Survey of physicians, residents, and medical students on the use of artificial intelligence technologies in healthcare and medicine

Abstract

Objective: To assess the attitudes of physicians, residents, and medical students toward artificial intelligence (AI) technologies, the prevalence and patterns of AI use in clinical practice and healthcare in Russia; and to characterize expectations and barriers to its implementation. Material and methods. A cross-sectional online survey was conducted from March 1 to July 15, 2026. The primary array consisted of 1169 returned questionnaires. After applying the exclusion criteria (age <18, technical data entry anomalies), 8 questionnaires were excluded. All subsequent calculations were performed on the final analytical sample of 1161 questionnaires. Statistical analysis involved Spearman's rank correlation, Kruskal–Wallis criterion, cluster analysis (k-means), principal component analysis (PCA). Results. The sample included 809 (69.7%) women, and 352 (30.3%) men, with the average age of 37.5±14.6 years (median 33 years, range 18–83 years). Among the respondents, 706 (60.8%) regularly or periodically use AI in medical activities, with the most common tool being large language model-based chatbots used by 620 (53.4%) participants. Thirty (2.6%) respondents fully trust the conclusions of AI systems, while 587 (50.6%) demonstrate insignificant trust. The perceived accuracy of AI diagnosis was 47.6% on average, while the perceived accuracy of diagnosis by physicians was 72.4% (a gap of 24.8%, stable in all age groups). Specialists <29 years old are more likely to use AI; the highest readiness was recorded in the group of 30–39 years. The use of AI outside of professional medical practice was the strongest predictor of its clinical use (ρ=0.65; p<0.001). Evaluation of diagnostic accuracy by physicians was not related to any indicator of AI acceptance (p<0.05). Cluster analysis (k=4) revealed four types of digital disposition: “Enthusiasts” (n=262), “Pragmatists” (n=310), “Observers” (n=269), and “Skeptics’ (n=320). The PCA confirmed a two-factor structure: acceptance of AI (PC1 39.4%) and trust in accuracy (PC2 15.0%). Conclusion. The gap between the declared acceptance of technology and actual behavioral engagement remains a persistent characteristic of the professional medical environment. A comparative analysis of age groups refutes the simplistic notion that “younger people are more supportive of AI, while older people more resistant to it”. Improving the overall digital literacy of medical professionals is a more effective and adaptable approach to reducing barriers to AI adoption than specialized medical AI training programs. At the same time, age predicts behavior rather than attitude: educational interventions aimed at changing attitudes towards technology are applicable to all age groups equally.