Invited Commentary | Ethics Integrating Artificial Intelligence Support in Patient Care While Respecting Ethical Principles Marianne Sharko, MD, MS; Curtis L. Cole, MD Responding to portal messages is generally considered cumbersome and time-consuming for physicians—perhaps an ideal task to be relegated to generative artificial intelligence (AI) support tools. Research has shown that AI can generate messages that are high quality and acceptable to patients and at times perceived to be more empathetic than physician-generated responses.1-4 This support can alleviate burdens on physicians and help address burnout. However, it is important to consider how patients feel about the use of AI-generated messages and their impact on patient- physician relationships. The study by Cavalier et al5 surveyed participants from a patient advisory group to assess attitudes toward AI- and physician-generated portal messages and authorship disclosure. While participants were generally satisfied with the responses, they preferred the ones written by AI, which tended to be longer and more detailed and were perceived as more empathetic. However, participants were less satisfied if they learned that these emails had been authored by AI. Participants were more satisfied if they were told that a physician wrote the email or if they were told nothing about the authorship. Reasons for the preference for physician-authored messages are not entirely clear. Perhaps patients value the personal connection, emotional response, reassurance that the physician cares about them, or knowledge that their physician has reviewed their data and ensured the accuracy of the message. Given current expectations that response messages are written by physicians, a disclosure of AI authorship may feel like a deception or may be perceived as a lack of care by their physician. Interestingly, the authors found that comfort with AI-generated messages was not associated with seriousness of the messages. This needs more exploration. The biggest value to physicians would likely be from automating the more voluminous routine messages. However, what is routine to a physician may be quite novel to an individual patient. To address this without violating trust or harming the intimacy of the patient-physician relationship, more time needs to be allotted for physicians to read through their messages and create longer, more detailed, and more empathetic responses when needed. This would be a win-win for the physicians and the patients. In addition, we must ask whether we are using AI for a task that should be done by a physician rather than using it for other burdens that are less worthy than expressing care for patients. If physicians need more time, why is talking to patients given up rather than the host of other nonclinical work that more rightfully should be done by a machine? The burden of messaging is a proximal cause of burnout in a larger context of how health care is provided today. The reality is that physicians are not supported to spend time crafting thought-out message responses. If patients and physicians value physician-authored messages, support for physician time spent creating responses needs to be acknowledged and valued, possibly through revised reimbursement policies. This study raises timely issues on the use and disclosure of AI support in the provision of patient care. Do patients have the right to know when AI is being used? As AI support in health care becomes more widespread, we need to establish acceptable uses of AI, determine when personal patient- physician interaction should be prioritized, and create guardrails around the use and disclosure of AI- generated support. The authors include a strong discussion of ethical guidelines that support the disclosure of AI-generated information to patients. Organizations, including the World Health Organization + Related article Author affiliations and article information are listed at the end of this article. Open Access. This is an open access article distributed under the terms of the CC-BY License. JAMA Network Open. 2025;8(3):e250462. doi:10.1001/jamanetworkopen.2025.0462 (Reprinted) March 11, 2025 1/3 Downloaded from jamanetwork.com by Weill Cornell Medical Library user on 05/02/2025 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2025.0449&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 (WHO), have addressed ethical issues related to the use of AI in health care.6 WHO principles to guide AI regulation include ensuring that information about the use of AI is transparent and easily understood to promote public discussion on its use. Transparency of AI use will prevent misinterpretation, foster patient trust, and ensure authenticity. However, it is not clear what constitutes the type of AI-generated support that needs to be disclosed. Does the use of spelling or grammar corrective tools qualify for disclosure? Imagine a scenario in which a physician has reviewed a patient’s data and wants to communicate information and empathy through a portal message. If the physician uses a large learning model to determine an effective way to express empathy in a portal message, does that need to be disclosed? If so, does that minimize the physician’s empathy and personal efforts? As the authors point out, attitudes toward AI use and disclosure will likely change as AI support becomes more widespread in patient care. Patients may become less surprised and disappointed to find out that messages have been generated by AI. The delineation of AI use that necessitates disclosure may be a moving target as we move forward with the use of AI. This study identified current patient expectations for physician authorship of portal messages. Given that AI authorship is relatively new, its use should be disclosed to avoid misrepresentation and to increase patient awareness. If the use of AI tools for this purpose becomes more standard and expected, there may come a time when its use is assumed, and disclosure is not necessary. We may want to consciously guide expectations to help patients appreciate why AI authorship could be helpful for messages that physicians consider routine. Future research will help mark changes in patient expectations. Attitudes may also be related to generational differences. Participants in this study tended to be older. Perhaps younger patients would be more satisfied with an AI-supported email message. If so, could this represent broader shifts in patient-physician expectations? Likewise, attitudes might differ among generations of physicians, a variable we have already seen in electronic health record adoption.7 As AI use becomes more prevalent, it will undoubtedly impact patient-physician expectations. Will broader use be associated with decreased value of patient-physician communication? Or will it enable increased opportunities for physicians to focus on tasks that support patient communication? This study highlights patient preferences for human connection translated through health care tasks. Physicians need to advocate for AI integration that improves medical care and decreases physician burnout while preserving patient-physician connection and respecting patient autonomy. Patients and physicians need to have a say on the role AI-generated support should play in patient care and the development of policies on ethical disclosure. Being transparent about AI support will help foster trust in physicians and patient awareness of the increased role of AI. Physicians need to be actively involved in developing policies that will leverage the power of AI-generated support while continuing to prioritize patient-physician connection and empathetic care. ARTICLE INFORMATION Published: March 11, 2025. doi:10.1001/jamanetworkopen.2025.0462 Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2025 Sharko M et al. JAMA Network Open. Corresponding Author: Marianne Sharko, MD, MS, Weill Cornell Medicine, 425 E 61st St, New York, NY 10065 (sharkom@med.cornell.edu). Author Affiliations: Department of Pediatrics and Population Health Sciences, Weill Cornell Medicine, New York, New York (Sharko); Department of Medicine and Population Health Sciences, Weill Cornell Medicine, New York, New York (Cole). Conflict of Interest Disclosures: None reported. JAMA Network Open | Ethics Integrating AI Support in Patient Care While Respecting Ethical Principles JAMA Network Open. 2025;8(3):e250462. doi:10.1001/jamanetworkopen.2025.0462 (Reprinted) March 11, 2025 2/3 Downloaded from jamanetwork.com by Weill Cornell Medical Library user on 05/02/2025 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2025.0462&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://jamanetwork.com/pages/cc-by-license-permissions/?utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 mailto:sharkom@med.cornell.edu REFERENCES 1. Goodman RS, Patrinely JR, Stone CA Jr, et al. Accuracy and reliability of chatbot responses to physician questions. JAMA Netw Open. 2023;6(10):e2336483. doi:10.1001/jamanetworkopen.2023.36483 2. Kim J, Chen ML, Rezaei SJ, et al. Perspectives on artificial intelligence-generated responses to patient messages. JAMA Netw Open. 2024;7(10):e2438535. doi:10.1001/jamanetworkopen.2024.38535 3. Garcia P, Ma SP, Shah S, et al. Artificial intelligence-generated draft replies to patient inbox messages. JAMA Netw Open. 2024;7(3):e243201. doi:10.1001/jamanetworkopen.2024.3201 4. Tai-Seale M, Baxter SL, Vaida F, et al. AI-generated draft replies integrated into health records and physicians’ electronic communication. JAMA Netw Open. 2024;7(4):e246565. doi:10.1001/jamanetworkopen.2024.6565 5. Cavalier JS, Goldstein BA, Ravitsky V, et al. Ethics in patient preferences to artificial intelligence–drafted responses to electronic messages. JAMA Netw Open. 2025;8(3):e250449. doi:10.1001/jamanetworkopen. 2025.0449 6. Stephenson J. WHO offers guidance on use of artificial intelligence in medicine. JAMA Health Forum. 2021;2(7): e212467. doi:10.1001/jamahealthforum.2021.2467 7. DesRoches CM, Campbell EG, Rao SR, et al. Electronic health records in ambulatory care—a national survey of physicians. N Engl J Med. 2008;359(1):50-60. doi:10.1056/NEJMsa0802005 JAMA Network Open | Ethics Integrating AI Support in Patient Care While Respecting Ethical Principles JAMA Network Open. 2025;8(3):e250462. doi:10.1001/jamanetworkopen.2025.0462 (Reprinted) March 11, 2025 3/3 Downloaded from jamanetwork.com by Weill Cornell Medical Library user on 05/02/2025 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2023.36483&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2024.38535&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2024.3201&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2024.6565&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2025.0449&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamanetworkopen.2025.0449&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://jama.jamanetwork.com/article.aspx?doi=10.1001/jamahealthforum.2021.2467&utm_campaign=articlePDF%26utm_medium=articlePDFlink%26utm_source=articlePDF%26utm_content=jamanetworkopen.2025.0462 https://dx.doi.org/10.1056/NEJMsa0802005