Future doctors, digital dilemmas: what are medical students’ perceptions of the ethical challenges of using machine learning applications in healthcare?

Introduction Machine learning (ML) is increasingly being integrated into healthcare, with potential to improve diagnostic accuracy, and patient care. However, its use introduces ethical concerns surrounding bias, accountability, transparency and informed consent. Despite national guidance recommending medical students are taught ML ethics in their curriculum, most UK medical schools are yet to implement this, potentially leaving students unprepared for future practice. This study aimed to investigate medical students’ knowledge of ML ethics, their perceptions of associated challenges, and views on integrating ML ethics into undergraduate medical education. Methods Fourteen medical students from two UK universities participated in focus groups where they discussed five ethical scenarios related to using ML in healthcare. Participants were recruited via convenience and snowball sampling, through emails, posters and referrals from clinical staff and students. Four focus group discussions were held, each containing three or four participants. Reflexive thematic analysis was used to interpret the collected data. Results Participants described minimal to no prior teaching and limited understanding of ML, yet identified several ethical concerns including exacerbating inequalities, responsibility uncertainty, lack of knowledge and maintaining trust. Students expressed concern that biased datasets may worsen healthcare inequalities and acknowledged uncertainty regarding clinician accountability when using opaque ML systems. Limited understanding of ML was perceived to compromise informed consent and clinicians’ ability to evaluate recommendations. Participants supported introducing ML ethics through case-based and reflective learning approaches. Discussion Although participants had limited prior exposure to ML, they identified several ethical concerns. Many believed clinician judgement could compensate for biased outputs, potentially underestimating the challenges of recognising automation bias and critically evaluating ML recommendations. Uncertainty about accountability reflected wider tensions between clinician responsibility and the opaque nature of many ML systems. Participants also acknowledged that inadequate understanding of ML may compromise informed consent, clinician confidence, and patient trust. These findings support including ML ethics education within undergraduate medicine. Integrating scenario-based and collaborative learning into existing curricula may develop ethical awareness, critical appraisal skills, and preparedness for responsible ML use in clinical practice.