Confidence With Machines: What Shapes AI Self-Efficacy in Future Health Professionals

Artificial intelligence is quietly rewriting the job description of every future doctor, nurse, pharmacist, and public health specialist, and a new study from Iran offers one of the most detailed portraits yet of how confident these students actually feel about using it. Researchers at Tabriz University of Medical Sciences surveyed 550 medical sciences students who already reported some experience with AI tools, measuring what psychologists call self-efficacy: a person’s belief in their own ability to execute a task successfully. The results, published in BMC Psychology, reveal a striking pattern in which academic achievement, employment, and the purpose of screen time matter far more than age or gender when it comes to feeling capable with AI.

The study team, led by Neda Gilani of the Department of Statistics and Epidemiology, used the Persian version of the Artificial Intelligence Self-Efficacy Scale, a validated instrument that asks respondents to rate their confidence across distinct dimensions of AI competence. On a standardized scale from 0 to 100, the students averaged 69.80 with a standard deviation of 13.34, a figure suggesting generally solid but far from universal confidence among those who had already dipped a toe into AI use. Importantly, the researchers excluded students who reported using AI not at all, meaning the sample captures the correlates of confidence specifically among young people with at least some hands-on exposure to the technology.

The methodological design was a cross-sectional survey, recruited through convenience sampling via online questionnaires distributed across university social media channels and student networks. Participants answered questions covering demographics, academic characteristics, socioeconomic status, and patterns of digital media use, alongside the self-efficacy scale. The researchers then applied both univariable and multivariable linear regression, a statistical approach that allows each factor’s independent association with AI self-efficacy to be estimated while holding all the others constant. This matters because raw correlations can be misleading: a student’s field of study, income background, and daily habits all interlock, and only multivariable modeling can begin to untangle which threads actually carry weight.

One of the most eye-catching findings concerned educational level. Compared with medical residents, students at every other stage of training reported higher AI self-efficacy. Bachelor’s students scored on average 15.34 points higher, master’s and PhD students 12.63 points higher, and professional doctorate students 18.63 points higher, with confidence intervals indicating these differences were unlikely to be statistical noise. The pattern is intriguing because residents are clinically senior, yet they emerged as the least confident group in the analysis. The authors do not overinterpret this, but the finding hints that younger students immersed in coursework and digital learning environments may be accumulating AI familiarity faster than physicians in the demanding final stretch of specialty training.

Academic performance also left a clear statistical fingerprint. Each increment in grade point average was associated with a 3.62-point rise in AI self-efficacy, with a 95 percent confidence interval of 2.44 to 4.80. In plain terms, students who excel in their formal studies also tend to feel more capable with AI tools, a relationship consistent with the broader psychological literature linking general academic self-belief to domain-specific confidence. Employment status added another independent boost: working students scored 5.25 points higher than their non-employed peers, with a confidence interval of 1.03 to 9.49. The authors suggest that workplace exposure may provide practical contexts in which AI skills are exercised and reinforced, though the cross-sectional design cannot confirm which way the causal arrow points.

Perhaps the most socially resonant result involved digital media use, where the purpose of screen time proved decisive. Students who spent three or more hours per day on educational digital media scored significantly higher on AI self-efficacy than those using such media less than one hour daily. The reverse held for leisure: recreational digital media use of three or more hours per day was associated with lower AI self-efficacy scores. The contrast is a textbook illustration of how the same medium can produce opposite psychological outcomes depending on how it is deployed. Hours spent watching tutorials, reading technical documentation, or practicing with software appear to build genuine perceived competence, while equivalent hours of passive entertainment do not, and may even displace the activities that would.

Just as revealing were the factors that dropped out of the final model. Age and gender were not independently associated with AI self-efficacy once the other variables were accounted for, a finding that challenges assumptions about technology confidence being a young person’s or a man’s game. Within this population of students who already use AI, the drivers of confidence appear to be what students do and achieve rather than who they are demographically. For educators and policymakers, that is an encouraging message, because behaviors and academic supports are far more amenable to intervention than fixed characteristics.

Beneath the headline score, the subscale structure of the Artificial Intelligence Self-Efficacy Scale added texture to the picture. The Assistance subscale, which reflects confidence in using AI as a supportive tool, recorded the highest mean score among the students surveyed. The Anthropomorphic Interaction subscale, which concerns confidence in engaging with AI systems that simulate human-like interaction, registered the lowest. That gap is intuitively plausible: asking an AI to summarize a paper or draft a message is a bounded, tool-like task, while interacting with conversational agents that mimic human behavior raises different and perhaps less familiar demands. As large language models and virtual agents become embedded in clinical settings, this dimension of confidence may become increasingly consequential for patient communication and professional training alike.

The study’s context gives it particular weight. Iran’s medical education system, like health systems worldwide, is confronting the rapid arrival of AI-driven diagnostics, clinical decision support, and administrative automation, and the students surveyed at Tabriz University of Medical Sciences represent the workforce that will operate these systems. The research protocol was approved and supported by the university, and the study received ethics approval from its institutional review board under code IR.TBZMED.REC.1403.359, with procedures conducted in accordance with the Declaration of Helsinki. Participation was voluntary and anonymous, and email addresses collected for a prize draw were kept separate from questionnaire responses, safeguards that matter for the integrity of self-reported data.

The authors are careful about the limits of what their design can show. Cross-sectional associations, they emphasize, do not establish causality: it is equally possible that confident students choose to spend more time on educational media, or that unmeasured traits drive both academic success and AI confidence. Convenience sampling through online channels also means the sample may not represent all medical sciences students, and the exclusion of complete AI non-users narrows the scope to those with prior exposure. The team calls for longitudinal studies to clarify the temporal direction of these relationships. Even with those caveats, the study delivers a clear and actionable signal: in the race to prepare health professionals for an AI-saturated workplace, grades, jobs, and purposeful digital habits appear to matter more than birthdays or gender, and the students who use their screens to learn rather than to scroll are the ones most ready to trust their own hands with the machines.

Subject of Research: AI self-efficacy and its academic, behavioral, and socioeconomic correlates among medical sciences students

Article Title: Artificial intelligence self-efficacy among iranian medical sciences students: academic, behavioral, and socioeconomic correlates

Article References: Gilani, N., Pourabbas, A., Haghshenas, R., & Dehghani, G. (2026). Artificial intelligence self-efficacy among iranian medical sciences students: academic, behavioral, and socioeconomic correlates. BMC Psychology. https://doi.org/10.1186/s40359-026-05735-4

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05735-4

Keywords: artificial intelligence, self-efficacy, medical education, medical sciences students, digital media, cross-sectional study, linear regression, academic performance, socioeconomic factors, Iran, BMC Psychology, health professions