Our findings demonstrated that medical students overwhelmingly believe that AI is important to the future of medicine and desire learning opportunities about AI, yet there remains a lack of educational opportunities across Canada at the medical student level. With the rapid progression of AI tools towards clinical implementation and more prevalent use of AI in medical research, educational opportunities about AI should be considered for inclusion in formal medical curriculum. Further, as the skillsets required to use AI may be different than those traditionally possessed by physicians, the desired learning formats, content interests, and perceived learning barriers of medical learners must inform the inclusion of AI content in medical curriculum.
Our findings are consistent with previous studies of medical learners, which have also identified limited knowledge of AI among medical trainees (Bisdas et al., 2021; Gong et al., 2019; Pinto dos Santos et al., 2019; Teng et al., 2022), A recent survey by Teng et al. (2022) found that medical students had limited knowledge about AI, suggesting that this indicated a need for urgent education. They noted that this growing knowledge gap would likely become a barrier to the development and use of AI in medicine, something that has been supported by other literature (Celi et al., 2016; Yu et al., 2018). Interestingly, this survey found that healthcare learners were optimistic of AI in their fields, although they were not sure it would be relevant in their field (Teng et al., 2022). We also found a degree of cognitive dissonance among participants, as they believed that AI would revolutionize medicine while simultaneously believing AI would not directly affect them or their future practise. These findings could reflect poor understanding of AI applications, sensational reporting of AI in media or medical literature, or the limited exposure to AI in a clinical setting. Some recent survey cohorts of medical students have found that their cohort is worried that AI may replace physicians in the future, while other surveys have reported this to be a non-issue for their study cohort (Gong et al., 2019; Pinto dos Santos et al., 2019). Gong et al. (2019) found that this anxiety has discouraged students from considering imaging based diagnostic specialities, such as radiology. Our findings are more congruent with a European survey performed by Pinto dos Santos et al. (2021), and did not support anxiety related to physician replacement among medical students. While fears about AI may vary given study cohorts, anxiety regarding the use of AI in a clinical context could be ameliorated by curriculum. Finally, as AI applications progress towards clinical implementation, a lack of understanding could present challenges in effective uptake by physicians (Pucchio et al., 2022a; Teng et al., 2022). Teaching could address this, in addition to other changing requirements such as the ethical and humanistic role of physicians (Pucchio et al., 2021). Urgent development in medical curriculum is required to accommodate for this growing need (Bisdas et al., 2021; Pucchio et al., 2021; Teng et al., 2022). As both this study and previous surveys have confirmed that medical students want AI incorporated into their formal medical curriculum, any such changes should be well received by the undergraduate medical learner population (Bisdas et al., 2021; Teng et al., 2022).
Previous literature has identified potential objectives for AI teaching, suggesting educational objectives including identifying what technology is appropriate in a specific clinical context, the humanistic and ethical components of AI, and identification of quality improvement applications of AI (Barbosa Breda et al., 2020; Pucchio et al., 2021; Teng et al., 2022). This study adds an assessment of existing educational opportunities, preferred formats of AI education by medical students, and potential barriers to uptake. We found that there are no existing formal opportunities to learn about AI nationally. Educational opportunities are similarly limited outside of Canada (Paranjape et al., 2019). Our respondents identified workshops as their preferred learning format, followed by lectures, and collaborative activities with other departments such as computer science. Our interviews identified one notable barrier to the inclusion of AI in formal curriculum, being non-AI content taking priority curricular inclusion. This could be mitigated by the development of non-longitudinal educational formats such as workshops. Although survey respondents did not believe technical knowledge would be a barrier to uptake of AI, interview participants did express concern about a lack of mathematical or computer science knowledge prevent effective learning about AI. Given the large spectrum of educational backgrounds and experience with technology, it is prudent for medical AI curriculum to restrain from exploring complex technical detail.
The study limitations include non-response and participant bias. While we did receive responses from all medical schools in Canada, our respondent population makes up only ~ 4.5% of the total undergraduate medical student population in Canada (The Association of Faculties of Medicine of Canada, 2019). This was in part due to the variable ability of undergraduate medical faculties to support survey dissemination, with some sending the survey as a newsletter, other posting it on the student portal, and other being unable to facilitate distribution. It is likely that participant bias affected study outcomes, and that respondents were more likely to possess interest or knowledge of AI and have a stronger technical understanding than non-respondents. Finally, medical students in later stages of training (e.g. 3rd and 4th year) and male respondents were underrepresented in the survey results.
A lack of understanding of AI has been demonstrated over multiple studies, as has the need for curriculum to address AI in medicine. With this survey providing insight into preferred formats of AI education and barriers to AI education, informed development of AI curriculum is possible. Medical education has been traditionally slow to adapt to technological changes, leaving students ill prepared to use technology in clinical practise (Wartman & Combs, 2018). However, as policy both in Canada and internationally begins to acknowledge the importance of AI in medicine, financial and institutional support for educational efforts will grow (Kocabas et al., 2021). Future research should seek to develop educational content in the formats indicated above and trial them in a medical student population.