Medical Students and Faculty Split Over Generative AI in Training

Generative artificial intelligence has swept through universities faster than almost any technology in living memory, but a new study from a large UK medical school suggests that the people at the heart of medical education are approaching it with a striking mixture of curiosity, caution and outright distrust. The research, published in Discover Education by Huiming Ding and Matt Homer of the University of Leeds, offers one of the first direct comparisons of how medical students and the faculty who teach them actually use, and feel about, tools such as ChatGPT in learning, teaching and assessment.

The study arrives against a backdrop of dramatic change. A UK-wide survey of 1,041 undergraduates in early 2025 found that 88 percent of students were using generative AI for assessment purposes, up from just 53 percent a year earlier. In healthcare, generative AI is already being explored for diagnosis support, disease risk prediction, patient communication and documentation automation. That raises an obvious question for medical educators: if tomorrow’s doctors will work in AI-enabled hospitals, shouldn’t medical schools be teaching them how to use these systems wisely today?

To find out what is really happening on the ground, the researchers ran an exploratory sequential mixed-methods pilot. They began with three focus groups, one each with preclinical students, clinical students and faculty members, and used reflexive thematic analysis of the transcripts to build a questionnaire. That survey was then distributed to all undergraduates on the five-year MBChB programme and to teaching staff at the school, yielding valid responses from 29 students and 32 faculty members between January and February 2025. The team anchored their analysis in established technology acceptance frameworks, the Technology Acceptance Model and the Unified Theory of Acceptance and Use of Technology, which hold that perceived usefulness and perceived ease of use, shaped by factors such as training and support, drive whether people actually adopt a new tool.

The headline finding is that generative AI is already woven into medical education, but thinly and unevenly. Across all respondents, average use frequency was just 1.08 on a scale from 0 (never) to 4 (daily), equivalent to rare use. Nevertheless, 58.6 percent of students and 40.7 percent of faculty reported using generative AI to some extent in a medical education context. Students most commonly deployed it to summarise specific topics, cited by 76.5 percent of student users, and to generate practice questions, exam stations or simulated medical cases, each reported by 41.2 percent. Faculty leaned on it mainly for improving their own writing, at 53.8 percent, and preparing teaching materials, at 38.5 percent. Notably, no faculty member reported using it to write feedback on student assignments or to mark them.

Confidence, however, lags far behind adoption. Respondents rated their overall self-confidence in AI-related knowledge and skills at just 1.38 on a 0 to 4 scale, sitting between slightly confident and moderately confident, with no statistically significant difference between students and staff. Among those who had never used the technology, the two groups diverged sharply in their reasons: 58.3 percent of non-using students said they simply did not trust generative AI, a figure significantly higher than the 15.8 percent of faculty who gave the same answer. Faculty non-users instead pointed to practical barriers, with 47.4 percent reporting no relevant training, 42.1 percent saying they had no time to explore the tools, and roughly a third of both groups unsure about the regulations governing acceptable use.

When it came to what the technology can do, respondents on average slightly agreed that generative AI has genuine capabilities in medical education, with time-saving the most strongly endorsed item at a mean of 3.92 on a five-point agreement scale. Students were significantly more positive than faculty about AI’s potential to make educational resources more financially accessible, echoing wider research suggesting that AI-generated simulations and virtual patients could lower cost barriers for schools and learners. But many participants insisted that effectiveness depends entirely on how the tools are applied. One faculty respondent argued that the question is less about the capabilities of generative AI and much more about the design of the processes it is involved in, noting that a badly designed learning activity remains badly designed whatever technologies are brought to bear on it. Another observed that educators and students have started using it without necessarily understanding it or having guidelines and boundaries for appropriate use.

The concerns were louder than the enthusiasm. On average, respondents agreed that generative AI carries substantial limitations, challenges and ethical risks, with a mean of 3.93, and inaccuracy, lack of transparency and bias topped the list. Faculty held significantly stronger concerns than students overall, and were particularly worried about inequitable student access to the technology and the opacity of training data and algorithms. Several respondents warned that AI could become a crutch that erodes creativity, discourages independent study and prevents students from developing problem-solving skills. One student captured the unease vividly, describing generative AI as having passed the point of being a tool and become instead a machine that carries out tasks for us rather than with us, with the potential to damage much of what makes human learning and even culture special. Others flagged the technology’s shallow grasp of scientific literature and its limited emotional intelligence, with one respondent noting that AI used exclusively for medical education would not effectively teach empathy and humane clinical practice, although some participants conceded that newer models show improving sensitivity on sensitive topics.

Assessment emerged as the most contested battleground. Many respondents said they could not yet judge whether AI’s impact on medical education was positive or negative, and both groups perceived a slightly negative effect on most aspects of the curriculum, with the exception of curriculum content, teaching methods and career preferences. Faculty worried that students could use AI to write assessments and go undetected, undermining the validity of qualifications in a profession where knowledge is a key requisite for being a doctor, and one student claimed that every assessment without intense invigilation or tab-blocking was already at risk. Suggestions for reform included making assessments more authentic and rethinking how critical thinking can be promoted and evaluated in ways that cannot be circumvented by AI, though one faculty member cautioned that academics feel unconfident and unequipped to design authentic assessments that use AI explicitly. Another respondent offered a provocatively optimistic spin, arguing that some of the problems AI poses for assessment are overdue spurs to innovate beyond long-standing, lazy approaches to examining students.

One of the study’s most striking results concerns gender. Male respondents used generative AI significantly more often than females, reported markedly higher confidence in their AI knowledge and skills, expressed fewer concerns about its limitations and ethical challenges, and held more positive attitudes towards its impact on medical education. The researchers note that such differences echo previous findings of a gender gap in AI adoption and argue that larger studies should investigate whether the technology’s growing embedment in medical education risks creating unequal access and unequal benefit across subgroups of future doctors.

What both groups agreed on, emphatically, was the need for support. Students most wanted open communication with educators about AI use, cited by 62.1 percent, followed by clearer institutional regulations and guidance, at 58.6 percent, and student training sessions, at 55.2 percent. Faculty demands were even more emphatic: 90.6 percent called for training for themselves, and majorities also wanted open dialogue with students, clearer policies, student training and advice from national advisory bodies, with faculty expressing significantly greater need for these resources than students did. Several faculty respondents went further, describing a cultural problem in which some colleagues deny that students are using AI at all and design assessments for a context that no longer exists, while stigma against students who use AI may push them to misuse it secretly rather than engage with it openly. The study’s authors conclude that ensuring ethical, effective and equitable use of generative AI in medical education will require strategic support from schools, universities and national regulators, and they call for national-scale research to track how this rapidly shifting landscape evolves. As one faculty respondent put it, generative AI is one of the biggest challenges to the way universities operate in a long time, and it will be the leadership and vision of institutions that determine whether its disruptive injection of change turns out to be for the good or destructive.

Subject of Research: Student and faculty experiences and perceptions of generative artificial intelligence in UK medical education

Article Title: Exploring student and faculty experiences and perceptions of generative artificial intelligence in medical education

Article References: Ding, H., & Homer, M. (2026). Exploring student and faculty experiences and perceptions of generative artificial intelligence in medical education. Discover Education, 5(1), Article 832. https://doi.org/10.1007/s44217-026-02106-4

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02106-4

Keywords: generative artificial intelligence, medical education, undergraduate medical students, faculty perceptions, technology acceptance model, UTAUT, assessment integrity, AI training, gender differences, ethical challenges, UK medical school, mixed methods