According to economists and futurists, traditional health care will become increasingly disintermediated by innovations in digital technology, including advances in artificial intelligence (AI)/machine learning (ML) [1-3]. These views are also held by many AI experts and health care informaticians, many of whom are physicians, who predict that ongoing developments in AI/ML will revolutionize the delivery of health care [4-7]. Moreover, digital innovations and AI/ML-enabled tools already play roles in health care by helping patients to monitor and manage their symptoms, supporting patient triage decisions via chatbots, informing clinical decisions, offering treatment recommendations via clinical decision support tools, and supporting health care resource management [8]. Despite these developments, in surveys, many medical professionals are skeptical about the impact and value of digital and AI/ML tools on their job, with surveyed physicians doubting the scope of technological innovations to replace clinicians in fundamental medical tasks [9-11]. Emerging surveys among students enrolled in a range of health care training programs, including medicine, dentistry, and clinical psychology, also revealed divergent opinions about the impact of AI/ML on their chosen profession, with participants reporting limited formal education on these topics [12-18].

We sought to explore the opinions of final-year medical students in Ireland on the impact of future technology on the job of general practitioners (GPs). We performed a brief scoping review of the literature using the terms “artificial intelligence,” “machine learning,” “education,” and “training” in the search engines of PubMed and Google Scholar, and explored the grey literature. Only a few surveys, which were conducted in Europe, the United States, and South Korea, explored the attitudes of medical or health care students about the encroachment of AI/ML in medicine, and most were single-site studies [12-18]. Our objective was to explore the opinions of final-year medical students across Ireland to obtain a better understanding of their forecasts about the capacity of future technology to fully replace or to partner with physicians in undertaking key components of the work of GPs. In addition, our aim was to explore both students’ longer-term predictions and comparatively shorter-term forecasts (25 years from now) about how technology might impact the work of GPs.

Participants in this convenience sample paper-based survey were final-year medical students at 4 of Ireland’s 7 medical schools (after survey administration, in August 2021, a new 8th medical school at the University of Ulster began enrolling students). Using the study team’s contacts, we sought to administer the survey in the country’s 4 geographical provinces. Between April 2019 and March 2020, the anonymous survey was distributed by lecturers after compulsory final-year classes at each institution to increase responses.

Institutional review boards at University College Cork (protocol #2018-188), National University of Ireland Galway (protocol #19-Dec-15), Queen’s University Belfast (protocol #19.28), and University College Dublin (protocol #LS-19-89) approved the study at their respective sites. Participation was voluntary, and all students who decided to participate provided written consent.

The survey (Multimedia Appendix 1) was divided into 5 parts (Sections A to E). Section A requested demographic information. In Section B, the study team replicated and also extended components of a survey instrument originally devised to investigate the views of UK GPs about the potential impact of technology on the primary care profession [9]. The survey by Blease et al [9] formulated a generic list of tasks common to primary care, including “analyze patient information to reach diagnoses,” “analyze patient information to predict the likely course of the patient’s illness,” “evaluate when to refer patients to other health professionals,” “formulate personalized treatment plans,” “provide empathic care to patients,” and “provide documentation (eg, update medical records) about patients,” and requested that respondents rate the likelihood of these tasks being replaced by future technology. An additional goal was to compare students’ responses with those in the original UK survey.

Replicating the original survey, the first set of 6 survey items in Section B opened with a brief statement: “Some people believe that machine learning/artificial intelligence will lead to significant changes in medical practice and that machines will one day replace the work of physicians; others deny that new technologies will ever have the capacity to replace this work.” We then asked respondents their opinion on the likelihood that, “future technology will be able to fully replace and not merely aid human doctors in performing each task as well as or better than the average GP.” Employing 4-level Likert items, we included the following response options: “extremely unlikely,” “unlikely,” “likely,” and “extremely likely.” Participants who responded that replacement was “likely” or “very likely” were asked a follow-up question about how soon in their estimation would technology have the capacity to perform the task as well as or better than the average GP, and were provided with a list of 5 response options: “0-4 years from now,” “5-10 years from now,” “11-25 years from now,” “26-50 years from now,” and “more than 50 years from now.” In all closed-ended questions in the survey, we avoided “don’t know,” “neutral,” or “no opinion” options on the grounds that participants often conflate these answers [19].

The study team also extended and developed the original survey instrument by asking students 2 additional questions in Section B. One question was “In 25 years, of the following options, in your opinion what is the likely impact of artificial intelligence/machine learning on the work of GPs?” Students were offered 1 of 4 response options: “no influence (GPs’ jobs will remain unchanged),” “minimal influence (GPs’ jobs will change slightly),” “moderate influence (GPs’ jobs will change substantially),” and “extreme influence (GPs’ jobs will become obsolete).” Participants who answered that there would be minimal, moderate, or extreme influence were then asked the following open comment box question: “Please briefly describe the way(s) in which you believe artificial intelligence/machine learning will change GPs’ jobs in the next 25 years.”

While Section B explored opinions about the potential capabilities of future technology to fully replace GPs on specific tasks, the aim of Section C was to explore students’ views about routine partnership between “man and machine,” that is, GPs and digital tools, in performing various tasks in primary care. Specifically, our aim was to explore students’ predictions about the roles of technology in triage decisions, clinical decision support, remote monitoring of symptoms, and patients’ access to their records. Using a 6-point Likert scale we asked students their level of agreement about the following 6 scenarios: “25 years from now…” (1) “…technology (eg, smartphone apps) will be used to decide when patients need to see a GP,” (2) “…GPs will routinely work in partnership with artificial intelligence/machine learning to diagnose patients,” (3) “…GPs will routinely work in partnership with artificial intelligence/machine learning to determine the likely course of a patient’s illness,” (4) “…GPs will routinely work in partnership with artificial intelligence/machine learning to devise patient treatment plans,” (5) “…remote monitoring of patients’ vital signs will be more common than in-person check-ups of vital signs with GPs,” and (6) “… patients will have greater access to their own medical records than they do today.”

Section D of the survey focused on students’ views about the potential benefits and harms of AI/ML in medicine, and Section E focused on students’ experiences and opinions about formal teaching of AI/ML in their medical degree program. The results of Section D will be published elsewhere, and the results of Section E have now been published [20].

The survey was devised in consultation with Irish, British, and American primary care physicians, and we piloted the survey with physicians in Ireland and the United Kingdom (n=6), and final-year medical students in the United Kingdom (n=5) to ensure face validity. The feedback process was conducted via one-on-one consultations involving think-aloud methods with primary care physicians and medical students.

In total, 60.7% (153/252) of students left comments describing the ways in which they believed AI/ML will change the work of GPs 25 years from now. Comments were short and had a mean of 7.21 (SD 6.96) words. Following inductive analysis (see above), 4 major themes emerged (see Textbox 1). Illustrative examples of themes and categories are provided below. For more elaborate comments, participant number, gender, year of birth, and chosen medical specialty have been mentioned (the latter information was provided by the respondents).

Few studies have explored the views of medical students about how AI/ML will impact the future of their job. This mixed methods study specifically explored forecasts of final-year Irish medical students about how future technology might influence the work of GPs. When requested to forecast the impact of AI/ML on the work of GPs 25 years from now, students were divided, with around half of all surveyed students believing the work of GPs will change minimally or not at all. Notably, students who did not intend to enter primary care predicted that AI/ML would have greater impact.

With regard to specific tasks, around one-third of students moderately or strongly agreed that 25 years from now, technology (eg, smartphone apps) would be used to decide when patients need to see a GP. Similarly, around one-third moderately or strongly agreed that GPs would routinely work in partnership with AI/ML to diagnose patients, determine the likely course of a patient’s illness (“prognosis”), or devise patient treatment plans. About 4 in 10 students moderately or strongly agreed that 25 years from now, remote monitoring of patients’ vital signs would be more common than in-person check-ups for vital signs, with 7 in 10 students agreeing that patients would have greater access to their medical records. Again, students who did not intend to enter primary care were more likely to forecast that AI/ML would impact key aspects of the work of GPs, including formation of decisions about when patients should see GPs, assisting GPs in diagnoses and prognoses, and helping patients obtain greater access to their medical records.

Results from the qualitative section of the survey supported and partially elaborated on these predictions. The dominant perspective was that 25 years from now, there would be a reduction in GPs’ workloads with less paperwork and greater efficiency in primary care. Other common themes encompassed forecasts that AI/ML-enabled tools would aid clinical judgment but only for a narrow range of symptoms, mostly pertaining to imagery. Another theme was the potential for AI/ML to aid with treatment and care management, including automatic prescribing. Fewer students envisaged a role for AI/ML in patient self-monitoring, and only a minority predicted an increase in telemedicine or patient access to health care. Although participants were divided about whether AI/ML might have an impact on the time GPs would spend with patients, most were skeptical about whether technological tools could ever replace the empathy provided by GPs.

Offering forecasts on the capacity for future technology to fully replace core aspects of the job, around 2 in 3 students believed it is unlikely or very unlikely that GPs would ever be fully replaced by AI/ML tools in performing diagnoses or formulating personalized treatment plans for patients. Students were split over whether prognoses or triage could ever be fully replaced. Consistent with the qualitative component, however, students were most skeptical about the scope of future technology to replace GPs in providing empathic care, with more than 9 in 10 predicting this was unlikely or very unlikely. In contrast and in keeping with predictions about the impact of technology over the next 25 years, students were most expectant about the scope of future technology to fully replace GPs in undertaking the role of documentation, with more than 8 in 10 believing this was likely or very likely. Among them, around 2 in 3 predicted this would happen in the next 10 years. Finally, we also found correlations between gender and students’ opinions, with male respondents more likely to believe future technology would fully replace GPs in undertaking diagnostics and prognostics, and in the provision of empathy. Students who did not intend to enter primary care professions were more likely to believe GPs would be replaced by future technology in making referral decisions to other specialists.

The results of this study mirror other recent medical student and GP surveys, which demonstrated a wide range of opinions among participants about the impact of AI/ML on health professions [14,17,18]. Conspicuously, students’ opinions about the prospects for technology to fully replace various primary care tasks revealed some similarities but also intriguing differences with the findings in a recent survey conducted with GPs in the United Kingdom [9]. Final-year medical students in Ireland and experienced GPs offered similar predictions about the capacity for future technology to replace GPs in key tasks; however, students tended to be more cautious and conservative in their estimations of time scales for when AI/ML advances might arise. Although these divergences might be associated with sampling effects, we noted that the original UK GP study [9] reported a weak correlation between respondent age and opinions, with younger GPs more skeptical about the imminence of technological advances.

The reasons behind associations between younger age/inexperience and relative technoskepticism are not fully understood, though 2 hypotheses might be considered. First, it is reasonable to hypothesize that, compared with established GPs, younger respondents may be more AI/ML savvy and less susceptible to hype about AI/ML, and as a result, they may be more reserved in their forecasts. However, a growing number of student surveys now indicate that there is scarce formal training in AI/ML in medical schools [12-18,20]. Indeed, in previously published findings that emerged from the present survey, 4 in 10 final-year students reported that they had not heard of the term “machine learning,” with 2 in 3 reporting spending no time learning about AI/ML during the entire period of their medical degree [20]. Therefore, it is unlikely that greater awareness about technology influenced comparatively more conservative predictions among our student respondents. A second and more plausible hypothesis is that younger age/inexperience and technoskepticism might be associated with well-documented optimism bias, which is the tendency of people to believe that they will not be affected by negative events. In short, student participants may be susceptible to interpreting information on AI/ML in ways that support the prospects of their own long-term career in medicine. Tentative support for this hypothesis comes from differences in opinions related to students’ choices of medical specialty, with those intending to enter primary care less likely to believe AI/ML would impact the work of GPs. Further support comes from the finding that students predicted that the administrative burdens of updating documentation would be outsourced to technology.

Like other studies, including those among psychiatrists [10], male respondents were more likely to predict that future technology will be able to fully replace GPs in some key tasks. The reasons for this difference are not fully understood, though findings from social psychology demonstrate sex differences when it comes to risk aversion [22]. It may be that males are slightly less cautious on average compared with females in offering professionally threatening predictions. It is worth emphasizing, however, that other surveys have not reported gender differences [9]. Only one-third of our respondents were male. For many years in Ireland, there has been a trend of more male medical students than female medical students [23]. Therefore, it is possible that gender disparities in respondents’ opinions in the present study might have been an artifact of sampling limitations.

We also observed contrastive predictions among our students compared with informaticians and other experts working in health care AI/ML and related fields. A Delphi poll of international health informatics experts reported consensus that in 10 years (by 2029), advances in AI/ML would prompt workforce changes within primary care, with a shift toward computing and engineering in the educational backgrounds of students entering medical school, and increasing demands on students to work with AI/ML-enabled tools in health care [24]. In contrast, when asked to reflect on what might change 25 years from now, a minority of students forecasted that GPs would partner with AI/ML tools in supporting clinical decision-making. However, such advances are already underway. In countries with electronic health records (EHRs) in primary care, the availability and uptake of clinical decision support systems, which are tools that link patients’ personal information held in EHRs to clinical software to inform patient-specific assessments or recommendations, appear to be widespread [25]. These tools are being increasingly powered by ML, and they use computerized reminders, alerts, and prompts linked to patients’ electronic records to help inform recommendations. Prescription alerts, for example, warn doctors about harmful dosing or risks of drug interactions, and clinical decision support systems have the potential to help standardize guideline adherence, and support diagnostic and prognostic decisions [26].

Other predictions associated with access to primary care and patient management of their care also diverged from expert predictions and current trends. For example, when asked to predict what might change 25 years from now, a minority of students agreed that technology, such as smartphone apps, would be routinely used to decide when patients need to see a GP, a finding supported by qualitative analysis. Although partly accelerated by COVID-19 and stay-at-home measures, so-called “digital first” gateways to online triage, such as AskMyGP, Engage Consult, and eConsult in the United Kingdom, are being increasingly adopted in primary care [27]. Although these systems are implemented with the goal of mitigating increased work burdens, it is important to note that there is scarce evidence such systems, as currently embedded into work practices, do in fact improve efficiencies, and they may even exacerbate pressures on physicians by identifying greater patient needs [28,29]. It is worth emphasizing, however, that predictions about increasing implementation of AI/ML tools in medicine are not the same as gauging views about their adequacy, safety, or ease of use, especially with respect to integration into GPs’ workflows. Notwithstanding, students’ predictions did appear to contrast with growing prepandemic secular trends.

A larger proportion (107/244, 43.9%) of students, though still a minority, moderately or strongly agreed that remote monitoring of vital signs will be more common than in-person check-ups in the near future. Nonetheless, few students elaborated on this in the qualitative section of the survey. Although students could not have predicted how the pandemic would catalyze an uptick in telemedicine, including the use of electronic communication to track, monitor, or manage symptoms or chronic conditions [30], interest and uptake in remote patient monitoring has grown in recent years [31,32]. Increasingly via smartphone photos, blood pressure cuffs, heart rate monitors, portable electrocardiography systems, and a host of other devices, patients can manage their health from their home with real-time readings relayed instantly to the patients and the clinical team. Moreover, there is evidence that so-called mobile health may improve precision [33-36] while driving down health care expenditure, including hidden travel costs, related to in-person appointments [37-39].

Finally, 1 prediction was fully in line with recent health care developments. Almost all students expected that access to medical records would increase in the next 25 years. Currently, in around 20 countries, including Australia, Canada, the Nordic countries, and the United States, patients are offered rapid online access to at least some of their EHRs, a practice that is growing.

A major strength was soliciting the opinions of a diverse range of medical students from institutions in geographically distinctive regions of Ireland. The survey offered insights into students’ forecasts about the potential impact of technology on the work of GPs both in the medium term during their own career span and in the longer term with regard to replacement of doctors. Going further than other investigations [9], the present study examined students’ views about the likelihood of full technological replacement of GPs in specific core roles while also examining participants’ predictions about the extent to which GPs might partner with machines in a variety of tasks. Combined with the mixed methods approach, the study permitted more nuanced students’ opinions about the impact of AI/ML on the work of GPs.

The survey has several limitations. We used a nonprobability convenience sample, limiting generalizations about the opinions of all final-year medical students in Ireland. In addition, the moderate response rate (43%) raises questions about representativeness, though this is a very strong response rate for survey research where participants do not receive compensation. It is also unknown whether the decision to complete the survey was influenced by prior awareness about AI/ML or whether response biases were influenced by participants who were more enthusiastic or more skeptical about the effects of AI/ML on primary care. Because of the limitations of open comment boxes, participants’ responses were often vague or truncated, and it was not possible to probe the views of students in depth. The survey was administered prior to the COVID-19 pandemic, which has been associated with considerable developments and attention regarding the role of AI/ML-enabled tools in digital epidemiology and public health. Conceivably, if the survey had been undertaken after the pandemic, participants’ views might have differed. Nonetheless, to date, no medical school included in this survey has modified their curriculum to include greater education about AI/ML.

Further survey research and curricular analyses could explore the extent to which medical students receive training about existing clinical decision support tools and their implementation in clinical work. Qualitative research methods, such as interviews and focus groups, could provide more nuanced findings on aspects of students’ views and understanding about the impact of AI/ML on medicine. Finally, future studies could explore the views of medical faculty about the impact of AI/ML-enabled tools on medicine, including their awareness, understanding, and appreciation of the scope for these applications and limitations associated with these applications. Such work might help illuminate potential obstacles to curricular advancement on these topics within medical education.

This mixed methods survey provides insights into what final-year medical students in Ireland think about the impact of AI/ML on primary care. A broad spread of opinions was apparent, with many forecasts contrasting with the considered opinions of health informaticists. Ireland is ranked as a leading technology capital in Europe [40], with the fastest growing technology workforce on the continent [41]. This survey combined with previously published findings [20] suggests that training regarding AI/ML in Irish medical education may be lagging behind advances in the field. We caution that without a firm curricular foundation on advances in AI/ML, students may rely on extreme perspectives involving self-preserving optimism biases that demote the impact of advances in technology on their choice of specialty on the one hand and technohype on the other. Ultimately, these biases may lead to negative consequences in health care. Improvements in medical education could help prepare tomorrow’s doctors to optimize and lead the ethical and evidence-based implementation of AI/ML-enabled tools in medicine for enhancing the care of tomorrow’s patients.