Introduction
In the modern world, information technologies are advancing at a remarkable pace. Humanity has essentially entered a new era of development, one marked by transformative changes across numerous aspects of life. Artificial intelligence (AI) is steadily expanding its influence over how we access and process information. At the same time, its impact on decision-making continues to grow, extending into critical fields such as medicine and medical education — particularly at the postgraduate level. This reality calls for a careful examination of both the tangible benefits and potential risks associated with its use. It seems essential to analyze what deserves our focused attention regarding AI in postgraduate medical training.
This work seeks to examine and analyze various aspects of AI application in medical education, particularly at the postgraduate level, while presenting our perspective on certain advantages and associated risks.
Artificial intelligence is a considerably broad term that is presently employed with varying interpretations. It seems reasonable to embrace the perspective of J. Camblong [1], who conceptualizes AI as a comprehensive toolkit, with its distinct variants representing individual instruments. Currently, AI encompasses such tools as statistical inference, pattern recognition, machine learning, deep learning, and generative AI.
Following comprehensive searches of PubMed, Embase, Cochrane, and CNKI databases, there was assessed the literature pertaining to “AI + medicine/medical education” from 2017 through 2022, establishing that current AI deployment in medical education concentrates primarily on clinical specialty training and continuing professional development [2]. The predominant fields of utilization encompass radiology, diagnostics, surgery, cardiology, and dentistry, with the fundamental role being to augment clinician efficiency and accuracy. However, the methodological quality of the examined research on “AI + medicine/medical education” remained low.
The realm of postgraduate medical education encompasses an extensive body of knowledge requiring assimilation through ongoing learning, one that must logically align with foundational principles and their evolving interpretation. Thus, AI application in this field can be exceptionally bene-ficial for organizing and structuring data. It may likewise be employed for analytical purposes. However, this demands a certain degree of intellectual and creative interpretation of the information obtained. Particularly when considering the risks accompanying the analysis of substantial data repositories, one should recognize that various factors can significantly impact results, including presentation bias, errors or deliberate data falsification, and the commercialization of scientific literature that favors selective reporting within vast information pools.
A notable advantage of AI integration in postgraduate education lies in its capacity to facilitate swift access to information from foreign-language sources through machine translation. The question of whether AI can entirely supplant instructors within postgraduate education systems represents a pertinent concern warranting careful examination.
Materials and methods
This investigation examined multiple dimensions of AI application in postgraduate medical education. Drawing upon the scholarly literature, we analyzed the principal advantages of incorporating AI tools into educational processes, with particular emphasis on the enhanced capacity for accessing contemporary medical information. We further investigated and assessed the risks attendant upon AI tool implementation in postgraduate medical training, encompassing concerns regarding competence, content quality, alignment of learning outcomes with educational objectives, bias, and ethical and legal considerations.
Results
Throughout the continuum of postgraduate medical education, various configurations of the aforementioned AI tools (statistical inference, pattern recognition, machine learning, deep learning, and generative AI) may be strategically deployed to refine existing competencies and acquire new knowledge and skills, tailored to specific objectives at successive educational stages. Notably, leveraging such AI instruments as deep learning and generative AI to facilitate efficient data sorting and analysis can substantially elevate the quality of educational content. Naturally, this prospect generates considerable enthusiasm regarding their pedagogical potential.
Concurrently, questions arise regarding the reliability of information obtained through generative AI. Despite the prevailing impression that ChatGPT, for instance, constitutes a competent and trustworthy source of medical knowledge, it remains inadequately aligned with such expectations. Notably, evidence indicates that ChatGPT furnished correct responses to approximately 60 % of randomly sampled questions from the European Examination in General Cardiology, yielding incorrect or indeterminate answers for the remaining 40 % [3]. ChatGPT’s performance in advanced specialty training for neurology and neuroscience was similarly assessed using 69 questions from the Specialty Certificate Examination (SCE) web-based question repository. These questions encompassed subtopics including clinical presentation, diagnosis, interpretation, and therapeutic management, with certain items addressing specific patient populations. Evaluation and comparison of ChatGPT 3.5 Legacy, ChatGPT 3.5 Default, and ChatGPT-4 revealed that the former two models demonstrated overall accuracy rates of 42 and 57 % respectively, falling short of the 58 % threshold required for SCE 2022 qualification. Conversely, ChatGPT-4 attained superior accuracy (64 %), surpassing the passing threshold and outperforming its predecessors across multiple disciplines and subtopics, thereby demonstrating the potential of AI models within this domain [4].
ChatGPT-4’s diagnostic performance in neurological contexts was further examined with respect to lesion locali-zation and differential diagnosis. GPT-4 achieved accurate localization in fewer than 50 % of cases and demonstrated diminished performance upon supplementary testing. The model exhibited greater precision in localizing and diagnosing less complex cases relative to more challenging presentations [5].
AI and generative language models offer substantial opportunities for enhancing medical education through realistic simulation, including digital patient models, standard and atypical clinical scenarios, and synthetic medical imaging. Nevertheless, ensuring content quality remains challenging, particularly given concerns regarding accuracy, reliability, and potential misuse [6].
A crucial aspect to consider in integrating generative AI into medical education involves the attendant ethical and legal challenges, including bias, safety, transparency, and confidentiality, which necessitate the development of comprehensive guidelines and policies governing AI use in medical education [6, 7].
Analyzing and evaluating AI’s existing and potential role in simulation-based education warrant particular attention. Several studies have been conducted at the Neurosurgical Simulation and Artificial Intelligence Learning Centre of McGill University’s Montreal Neurological Institute (Canada), examining surgical performance metrics among medical students completing an AI-enhanced training curriculum compared to a control cohort receiving standard instruction without AI-based feedback. Notably, certain undesirable outcomes emerged regarding technical competencies within the AI-enhanced training group. Specifically, the AI-augmented bimanual surgical skills curriculum produced unintended effects, enhancing safety performance while simultaneously compromising other performance metrics [8].
AI systems in medical contexts confront multiple challenges, among which is the tendency to prioritize predictive accuracy over rigorous scientific validation and clinical assessment [9]. This inevitably affects the informational foundation of postgraduate medical training and constrains the development of sound clinical reasoning among learners.
Overall, foundation models such as ChatGPT depend critically on the vast data repositories uniquely afforded by the Internet, encompassing material of widely varying logical coherence, factual accuracy, and reliability, content whose complex semantic filtering cannot be reliably executed without human oversight. This necessitates extensive human supervision not only of training data inputs but also of model outputs, inevitably imbuing any model with subjectivity reflecting its developers’ perspectives [10]. Thus, while integrating AI tools, particularly chatbots, into educational processes offers substantial benefits, it remains essential to employ AI instruments as supplementary resources rather than primary authorities [11].
A critical factor in postgraduate medical education, particularly for accessing current scientific information in real time, is physicians’ foreign language proficiency [12]. The application of AI tools (continuously improving machine translation) has become increasingly vital in acquiring such information. These tools help overcome language barriers, enhance learning efficiency, and offer promising prospects for further advancement.
Discussion
Based on analyzing various aspects of AI implementation in educational processes, including postgraduate medical training, several factors emerge as either advantages or risks. A paramount question concerns whether information obtained through generative AI can be fully trusted and relied upon, and if not, to what extent it merits confidence and how it should be verified. Numerous examples now raise concerns regarding potential inadequacies in AI-generated data, including discrepancies between generated content and primary sources or foundational knowledge. These issues create substantial risks to the reliability of information employed in educational contexts. Given these considerations, AI-generated educational content requires ongoing scrutiny and rigorous quality assessment.
Implementing models such as GPT-4 into educational processes and clinical practice requires targeted training with specialized neurological focus, particularly given that GPT-4 achieved accurate neurological lesion localization in fewer than 50 % of cases and demonstrated declining performance under additional testing [5]. Overall, analyzing this evidence within the context of medical practice reveals that even minimal error rates from fully trusted AI tools constitute substantial medical risks, especially in clinical settings. Moreover, answering examination questions correctly, absent direct patient care experience, cannot establish genuine professional competence. Thus, while AI represents a valuable auxiliary instrument, it cannot supplant clinical reasoning or hands-on practice, nor does it diminish their fundamental importance.
Naturally, AI-generated products such as realistic simulation through digital patient models, standard and atypical clinical scenarios, and synthetic medical imaging contribute substantially to enhancing medical education. These applications can benefit not only learners but also educators in diversifying instructional materials, refining pedagogical approaches, and advancing professional development.
The previously discussed findings regarding AI-enhanced bimanual surgical skills training, which produced unintended effects improving safety metrics while compromising other performance indicators [8], merit careful consideration. These results underscore the critical need for continuous monitoring and assessment to ensure alignment between learning outcomes and educational objectives when integrating AI into practical training curricula.
Optimizing AI implementation in medical education requires ongoing collaboration among educators, clinical experts, and AI developers. When integrating generative AI into medical education, accompanying ethical and legal challenges must be addressed, including bias, safety, transparency, and confidentiality. Resolving these issues demands sustained effort and coordinated engagement from all stakeholders.
Several additional critical aspects warrant consideration. The potential dehumanization of educational processes necessitates careful deliberation when integrating AI into learning systems, particularly within medical education. Furthermore, from a neurophysiological perspective, optimal cognitive development and function require diverse and active mental processes. Substituting these processes with AI tools during learning poses potential risks to cultivating individual intellectual capacities. Therefore, achieving an appropriate balance between human cognition and AI assistance remains essential, including throughout postgraduate training.
Given the foregoing considerations, AI cannot currently replace postgraduate medical educators, nor is complete substitution of human instructors with AI tools likely to serve humanistic values in the future.
Conclusions
Employing AI tools in postgraduate medical education represents a promising avenue for enhancing technical capabilities for both learners and educators, yet their application entails considerable risks requiring careful management. Optimizing AI integration in medical education necessitates ongoing collaboration among educators, clinical experts, and AI developers.
Received 04.05.2026
Revised 08.06.2026
Accepted 17.06.2026
Список литературы
1. Camblong J. Where healthcare needs to focus for AI. Fast Company. 2023 Oct 04. Available from: https://www.fastcompany.com/90958948/where-healthcare-needs-to-focus-for-ai.
2. Sun L, Yin C, Xu Q, Zhao W. Artificial intelligence for healthcare and medical education: a systematic review. Am J Transl Res. 2023 Jul 15;15(7):4820-8.
3. Haverkamp W, Tennenbaum J, Strodthoff N. ChatGPT fails the test of evidence-based medicine. Eur Heart J Digit Health. 2023 Jul;4(5):366-7. doi: 10.1093/ehjdh/ztad043.
4. Giannos P. Evaluating the limits of AI in medical specialisation: ChatGPT’s performance on the UK Neurology Specialty Certificate Examination. BMJ Neurol Open. 2023 Jun;5(1):e000451. doi: 10.1136/bmjno-2023-000451.
5. Galetta K, Meltzer E. Does GPT-4 have neurophobia? Loca-lization and diagnostic accuracy of an artificial intelligence-powered chatbot in clinical vignettes. J Neurol Sci. 2023 Oct 15;453:120804. doi: 10.1016/j.jns.2023.120804.
6. Karabacak M, Ozkara BB, Margetis K, Wintermark M, Bisdas S. The Advent of Generative Language Models in Medical Education. JMIR Med Educ. 2023 Jun 6;9:e48163. doi: 10.2196/48163.
7. Katznelson G, Gerke S. The need for health AI ethics in medical school education. Adv Health Sci Educ Theory Pract. 2021 Oct;26(4):1447-58. doi: 10.1007/s10459-021-10040-3.
8. Fazlollahi AM, Yilmaz R, Winkler-Schwartz A, Mirchi N, Ledwos N, et al. AI in Surgical Curriculum Design and Unintended Outcomes for Technical Competencies in Simulation Training. JAMA Netw Open. 2023 Sep 5;6(9):e2334658. doi: 10.1001/jamanetworkopen.2023.34658.
9. Khanna NN, Maindarkar MA, Viswanathan V, Paul S, Bhagawati M, Ahluwalia P, et al. Economics of Artificial Intelligence in Healthcare: Diagnosis vs. Treatment. Healthcare (Basel). 2022 Dec 9;10(12):2493. doi: 10.3390/healthcare10122493.
10. Ruffle JK, Foulon C, Nachev P. The human cost of ethical artificial intelligence. Brain Struct Funct. 2023 Jul;228(6):1365-9. doi: 10.1007/s00429-023-02662-7.
11. Ghorashi N, Ismail A, Ghosh P, Sidawy A, Javan R. AI-Powered Chatbots in Medical Education: Potential Applications and Implications. Cureus. 2023 Aug 10;15(8):e43271. doi: 10.7759/cureus.43271.
12. Tkachenko OV, Lichman LYu, Trishchynska MA, Kononets OM, Stetsyuk RA, Khanenko NV. Advancing foreign language competence in neurologists: challenges, trends, and strategies for improvement. Int Neurol J. 2025;21(3):13-8. doi: 10.22141/2224-0713.21.3.2025.1171.
