The authors developed a pilot curricular innovation to prepare postgraduate Internal Medicine residents to integrate artificial intelligence (AI) tools responsibly into patient care. The curriculum aimed to teach foundational prompting strategies for AI-assisted clinical reasoning and the ethical and legal considerations relevant to clinical AI use. The intent was to move beyond learner attitudes and assess both perceived knowledge and objective knowledge outcomes.
The educational intervention, delivered during the 2025–2026 academic year, consisted of two linked components: an initial didactic and a subsequent hands-on, interactive small-group session. The didactic introduced core concepts of ethical AI use, legal considerations such as physician liability and informed consent, and fundamentals of effective prompting for AI systems. The interactive component integrated AI prompts and AI outputs into clinical vignettes to simulate real-world decision-making.
Teaching strategies emphasized near-peer facilitation and situated learning: trainees worked through cases in small groups with guidance that mirrored clinical workflows. Figures included in the full text illustrate the session agenda and the small-group activity structure used to achieve session objectives.
The pilot was implemented with Internal Medicine resident physicians at a single academic site. Of 96 eligible residents, 59 (61.5%) completed the presession survey and 52 (54.2%) completed the postsession survey. Fifty-one residents participated in both the pre- and post-session knowledge tests, enabling paired assessment of objective knowledge change.
The study used a pre–post intervention design. Perceived knowledge across five predefined learning objectives was assessed via surveys administered before and after the session. Objective knowledge was measured using an 8-item knowledge test given before and after the intervention. Statistical comparisons used the Wilcoxon Rank-Sum test. Effect sizes were reported for changes in perceived knowledge and for the objective test score changes.
Perceived knowledge increased significantly across all five learning objectives, with effect sizes described as moderate to large in the abstract. For objective knowledge, the median pre-session test score was 6 out of 8 (interquartile range [IQR] 4–8). The median post-session score rose to 7 out of 8 (IQR 5–8). This improvement reached statistical significance (p < 0.001) with a moderate effect size reported as 0.33. Visual data in the article include a radar overlay comparing pretest and posttest performance by item and a scatter plot showing item discrimination and difficulty indices for the posttest.
The combined didactic and interactive small-group format improved both residents’ self-reported understanding and measurable knowledge related to prompting strategies and the ethical and legal aspects of clinical AI use. The authors position this curriculum as a feasible approach to equip trainees with foundational competencies needed in an AI-integrated clinical environment. They underscore the relevance of teaching both technical prompting skills and non-technical domains such as informed consent and physician liability when AI tools influence patient care decisions.
The abstract reports results from a single-site pilot with partial survey response rates; detailed information about long-term knowledge retention, translation to clinical behavior, or patient-level outcomes was not reported in the abstract. The authors note that future work should focus on developing clinical assessments of trainee skills using AI tools to evaluate competence in applied settings. Additional details regarding curricular duration, facilitators’ training, exact assessment items, and how legal concepts were operationalized are not reported in the abstract and would require consultation of the full text.
The authors declared no conflicts of interest in the published report.
Note: This summary is based on the abstract and accessible information from the PubMed/PMC record. Full methods, assessment instruments, and additional results are available in the full text for readers seeking detailed procedural and psychometric data.