trusted testers with our team. This project involved collaboration with multiple teams from Google DeepMind, Google Research, Google Cloud, and Google for Health. We thank our team members for their insightful discussions and contributions.
In particular, the Artificial Intelligence Clinician Collaboration would not be possible without the fundamental research and engineering efforts of Aniruddha Raghu, Arthur Chen, Charlie Taylor, CJ Park, David Stutz, Devora Berlowitz, Doug Fritz, Dylan Slack, Eliseo Papa, Jack Chen, JD Velasquez, Jing Rong Lim, Katya Tregubova, Kelvin Guu, Meet Shah, Richard Green, Ryutaro Tanno, Sukhdeep Singh, Victoria Johnston, Adam Rodman.
We thank our many collaborators for their invaluable contributions, including Ali Eslami, Aliya Rysbeck, Andy Song, Anil Palepu, Anna Cupani, Bakul Patel, Bibo Xu, Brett Hatfield, David Wu, Ed Chi, Emma Cooney, Erica Oppenheimer, Erwan Rolland, Euan A. Ashley, France Pietra, Resca-F. Gordon Turner, Gregory Wayne, Hannah Gladman, Irene Teinemaa, Jack O’Sullivan, Jacob Koshy, Jan Freyberg, Jason Gusdorf, Joelle Wilson, Katherine Tong, Juraj Gottweis, Michael Howell, Mili Sanwalka, Pavel Dubov, Pete Clardy, Peter Brodeur, Sico Dale, Suman Wailan, Rachel Manth, Manth Cemgil, Tim Strother, Uchechi Okereke, Valentin Lievin, Vishnu Ravi, Yana Lunts, Yun Liu, Simon Staffell, Rachel Teo, Adriana Fernandez Lara, Armin Senoner, Danielle Breen, Paula Tesch, Leen Verburgh, Dimple Vijaykumar, Juanita Bawagan, Muinat, Maria Ash Montes and Rob Abdul. The feature films were produced by Christopher Godfree, Matt Mager, Emma Moxhay and Simon Waldron.
We thank James Manyika and Demis Hassabis for their insightful guidance and support throughout the research process.