Why using AI saves doctors time but won’t fix physician burnout

AI adoption in medicine is accelerating: more than 12,000 primary care clinicians have enrolled in Canada Health Infoway’s national AI scribe program, about 5,000 Ontario physicians use AI scribes and roughly one in five Canadian physicians now use AI tools. Almost 60 per cent report a cut in administrative time. AI scribes are tools that, with the patient’s consent, listen to a clinical visit and automatically draft the physician’s notes. We’re writing as an emergency physician who has led digital transformation in emergency departments for six years and advises on AI readiness (Ali Abdalvand), and an organizational psychologist studying burnout and responsible AI adoption (Ehsan Etezad). Our observation is that AI keeps its narrow promise of saving time. But without proper frameworks, it quietly adds new burdens that absorb those savings. We believe the fix is organizational. Burnout stems from mismatches in workload, control, reward, community, fairness and values; not from any single task such as paperwork. One study found electronic health record design and use accounted for just 6.8 per cent of the differences in burnout between clinicians. When work conditions such as a chaotic atmosphere and lack of control over workload were also considered, that number rose to 36.2 per cent. In emergency medicine, physicians are interrupted 7.9 times an hour, and prescribing error rates nearly triple with interruptions. No scribe changes that. Emergency physicians are interrupted 7.9 times an hour, and prescribing error rates nearly triple with interruptions. (Pexels/Michael Berdyugin) How AI quietly adds to physicians’ work I think we can all relate to the idea that when you complete one task efficiently and save some time, new demands fill the space, making it seem as though nothing was saved in the first place. But as AI introduces new ways of reducing the time physicians spend on documentation, it comes with novel ways of adding work, too. The time dividend gets spent. Saved time rarely comes back to clinicians. In Canada’s national AI scribe program, half of clinicians put the time they saved into other administrative work, and one in five used it to add appointments. Studies from the United States show the same pattern: AI scribe adopters across five health systems saved 13.4 minutes of daily EHR time and delivered 0.49 more visits per week. Errors do happen and patient safety is at stake. In two simulated tests, Ontario’s auditor general reported that most of the 20 government-approved AI scribes made errors. Twelve recorded the wrong drug, nine invented treatment steps and 17 missed key mental-health details in at least one test. The province says there are no known reports of patient harm. OntarioMD, a subsidiary of the Ontario Medical Association that helps physicians adopt digital health tools with funding from the Ontario Ministry of Health, responded that physicians “know they must review every clinical note generated by their AI scribe.” That safeguard is essential — and it is also work. Used to be authors and now editors. Automation often leaves humans the hardest job, which is watching for rare but serious errors. When physicians at one U.S. health system got AI-drafted replies to patient messages, replying time did not decline as physicians still have to edit the text to make sure it was accurate. More tools, more switching. In our experience, many AI tools sit outside core clinical systems. As an example, in one of our emergency departments, starting a night shift means logging into the work station, the health record, the dictation system, the imaging viewer, the EKG viewer and, most recently, an AI scribe; six systems before seeing the first patient, and a copy-and-paste at the end of every interaction because the scribe lives outside the chart. Thirty-nine per cent of Canadian physicians named better-connected, interoperable systems as a priority. Always learning because it’s always changing. Keeping up with constant learning is a job demand in its own right and it can wear people down. This is also known as AI fatigue, defined as the overwhelming pressure of keeping up with AI tools for people’s jobs. Change fatigue is a predictable response to how change is managed by the health-care system and health-care leaders and it is not a reflection of someone’s failing to learn or to adapt. A better way We propose a practical framework that can be used by health-care organizations to improve the adoption of AI. This framework, GRACE, was co-developed by one of the authors, Ehsan Etezad, and it sets out five principles for moving toward responsible adoption. GRACE framework for responsible adoption of AI. (Ehsan Etezad), CC BY Govern adoption. Include front-line clinicians on AI committees. Give every pilot a stop-rule written in advance. If the pilot doesn’t return measurable time, or audits find errors above an agreed threshold, it ends or is redesigned. Respect autonomy. Keep use voluntary and let clinicians choose among vetted tools. Respecting autonomy is an essential component of responsible AI adoption. Advance access. Build tools into the core clinical record so no one has to copy and paste. Fund access fairly. Cultivate competence. Offer protected, paid learning time and peer “AI champions.” Evaluate effectiveness. Track burnout, after-hours work, note accuracy and whether time actually comes back. Another practical tool, the AI Adoption Matrix, maps organizations on two dimensions. Enablement is how broadly people can access, use and build skills with AI. Governance is how well policies, oversight and accountability work. We encourage health-care organizations to map themselves on this quadrant and work toward being “Responsible Pioneers” with GRACE. We encourage health-care organizations to map themselves on this quadrant. (Ehsan Etezad), CC BY The real test There is increasing pressure to move fast with AI adoption in health-care systems and it will likely continue. We would like to clarify that we’re not arguing against adoption of AI in health care. We actually encourage it and are excited about the future it will bring to all of us. That said, we encourage health-care organizations to adopt AI responsibly because stakes are high in medicine and patient safety could be at risk. We believe success should be measured by how happy a physician can feel at their job and about their job, not just how many patients they can see in one shift. None of us wants our loved ones to be treated by an exhausted and overwhelmed physician, so let’s ask ourselves how can we reimagine a health-care system that won’t burn out our physicians.