Physician burnout is a public health problem that affects a high proportion of practitioners. It is linked to early departure from the profession, reduced productivity, more frequent medical errors, and lower patient satisfaction1. There is no established definition of burnout, but criteria typically include emotional exhaustion, depersonalization, and diminished sense of accomplishment. Female, young, and emergency room providers appear particularly prone, with psychiatrists and ophthalmologists relatively less impacted1. While the overlap with depression is clear, burnout is considered job-related, whereas depression is less context-dependent, although the greater stigma of psychiatric diagnoses may have helped popularize burnout as a more “acceptable” condition. The tangled psychological, demographic, specialty and systemic factors have made solutions elusive and prompted some to look to artificial intelligence (AI) for innovative remedies that could address physician burnout across its myriad fronts and contributors2, 3. But what if the very tools we hope will alleviate burnout inadvertently create new, more insidious sources of strain? Analyzing AI's impact on time and mental demands – two key dimensions of the NASA Task Load Index4, a standard tool for assessing workload in health care – provides a path for understanding its effects on burnout. A critical factor in high-burnout specialties is overwhelming time demands, characterized by having insufficient time to see patients and meet the charting, billing, credentialing, continuing education, and other administrative and teaching responsibilities, leading to work extending into personal and family time. AI might alleviate time demands by automating clinical charting, billing forms, medical leave requests, and insurance correspondence. Automating note generation, for example, could free up a significant portion of a physician's day, reducing the need to extend work hours. By using ambient scribes to write summaries or natural language tools to access patient data (e.g., “give me this patient's hospitalization history”), AI could make physicians more efficient. AI could also aid in efficiently communicating with patients by adjusting text to match literacy level. In the emergency department's high-burnout context, AI could improve workflow by prioritizing critical cases, rapidly identifying urgent findings on imaging studies, and performing emergency triage tasks, thereby speeding up initial patient assessments. The resulting time savings could mean more time spent in direct patient interactions – an activity that has been considered protective against burnout5. However, it is also possible that some new tools may end up becoming an AI-age example of the “productivity paradox”, a long-identified problem with adoption of information technologies in medicine6. If AI tools are poorly designed, not seamlessly integrated into existing workflows and electronic health records, or have usability issues, they could introduce new inefficiencies, disrupt established routines, diminish physicians' control over their day, and increase time demands. The initial phase of implementation of AI systems also often requires a time investment for training, which could temporarily accentuate time pressures. In the case of AI scribes, high “hallucination” rates necessitate that physicians spend considerable effort editing and correcting, potentially requiring more time than if the assessment had been done manually. Mental demands are another common source of physician burnout. By using clinical decision support systems that increase the efficiency of information synthesis, present critical data clearly, offer diagnostic suggestions, propose treatment interventions, and highlight potential risks, AI can reduce the cognitive load associated with complex case management. Similarly, by enhancing information retrieval from dense electronic health records through data visualization, and by automating the tracking of patient progress and pending results, AI can allow physicians to maintain better situational awareness with less mental effort. Early research suggests significant potential for AI in this arena, with large language models demonstrating a strong capacity for complex medical reasoning and multi-step clinical decision-making7. But AI tools can also increase mental demands, including via “alert fatigue” from excessive or low-specificity notifications or a need for hyper-vigilance. It is entirely possible that physicians will increasingly face an overwhelming volume of AI-generated data that requires extensive cognitive processing to assess for relevance and accuracy. While AI tools may reduce cognitive load in one area, it is conceivable that overall cognitive effort may actually increase due to the need to vigilantly monitor AI outputs for errors that can negatively impact care or make it into the medical record, thereby also increasing practitioners’ medico-legal risks. Such vigilance will likely need to heighten to confront a dangerous phenomenon being borne out in research, namely the tendency to over-trust AI-generated medical advice when it appears to “think” or when it generates inaccurate but believable-seeming citations to support its reasoning8. To avoid this automation bias and prevent medical errors that can harm patients or lead to malpractice claims against doctors, physicians may be forced into a taxing, high-alert state that compounds mental strain. In the context of an overstretched health care system, it is not surprising that AI has been seen as a potential panacea. In the most optimistic scenario, AI could automate burdensome or tedious tasks, freeing up precious physician time. AI may also act as a powerful cognitive partner, assisting with complex diagnostics and personalizing management and care. Still, it would be simplistic to assume a mostly positive outcome on burnout from the medical AI tools that many health systems are rolling out. The integration of new technologies into medical practice has often been a double-edged scalpel, and the widespread adoption of electronic health records nearly 20 years ago is one cautionary lesson. They were intended to streamline information and strengthen care coordination, and early versions were lauded for their real-time alerts and helpful reminders9. Since then, however, electronic health records have emerged as a rather consistent culprit in physician burnout studies, largely due to the substantial increase in administrative tasks that they have enabled, cumbersome system interfaces, and workflows that detract from direct patient engagement and extend physicians’ workdays9. Such stories should serve as a reminder that tools designed to improve care can inadvertently introduce new burdens if not implemented with careful consideration for the human end-user and without an attempt to predict, and protect against, potential downsides. AI holds meaningful promise to alleviate some drivers of physician burnout, but the promise is contingent on addressing significant challenges. These include ensuring high levels of accuracy to prevent increased verification burdens or medical errors, mitigating the risk of over-trust in imperfect systems, ensuring seamless workflow integration to avoid new inefficiencies, controlling the “task creep” that may happen because a new technology has now made new tasks possible, and protecting against the more insidious intrusions suddenly feasible, such as increased surveillance of physician productivity by AI-empowered administrators. There is also the real concern of deskilling, where over-reliance on AI could reduce physician proficiency, and the psychologically destabilizing fear of total physician displacement by AI in some medical specialties. For medical AI tools to be a genuine solution rather than another source of physician burnout – or even a trigger for a mental health crisis – the development and deployment of these systems must be physician-centered and informed by high-quality research. Along the way, input from psychiatrists, who seem to have lower burnout rates themselves and may understand burnout particularly well due to its overlap with depression, could be uniquely helpful. Until then, circumspection around the adoption of medical AI as a sort of anti-burnout cure-all would seem in order.
Bouguettaya et al. (2026) studied this question.