In a 2021 American Family Physician editorial, Drs. Liaw, Kakadiaris, and Yang shared a future in which family physicians are aided by artificial intelligence (AI) to provide better care, devote more time to patients, and spend less time on administrative tasks.¹ Since then, the future has increasingly become the present with the advent of ChatGPT and other Large Language Models (LLMs) finding their way into the physician-patient encounter. Based on one's experience, AI may be viewed as the cause of the inevitable dehumanization of health care or the solution that restores the joy of practice by assisting with point-of-care clinical decision-making and decreasing administrative burden.

A recent study estimated that for 2,500 patients, a primary care physician working alone would require 26.7 hours each day to provide preventive, chronic, and acute care; document that care; and attend to the electronic inbox. In a team setting, with many tasks delegated to others, that same physician would require 9.3 hours daily.² Another study showed that primary care physicians spend more time overall and more time after hours in the electronic health record than those in other medical and surgical specialties.³ Greater administrative burden has been directly linked to physician burnout. A systematic review found an average burn-out prevalence of 35% among family physicians and even higher burnout in those who spend more time on patient documentation and feel like they have less control over their workload.⁴

Health care practices are increasingly turning to LLMs, the backbone of chatbots, to help decrease administrative burden and combat burnout. LLMs are trained on a huge amount of data to be able to predict the next correct text in a sequence, which allows them to simulate the type of output a human would generate. They are already being used to dictate notes with a high level of accuracy, decreasing the time physicians spend documenting visits.⁵ Other LLMs are used to monitor patients with specific health conditions, engaging them in chats that are elevated as needed to a physician or other team member based on a patient's responses.⁶ Still other chatbots generate draft replies to nonemergent patient messages for a member of the clinical team to edit (or delete) before responding, saving the team time while maintaining the patients' electronic access to their primary care team.⁶

Helping with documentation, patient monitoring, and patient inquiries is a solid series of first steps for LLMs in primary care. LLMs are being explored as clinical decision support tools. A future iteration will no doubt see LLMs integrated into our electronic health records. During a clinical encounter, the LLM will quickly review a patient's prior health care information and, based on how the appointment evolves, suggest diagnostic possibilities, recommended ancillary evaluations, and possible treatment strategies—with orders ready to be signed depending on the chosen settings. Taking into account the totality of a patient's history and presenting options for consideration will help physicians make even better-informed decisions, taking the routine work off of their hands and giving them more time to do the things that only a family physician can do—complex medical evaluation and decision-making conversations performed in a comprehensive, contextualized way. In addition, LLMs may find a place in patient care between appointments, helping to close preventive health service gaps and monitoring and responding to chronic care conditions.⁷

The use of LLMs in primary care is not without risks. The World Health Organization has recognized that the data used to train AI systems may be biased against certain people groups, LLMs may generate authoritative yet wrong answers to queries, and LLMs can be made to generate dis-information.⁸ In addition, some have voiced concerns that LLMs may come to minimize, or possibly eliminate, the family physician's role on health care teams.⁹ Some believe that in the drive for profit, health systems may choose to replace family physicians with other AI-enhanced clinicians, such as advanced practice providers, who are allowed to practice with a greater scope than they had previously. These concerns are valid. However, with rare exceptions, technological change has only moved us forward. It does not appear that backward is even an option—the ship has sailed and is underway.

The best defense against AI risks becoming realities is conscientious physicians guiding the development and implementation of LLMs into clinical care settings, pointing out what LLMs can do and what they cannot. In family medicine, no LLM can yet address a complex patient in a unique sociocultural situation with overlapping comorbidities and health states from the vantage point of a longitudinal relationship.

LLMs in primary care are here now, and their presence will only increase over time. The key to making them work for physicians and patients is for family physicians to be involved in their implementation, their evolution, and the assessment and research surrounding them. We do this by engaging with and actively beta testing new technology in our practices, educating ourselves so that we can stay abreast of and even contribute to progress, and collaborating with our colleagues in medical informatics to help them operationalize the tools that are being placed in our patient care areas. Without our guidance, these systems may not meet the needs of our patients or our practices and could potentially be yet another electronic tool taking us further from our core mission of patient care. If we as a community do not take ownership of the development of these systems, others will, and we may not like the end result.

Editor's Note: Dr. Saguil is an assistant medical editor for AFP.