Background: Artificial intelligence (AI) is increasingly intersecting with clinical care and medical education, yet adoption and perceptions vary across training stages. We surveyed medical students and medicine/medicine subspecialty trainees and faculty within a large academic health system to characterize current AI use, barriers to adoption, perceived impact, and educational needs.
Methods: We conducted a cross-sectional Qualtrics survey of medical students and medicine/medicine subspecialty residents, fellows, and attendings affiliated with an academic health system. Recruitment occurred via departmental and student listserv emails and announcements at meetings. Quantitative data were summarized with frequencies and Likert distributions; comparisons by training level used chi-square or Kruskal–Wallis tests; and associations with years in practice used Spearman correlation.
Results: Of 113 respondents, 108 were eligible and included: 41 medical students, 32 residents/fellows, and 35 attendings. Overall, 49% described their institution’s approach to AI as “early adopter,” with similar proportions across training levels. 87.5% of students, 85.0% of residents/fellows, and 68.0% of attendings reported using AI for education. Students primarily used AI to explain course concepts, proofread assignments, and support literature review; clinicians used it most often to learn about diseases/treatments, generate board-style scenarios, and summarize complex readings. Half of clinicians reported current clinical AI use, concentrated in administration, patient care, and diagnostic support. Attendings more often used AI for administrative tasks (82.4% vs 29.4%, p=0.007), whereas residents/fellows more often used it to monitor side effects/complications (52.9% vs 12.5%, p=0.001). Perceived AI impact was largely positive for administration/record keeping, diagnosis, and interpretation, but mixed for malpractice risk management, with students notably more optimistic than others. Respondents strongly endorsed the need for clinicians to be educated about AI and expressed interest in learning more, while few felt they had received adequate formal AI training. Priority curriculum areas included clinical applications, tasks AI can replace, and AI in research. Among non-users, the most common barriers were concerns about accuracy/reliability, lack of awareness, and limited access to tools.
Conclusions: Across training stages, AI is already widely used for education and by approximately half of medicine/medicine subspecialty clinicians in practice. Adoption patterns diverge by seniority—attendings primarily leverage AI for administrative efficiency, while residents/fellows report more clinical monitoring and decision-support use. Although overall perceptions were positive, substantial uncertainty remains regarding reliability, malpractice implications, and boundaries of appropriate use. Respondents strongly endorsed a need for formal, clinically oriented AI training.These findings highlight a time-sensitive opportunity for medical schools and medicine/medicine subspecialty training programs. AI literacy is becoming foundational for competent clinical practice, yet formal preparation remains limited. Developing structured curricula that address confidence gaps, ethics, safety, and clinical integration will be critical to ensure responsible and effective use of AI in patient care.