Sample description and interest in and exposure to AI in health care
The sample consisted of 354 medical students (63.8% male) whose mean age was 21.4 years (SD = 1.8). The sample characteristics are presented in Table 1. Most students (77.1%) had never taken a course in AI, but a similar proportion of students reported at least some level of interest in the field (64.7%). GPA was not associated with either AI exposure or interest. Students in different cohorts did not differ significantly in the number of AI courses they had taken or their level of interest in AI. As expected, students who reported having received formal AI training in their education also reported having taken more AI courses (t (350) = 11.18, p < 0.001) and a higher level of interest in AI (t (350) = 2.29, p = 0.022). Students in their clinical years exhibited significantly lower GPA values compared to those in their basic years (t (352) = 2.76, p = 0.006; Table 1).
When the participants were asked where they had been exposed to AI, the majority (78.2%) named public media (e.g., television, YouTube, and Twitter). Fewer than half reported exposure through family and friends (41.2%), followed by online sources (24.3%). Some students reported exposure through research projects (12.4%), peer-reviewed articles (10.2%), books (9.9%), lectures (9.0%), or conferences (4.5%).
Knowledge and perceptions of AI in medicine
The students’ responses to the questions concerning their understanding of AI are presented in Fig. 1. Students in their basic years of education were significantly more likely to report being able to list the advantages and benefits of AI in medicine (57.7%) than those in their clinical years (45.3%, χ2 = 4.69, p = 0.030). Beyond these questions, no significant differences based on the year of education were observed.
Overall responses to the questions regarding AI
Most students felt that AI would play a significant role in medicine during their lifetime (65.8%) and were excited about the possibility of using AI technology as future physicians (59.0%). However, few understood fundamental AI concepts (e.g., cross-validation, 18.4%), could list examples of clinically relevant AI research (25.1%), or felt that their school offered resources to explore AI in medicine (20.1%).
Influence of information sources on knowledge
The participants reported that they had learned about AI concepts primarily from online forums (n = 86), books (n = 35), lectures (n = 32), media platforms such as Twitter or YouTube (n = 277), family and friends (n = 146), peer-reviewed journal articles (n = 36), professors and doctors (n = 49), and research projects (n = 44). Because of the small sample size, we did not consider students who reported that they had learned from conferences (n = 16).
We used linear regression to test the association between the number of AI courses and information sources. Participants who learned from online forums (t(345) = 2.05, p = 0.041), books (t(345) = 3.12, p = 0.002), or lectures (t(345) = 2.44, p = 0.015) had taken significantly more AI courses than students who reported learning from friends and family (t(345) = − 3.61, p < 0.001) or media platforms (t(345) = − 3.78, p < 0.001).
We tested the participants’ levels of agreement (“agree” or “strongly agree”) with each survey question via logistic regression. Each answer was tested for associations with self-reported use of all information sources. Students who relied on media were more likely to indicate that AI would play a significant role in medicine in their lifetime (OR = 2.76, Z = 3.74, p < 0.001), they could list the advantages and benefits of the use of AI in medicine (OR = 1.95, Z = 2.43, p = 0.015), training in these concepts would be useful in their careers (OR = 2.27, Z = 3.01, p = 0.003), and they wanted to learn what medical students should know regarding AI in medicine (OR = 1.89, Z = 2.38, p = 0.017).
Participants who received information from family and friends were significantly more likely than others to be excited about using AI as future physicians (OR = 1.81, Z = 2.56, p = 0.010), as were students who learned from research projects (OR = 2.47, Z = 2.16, p = 0.031). Participants who relied on books were more likely than others to report that they understood AI concepts, such as convolutional neural networks and cross-validation (OR = 5.94, Z = 4.13, p < 0.001), be able to separate “hype” from clinically relevant AI articles (OR = 2.38, Z = 2.13, p = 0.033), and report that their school offered resources to explore AI in medicine (OR = 2.35, Z = 2.06, p = 0.040). Students who learned from professors or doctors reported that their school provided resources to support exploration at similar levels (OR = 2.29, Z = 2.31, p = 0.021).
Students who reported higher levels of interest in AI tended to respond more positively to survey items focused on AI knowledge and perceptions than those with lower levels of interest. Students who exhibited higher levels of interest were more excited to use AI technology as future physicians (Q2, r = 0.154), reported that they understood AI concepts (Q3, r = 0.153), indicated that they could list recent examples of clinically relevant research (Q4, r = 0.157) and the advantages of AI in medicine (Q5, r = 0.106), believed that training in AI concepts would be helpful (Q10, r = 0.110), and wanted to learn more about AI in medicine (Q13, r = 0.145). Notably, although all these correlations were significant, they were generally weak (range: 0.106 to 0.157).
Students who had taken more AI courses were significantly less likely than those who had taken fewer courses to believe that AI would play a significant role in medicine during their lifetime (Q1, r = -0.193), be excited about using AI technology as future physicians (Q2, r = -0.155), view training in AI concepts as useful for their future careers (Q10, r = -0.162), and want to learn more about AI in medicine (Q13, r = -0.142). However, these students reported that they could list recent examples of clinically relevant AI research (Q4, r = 0.109). Again, all the correlations were generally weak (range: 0.109 to 0.193).
What students want from AI education
On average, students reported that they were willing to spend 1.89 h per month learning about these topics (SD = 1.43). Students who reported higher levels of interest in AI were significantly more willing than those with lower interest to spend more time learning about these topics, after adjusting for age and sex (t (349) = 3.40, p < 0.001). However, the reverse was true of students who had taken more AI courses, who were less willing to spend time learning (t (349) = -2.36, p = 0.019); this may be because they had already learned about this topic. When participants were asked about the most useful ways in which their school could promote AI exposure among students, approximately half mentioned short lectures on the fundamentals of AI in medicine (51.9%), question-and-answer panels with leaders in the field (46.3%), and workshops on programming AI models (41.2%).
When participants were asked about the subjects that would be most interesting to explore, the majority mentioned the topics of when to use AI in medicine (62.9%) and its strengths and weaknesses (51.7%). Other topics of interest included AI ethics (43.8%) and the use of AI in medical research (40.1%). Relatively few expressed interest in concrete elements of AI implementation and use, such as the roles played by individuals on multidisciplinary teams researching AI (10.4%), ways of critiquing AI models (11.0%), and the process of developing AI models (18.9%).
Perceived influence of AI on medical specialties
Only a minority of students (22.3%) reported that they were less likely to work in specialties in which AI was perceived to have the most influence. The majority indicated that their specialty choice would not be affected by this consideration. When participants were asked about their intended field of practice, students in their basic years were most likely to indicate general surgery (30.6%), dermatology (29.7%), neurosurgery (21.6%), internal medicine (20.7%), and emergency medicine (20.7%). Students in their clinical years were most likely to indicate internal medicine (37.0%), followed by family medicine (25.9%), general surgery (24.7%), emergency medicine (19.3%), and ophthalmology (18.9%).
When participants were asked about the fields that they believed would be affected most strongly by AI, students in their basic years of education named diagnostic radiology (34.2%), general surgery (30.6%), pathology (24.3%), anesthesiology (23.4%), and family medicine (19.8%). Students in their clinical years shared similar opinions but at considerably higher rates, citing a higher likelihood of impacts on radiology (52.7%), pathology (34.6%), and family medicine (28.4%). Students in their clinical years were less likely to expect impacts on general surgery (20.6%) or anesthesiology (20.6%). Both groups reported that the impacts of AI were least likely to be observed in otolaryngology (0.9% basic, 1.2% clinical), urology (3.6% basic, 2.1% clinical), and obstetrics and gynecology (0.9% basic, 3.3% clinical).
