In a recent editorial in the Monash Bioethics Review, researchers Marco Paglialonga and Cristiana Simonetti argue that integrating AI technology with the humanities “is key to promoting patient-centered care while preserving the core values of therapeutic relationships.” This could mean AI-generated discharge instructions that patients actually understand, chatbots that respond to after-hours questions with warmth, or decision support systems that enhance rather than erode the patient-clinician relationship. They ask us to see AI in healthcare not as a human substitute, “but as a tool to be humanized” where “human dignity is not sacrificed to efficiency but defines its purpose.” They note that if we can use this tool well it holds the potential “to amplify patient voices, break communication barriers, and restore dignity to those excluded from clinical dialogue.”
This vision for humanism in AI in medicine mirrors the Gold Foundation’s vision of keeping healthcare human. The Gold Foundation works toward this vision with a mission to foster kindness, safety, and trustworthiness. This Research Roundup applies that mission lens to AI in healthcare. Below, you’ll read about three studies exploring the potential of AI to enhance patient safety, foster kindness in the clinical encounter, and support clinician trustworthiness. Wherever you are on the spectrum from excited to curious to cautious to alarmed about the future of AI in medicine, these studies clarify that all of us — practitioners, patients, researchers, educators, and administrators — have a role to play in shaping that future.
Safety
Studies have shown that AI can generate patient communication materials; this study from Germany asks how well it does so, with implications for AI’s ability to increase discharge process safety.
What’s going on with me and how can I better manage my health? The potential of GPT-4 to transform discharge letters into patient-centered letters to enhance patient safety: Prospective, exploratory study. Eisinger, F., Holderried, F., Mahling, M., Stegemann–Philipps, C., Herrmann–Werner, A., Nazarenus, E., Sonanini, A., Guthoff, M., Eickhoff, C., and Holderried, M. (2025). Journal of Medical Internet Research 27: e67143. Access the Free Article
What: This study explores AI’s promise to provide “personalized and scalable support for helping patients understand medical information” by asking: How well does GPT-4 identify and transform patient safety-relevant information from traditional discharge letters into patient-centered letters? Authors developed discharge letters for 3 common medical conditions containing both “remember” and “understand” levels of educational content. Due to variability in AI output, they generated each letter 5 times using the same prompt. Two clinicians analyzed letters for medical quality, patient centricity, and ability to convey safety-related information. Overall, GPT-4 did not fully demonstrate that promise.
So What: A positive: GPT-4 letters significantly outperformed traditional discharge letters in patient-centricity — they were easier to read, contained fewer abbreviations and medical terms, and more directly addressed the patient. The drawbacks: Approximately 4% of AI-generated sentences contained medical errors (from imprecision or incompleteness), and no AI letters included all relevant information (78% of learning objectives were captured). “Understand” level information was omitted more often than “remember” instructions, with more frequent omissions for complex medical requirements.
Now What: The authors conclude that while GPT-4 shows potential to enhance patient-centeredness of discharge letters, it is not yet suitable for patient care without medical professional review, particularly given the omissions and occasional hallucinations (factually incorrect AI-generated content). Further advances in prompting techniques and targeted development of medical language models could minimize these limitations. If addressed, GPT-4 could meaningfully support healthcare professionals in patient-centered communication and improve patient understanding — a significant step toward better patient safety and care quality.
Kindness
The prior study found that AI excelled at being patient-centric. Such findings suggest our concerns about AI making medicine less human might be misplaced. This UK study provides support for that conclusion.
AI chatbots versus human healthcare professionals: a systematic review and meta-analysis of empathy in patient care. Howcroft, A., Bennett-Weston, A., Khan, A., Griffiths, J., Gay, S., and Howick, J. (2025). British Medical Bulletin 156, no. 1: ldaf017. Access the Free Article
What: This systematic review examined 15 empirical studies comparing conversational AI chatbots using large language models with human healthcare professionals on empathy measures. Studies included real patients, healthcare users, or patient-authored communications (emails, portal messages, public forums). All used blinded evaluations (except one unspecified); 14/15 used non-validated empathy instruments; 14/15 were published in 2024. Participants functioned as observers rather than active chat participants. Thirteen studies reported statistically significantly higher empathy ratings for AI; two dermatology studies favored humans. Meta-analysis of 13 studies found AI rated approximately 2 points higher on 10-point empathy scales.
So What: With 73% likelihood of AI being perceived as more empathetic than human practitioners, these findings suggest AI-driven interactions are unlikely to cause harm through empathy deficits. The text-based nature limits transferability, though this limitation may diminish as more patient-practitioner interaction occurs via text.
Now What: While methodological limitations (unvalidated scales, text-only evaluations) temper these results, the evidence challenges longstanding assumptions about human clinicians’ exclusive capacity for empathic communication. The authors call for future research using voice-based interactions and direct patient feedback, with rigorous validation through randomized trials to ensure clinical reliability.
Trustworthiness
Just as AI’s empathy depends on how information is presented, its trustworthiness depends on factors beyond technical accuracy. This study explores what influences patients’ willingness to trust AI-supported clinical decisions.
The effect of artificial intelligence on patient-physician trust: Cross-sectional vignette study. Zondag, A., Rozestraten, R., Grimmelikhuijsen, S.G., Jongsma, K.R., van Solinge, W.W., Bots, M.L., Vernooij, R.W., and Haitjema, S. (2024). Journal of Medical Internet Research 26: e50853. Access the Free Article
What: This study out of the Netherlands explored acceptability of AI-based Clinical Decision Support Systems (CDSS). 398 participants were randomized to four vignettes: clinician decision-making with or without AI support, in both low-risk (rheumatoid arthritis flareups) and high-risk (neonatal sepsis) scenarios. Intervention vignettes explicitly stated the doctor received support from an advanced computer system combining medical information for risk prediction. Participants completed a modified “Trust in Physician” scale measuring three dimensions: competence, integrity, and benevolence.
So What: In high-risk vignettes, women reported lower trust in the AI intervention group, with lower perceptions of physician competence and integrity. Authors found positive correlation between general technology trust and measures of benevolence and integrity in low-risk intervention groups. This suggests higher-stakes scenarios may yield greater patient preference for traditional human decision-making.
Now What: These findings reveal that trust in AI-supported care isn’t just about technological competence — it’s deeply influenced by patients’ prior experiences, gender, risk level, and broader trust in technology. Building trustworthy AI in healthcare requires addressing these human factors alongside technical accuracy. For successful AI adoption in clinical practice, patients must be involved in both development and implementation, and broader societal discussion about human values and AI in healthcare is needed.
These three studies demonstrate that AI in healthcare is neither inherently humanistic nor dehumanizing — it is a tool that amplifies the intentions of its designers and users. The challenge ahead is ensuring that patient voices, clinical wisdom, and humanistic values shape AI’s development and deployment, rather than allowing efficiency and automation to define its purpose.
Researchers, clinicians, administrators, and others in the United States committed to the Gold Foundation’s mission should work toward bringing the insights and patient-centered focus of the international authors cited above to a U.S. context to ensure that our healthcare system develops AI not as a replacement for human connection, but as a tool that amplifies our capacity for kindness, enhances our commitment to safety, and earns the trust of the patients we serve.
To receive email notification of future editions of the Jeffrey Silver Humanism in Healthcare Research Roundup, enter your information here and select “Research Roundup” from the checkboxes at the bottom. Read previous posts in this series.
