Changes in medical students’ ethical and acceptability-related perceptions of an AI-supported doctor–patient communication learning pathway: a longitudinal empirical ethics survey

Artificial intelligence (AI)-supported learning is increasingly used in medical education, including doctor–patient communication training. Such systems may provide virtual dialogue practice, immediate structured feedback, reflection prompts and opportunities for further practice after feedback. However, educational usefulness does not necessarily imply ethical acceptability. This study examined changes in medical students’ ethical and acceptability-related perceptions before and after participation in an AI-supported doctor–patient communication learning pathway. This longitudinal empirical ethics survey was nested within a course-based randomised educational trial but was analytically independent of the parent trial’s educational outcome analysis. Surveys were administered after random allocation to the AI-supported pathway but before exposure to AI-supported learning functions (T0), at the end of the 8-week course (T1), and at the 3-month follow-up (T2). The main analysis focused on students in the AI-supported pathway with complete T0–T1–T2 ethics-related scale data and analysable T2 open-text responses ( n = 31). The open-text question was included only at T2. The primary comparison was the within-participant T0–T2 change across six ethics-related domains. Two domains showed statistically significant T0–T2 changes. Trust in AI feedback decreased from 3.86 (0.55) to 3.20 (0.62), with a mean change of − 0.66 (95% confidence interval [CI] − 0.88 to − 0.42), Cohen’s d z = − 1.04 and Holm–Bonferroni-adjusted p < 0.001. Room to question AI outputs increased from 2.71 (0.68) to 3.67 (0.58), with a mean change of 0.96 (95% CI 0.69 to 1.23), Cohen’s d z = 1.29 and Holm–Bonferroni-adjusted p < 0.001. The remaining four domains showed small, non-significant changes after Holm–Bonferroni correction. T2 open-text responses indicated that students recognised the formative value of AI feedback while emphasising critical use, teacher supervision, preservation of interpersonal interaction and clearer data governance. This exploratory longitudinal empirical ethics survey showed that, following participation in an AI-supported doctor–patient communication learning pathway, two of the six ethics-related domains showed statistically significant changes: trust in AI feedback decreased, while perceived room to question AI outputs increased. These findings support continued attention to ethics-focused evaluation and clear governance arrangements in AI-supported learning pathways.