Inside the AI Classroom: How Medical Students and Professors Really Use Chatbots

Generative artificial intelligence has swept into lecture halls, libraries, and hospital corridors faster than almost any technology in the history of medical education, and a new qualitative study from Iran offers one of the most detailed portraits yet of how the people at the center of this transformation actually live with it. Published in BMC Medical Education, the research by Leila Homayouni and Zahra Karimian of Shiraz University of Medical Sciences set out to capture something that large surveys cannot: the texture, ambivalence, and quiet anxieties of faculty members and students who have already woven AI applications into their daily academic routines. Their findings paint a picture that is neither utopian nor dystopian, but conditional, negotiated, and deeply shaped by context.

The study, conducted between November 2024 and April 2025, recruited sixteen participants, eight faculty members and eight students, drawn from diverse medical disciplines at Shiraz University of Medical Sciences. All had prior experience using AI tools, and the researchers employed purposive and snowball sampling to reach participants with genuine, hands-on familiarity rather than secondhand impressions. Data came from semi-structured, in-depth interviews lasting fifty to sixty minutes each, supplemented by field notes, and the interviews continued until the team reached data saturation, the point at which new conversations stopped yielding new themes. Every transcript was analyzed verbatim using Graneheim and Lundman’s qualitative content analysis approach, supported by the MAXQDA 24 software package, and the team followed COREQ reporting guidelines to ensure methodological transparency and trustworthiness.

From this analysis, five overarching themes emerged, accompanied by sixteen subthemes: the functional and practical value of AI; human oversight, ethics and authenticity; perceived capacity building and institutional readiness; cognitive, emotional, and behavioral influence; and socio-cultural dynamics and future transformation. The architecture of these themes matters because it resists the simple binaries that dominate public discourse about AI in education. Participants did not describe the technology as either a miracle or a menace. Instead, they described a continuous negotiation, weighing efficiency gains against ethical risks, cognitive costs, and institutional constraints that vary from one classroom, one country, and one user to the next.

One of the study’s most striking findings is how differently the two groups used the same tools. Faculty members reported a strategic, almost managerial relationship with AI applications. They delegated routine, repetitive tasks, drafting administrative communications, summarizing material, handling the mechanical scaffolding of academic work, precisely so that their own cognitive effort could be preserved for complex judgment calls that demand expertise. Students, by contrast, exhibited what the researchers characterize as a more relational reliance on AI. Rather than treating the technology as a disposable assistant, students described a bond of habitual dependence, and that dependence appeared to heighten their vulnerability to automation bias, the well-documented tendency of human users to accept machine-generated output uncritically, even when it is wrong.

Automation bias is not a hypothetical concern in medicine. A medical student who accepts an AI-generated summary of a clinical guideline without checking the primary source, or who trusts a chatbot’s explanation of a drug interaction without verifying it, is practicing a habit that could later surface in patient care. The Shiraz study suggests that this risk is not evenly distributed: it concentrates among learners whose relationship with AI is intimate and habitual rather than instrumental and bounded. The researchers’ framing implies that the very population with the most to gain from AI’s speed and convenience, students still building their foundational knowledge, may also be the population most exposed to its cognitive costs.

Both faculty and students voiced a shared and deeply felt worry that the researchers label intellectual deskilling. The concern is that if AI systems routinely perform the cognitive heavy lifting of academic work, searching, synthesizing, drafting, and explaining, then the muscles those tasks once trained may atrophy. In medical education, where the goal is not merely to produce documents but to shape clinical reasoning, diagnostic intuition, and ethical judgment, deskilling is not an abstract worry. It strikes at the core purpose of the enterprise. The authors argue that mitigating this risk requires human-in-the-loop supervision, a design and pedagogical principle in which a human being remains the final arbiter of AI output, checking, correcting, and taking responsibility for what is ultimately used or taught.

The study also surfaces a dimension that is often missing from Western-centric AI-in-education literature: infrastructure and geopolitics. Participants described barriers to access that their counterparts in well-resourced institutions rarely confront, including the need for workarounds such as virtual private networks to reach AI services that are restricted or unavailable in their region. The researchers conceptualize this as a geopolitical algorithmic divide, a structural inequity in who can participate in the AI transformation of education and on what terms. This framing reframes the global conversation: the digital divide of the previous era concerned connectivity and hardware, while the new divide concerns access to the large language models and generative platforms that increasingly mediate academic work. Equitable integration of AI into medical education, the authors suggest, will require institutional support to bridge this gap, not merely individual initiative.

Underlying all of these findings is a methodological point worth appreciating. Most existing evidence on AI in medical education is quantitative and drawn from Western institutions, which limits the field’s ability to understand how socio-cultural and economic context shapes adoption. By choosing a qualitative descriptive design and grounding it in a constrained educational setting in Iran, Homayouni and Karimian demonstrate that the experience of AI is not universal. Attitudes toward authenticity, the ethics of outsourcing thinking, and the practical feasibility of using these tools are all filtered through local conditions. A student in Shiraz navigating service restrictions and VPN workarounds is having a categorically different experience from a student in a Silicon Valley-adjacent university with seamless, subsidized access, even if both are typing prompts into the same model.

The study’s conclusions point toward a pedagogical shift that many educators have begun to advocate: moving away from assessing the products of learning and toward cultivating evaluative judgment. If AI can generate an essay, a summary, or a study plan in seconds, then the scarce and valuable skill is no longer production but evaluation, the capacity to judge whether an AI output is accurate, appropriate, and safe. The authors argue that sustainable integration of AI into medical education depends on this shift, alongside institutional policies and support structures, rather than on ad hoc individual experimentation. Without such a framework, the gap between enthusiastic early adopters and cautious skeptics within a single institution can itself become a source of inequity and confusion.

Finally, the researchers are candid about the limits of a snapshot. Their study captures a single moment in a rapidly evolving landscape, and they recommend longitudinal research to track how perceptions, ethical concerns, and usage patterns evolve as AI becomes more deeply embedded in medical training. The question their work leaves open is one that every medical school now faces: how to harvest the genuine efficiency of these tools without hollowing out the very expertise they are meant to serve. The answer, this study suggests, will not be found in blanket bans or uncritical embrace, but in the deliberate, context-sensitive negotiation between human judgment and machine capability that the Shiraz participants are already, sometimes uneasily, performing every day.

Subject of Research: Experiences of medical sciences faculty and students using artificial intelligence applications in medical education

Article Title: Exploring the experiences of medical sciences faculty members and students in using artificial intelligence applications: a qualitative descriptive study

Article References: Homayouni, L., & Karimian, Z. (2026). Exploring the experiences of medical sciences faculty members and students in using artificial intelligence applications: a qualitative descriptive study. BMC Medical Education. https://doi.org/10.1186/s12909-026-10331-6

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10331-6

Keywords: artificial intelligence, medical education, generative AI, qualitative research, automation bias, intellectual deskilling, faculty, students, large language models, digital divide, human-in-the-loop, Shiraz University of Medical Sciences