Dr. Robert Wachter, the chair of the Department of Medicine at the University of California, San Francisco (UCSF), has long contemplated the intersection of technology and medicine. He introduced the term “pajama time,” referring to the hours doctors spend after clinic hours on documentation. His notable 2015 book, The Digital Doctor, explores both the potential benefits and drawbacks of the
In his latest work, “
Medical Economics interviewed Wachter about the actual benefits of AI scribes, their limitations, and his belief that their most significant contribution may be beyond mere time-saving.
Related content: Take note: The AI scribe era is here
Can AI scribes effectively address the pajama time dilemma?
Wachter was cautious in his expectations regarding the documentation benefits of AI scribes. While they indeed alleviate some of the time doctors spend on documentation during “pajama time,” the influx of patient messages and portal communications that fill evening hours requires separate solutions. More importantly, he emphasized that the true impact of AI scribes lies in enhancing the in-appointment experience.
The introduction of electronic health records (EHR) altered the nature of patient encounters significantly. It created an audience—Medicare, insurers, malpractice attorneys, and quality measure administrators—all seeking greater insight from clinical notes, thus inflating their complexity. Physicians found themselves functioning more like “expensive, grumpy data entry clerks,” as Wachter put it. AI scribes hold the potential to relieve much of this burden, allowing doctors to reclaim eye contact, presence, and the ability to listen actively.
“This technology is designed to give physicians back not only their time but the very humanity of the doctor-patient visit,” he remarked.
The time savings, however, have proven to be more modest than some had anticipated. Recent studies, including one published in
Why is documentation a fitting starting point for AI in medicine?
Wachter describes AI scribes as intentional “training wheels.” Their initial deployment was chosen not because it’s the most pressing issue AI can address in healthcare, but because it occupies a space on the two-by-two matrix that is relatively simple to execute and carries minimal risk if mistakes occur.
He draws parallels to the first wave of AI in healthcare during the late 1970s and early 1980s, when the founders focused on diagnosis—a challenging problem. The technology and necessary infrastructure were not sufficiently matured, leading to eroded trust in AI when it failed.
“You don’t tackle the hardest challenge first,” Wachter explained. “You address simpler, manageable problems to build experience and trust.”
Scribes fit into this strategy perfectly. UCSF already had a model utilizing human scribes, often pre-medical students, documenting for nearly a decade. The AI version had a clear lineage, identifiable problems to fix, and a low risk of significant harm from mistakes. Errors could occur in notes, but they are manageable.
Currently, about 70% of UCSF’s roughly 3,000 to 4,000 physicians utilize AI scribes, and a staggering 90% of them express enthusiasm for the tool.
“We’ve nearly reached a point where, if we were to discontinue its use, we could potentially lose a significant number of doctors,” Wachter noted.
What limitations do AI scribes have?
Wachter acknowledged several constraints. AI scribes currently struggle to adapt to different sub-specialty documentation styles. The narrative component of physical exams, a task physicians are unaccustomed to doing, remains a challenge. Some doctors have tested these tools, found the error rates troubling, and opted to revert to traditional methods.
He raised an additional nuance: there may be cognitive benefits from the act of writing notes that might be lost with AI-generated drafts. Physicians sometimes identify overlooked details or relationships between symptoms during documentation that might not be as apparent when simply reviewing an AI-generated summary.
“There’s something valuable lost when I’m not typing my own notes,” he confessed, acknowledging the difficulty in quantifying this loss.
The notion of a “human in the loop,” commonly proposed to alleviate concerns about AI safety, deserves more critical evaluation. If a physician reviews multiple flawless notes consecutively, their attention to the next one is likely to wane. Over-reliance on these tools presents a genuine risk, and healthcare systems have yet to develop comprehensive strategies to mitigate this risk, though Wachter noted that discussions about innovative approaches, such as embedding deliberate errors to maintain alertness, are in progress.
Are AI scribes beneficial for physician practices?
The financial return on investment for AI scribes today is less about throughput and more about the physician experience. A recent study published in
Considering the financial impact of replacing a primary care doctor, a tool that enhances physician satisfaction and retention fundamentally alters the financial equation.
Wachter posited a larger strategic argument: AI scribes are merely the “singles,” while more impactful technologies related to clinical decision support, diagnosis, test selection, and treatment are still on the horizon. Building trust is essential, and AI scribes facilitate this process.
“The ultimate value of scribes, when the narrative is fully developed, will be found in winning hearts and minds, cultivating openness for more expansive tools,” he asserted.
For a deeper understanding, listen to our complete episode of
