The medical community should be willing to experiment with AI systems that make autonomous clinical decisions, rather than insisting physicians always have the final say, Ezekiel Emanuel, MD, PhD, and venture capitalist Vinod Khosla wrote in a Sept. 8 article for The Atlantic.
Dr. Emanuel is an oncologist, bioethicist and vice provost for global initiatives at the University of Pennsylvania. Mr. Khosla is an entrepreneur, technologist and founder of venture capital firm Khosla Ventures.
The essay expands on an Aug. 17 commentary the two published in JAMA, co-written with Abe Baker-Butler, also of the University of Pennsylvania, and Neal Khosla, Vinod Khosla’s son and CEO of AI virtual care company Curai Health. Vinod Khosla disclosed in the Atlantic piece that he has invested in the virtual primary care startup, co-founded by his son, in which physicians supervise AI assistants.
The authors argue that much of modern medicine — managing hypertension, determining cancer treatment, addressing chemotherapy side effects — follows algorithmic steps AI can execute reliably. Despite their best intentions, physicians can easily make mistakes carrying out these steps; the authors noted up to 15% of Americans are misdiagnosed annually, contributing to tens of thousands of preventable deaths.
Dr. Emanuel and Vinod Khosla point to a study in which Google’s Articulate Medical Intelligence Explorer outperformed physicians at eliciting patient information, a 2023 trial in which autonomous AI adjusted insulin doses for type 2 diabetes patients faster than physicians did, and a 2025 review of 52 studies that found physician-AI hybrid teams did not outperform AI alone. The latter study, they argue, undercuts the common argument that keeping a physician in the loop is inherently safer.
Another common concern is that AI can’t replace the human connection of medicine. However, Dr. Emanuel and Vinod Khosla argue that physicians can also come across as robotic themselves and fail to develop deep, trusting relationships with patients. They cited a Gallup survey in which 21% of respondents said they turned to AI after feeling dismissed by a healthcare provider.
Rather than limiting progress with strict physician oversight, the authors argue the field should focus on resolving the open questions autonomous AI raises: how to design ethical, real-world testing for these models, how regulators and professional bodies should oversee them, and how medical education needs to change so physicians can focus on the skills only humans bring.
The authors cited the American Medical Association’s opposition to their JAMA commentary. AMA’s own position — laid out in an Aug. 18 framework developed with the Digital Medicine Society — emphasizes physician decision-making authority. The framework identifies five “enduring responsibilities” physicians must retain as AI expands in medicine, including preserving human connection, demonstrating clinical judgment and stewarding safe AI integration.
Even so, the authors concede the medical establishment isn’t uniformly opposed to loosening the reins. They cite a position paper the American College of Physicians published last week that states “fully autonomous AI is possible,” though the group recommends limiting its use to low-risk, low-complexity decisions with a clinician available to step in.
“Doctors have been presented with a powerful tool that has the potential to improve care for millions of people. It’s their duty not to stand in the way,” they wrote.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
