Robert Wachter, a physician and author, recently shared his perspective on artificial intelligence in healthcare during a presentation at the SHM Converge conference. Drawing from over 100 interviews for his book A Giant Leap: How AI is Transforming Healthcare and What That Means for Our Future, he argued that AI is shifting from a novelty to a standard tool in medicine—one that could help address long-standing problems in the system.
Wachter described a healthcare system still struggling with access and administrative burdens. The electronic health record, once promised as a solution, fell short. Now, AI is being tested in roles like chart summarization, clinical decision support, and documentation. His colleagues have welcomed the rollout of AI tools, including scribes, chart summarization tools, and discharge summary drafts, which save time and allow clinicians to spend more time communicating with patients.
His optimism isn’t unchecked. He acknowledged that while some early concerns—like AI “hallucinations” or bias—have diminished, new risks have emerged. Privacy and security remain major issues, particularly with the potential for HIPAA violations. But his biggest worry is disinformation. To illustrate, he showed a deepfake video of himself making false statements in a hospital hallway, a stark reminder of how convincing AI-generated misinformation can be.
Speed over perfection
Wachter urged healthcare leaders to move quickly, even if imperfectly. He quoted Dr. Gianrico Farrugia: “The risk of not going fast enough is far greater than the risk of going too fast.” Governance, he argued, should focus on adaptable oversight, staff training, and identifying skill gaps—not waiting for flawless systems.
Related: Giving Feedback That Actually Works
For clinicians, the shift raises questions about job security and deskilling. Wachter compared the risk to losing a phone: if he lost his phone, he might never see his wife again because he relies on it to store her number. He placed hospital medicine in a middle ground—less vulnerable than specialties like radiology or pathology, which rely heavily on pattern recognition, but not as insulated as surgery. He emphasized that AI won’t achieve perfect accuracy, meaning human judgment will remain essential.
Wachter recommended medical education curricula that teach learners how to serve effectively as the human in the loop, prioritizing diagnostic reasoning and avoiding deskilling from AI use.
Despite the risks, Wachter remains optimistic. AI, he said, is becoming less of a novelty and more of a routine part of healthcare. The challenge for leaders isn’t whether to adopt it, but how to do so responsibly—balancing speed with caution, and automation with human oversight.
For now, the tools are still evolving. But the early signs suggest they might help address some of the challenges that electronic health records never could.
