Exploring medical students’ attitudes and perceptions toward artificial intelligence in medicine in Shandong Province, China Authors: Mingchan Liu, Yi Cheng, Shu Li, Shanshan Wang, Feng Du, Xiaonan Wei, Zhiying Ai, Siyuan Yan Published: 2025 ABSTRACT BackgroundThe integration of artificial intelligence (AI) into medical education has transformative potential, yet understanding medical students’ attitudes toward AI remains critical for its effective implementation. This study investigates the attitudes, perceptions, and the factors influencing them among medical students in Shandong, China, toward AI in education.MethodsA cross-sectional survey was conducted from May to June 2025 involving medical students at five medical universities in Shandong, China. Employing convenience sampling, 788 validated participants completed questionnaires that assessed variables including AI familiarity, perceived usefulness (PU), perceived ease of use (PEU), and ethical concerns. Statistical analyses comprised descriptive statistics, independent t-tests and one-way ANOVA tests.ResultsBased on 788 valid responses, the study revealed high levels of both familiarity with and usage of AI tools among medical students (47.33% and 91.24%). While they hold positive perceptions of AI’s PU (3.60 ± 0.85) and PEU (3.66 ± 0.76), significant ethical concerns exist, including privacy issues (48.48%), fears of eroding critical thinking (61.93%), and academic integrity worries (55.84%). Male students (p = 0.020) and those in higher academic years (p < 0.001) demonstrated stronger AI competency, with ethical apprehensions increasing notably as students’ progress through their medical education (p < 0.001). Institutional affiliation had little impact on these patterns (p > 0.05).ConclusionsShandong medical students demonstrate cautious optimism regarding AI adoption, recognizing its educational potential while emphasizing the necessity of ethical frameworks and operational safeguards. Curriculum adaptations and transparent governance mechanisms should be implemented to ensure congruence between technological implementation and educational objectives.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12909-025-08465-0. Introduction Artificial Intelligence (AI) has emerged as a transformative force across industries, particularly in healthcare, where its applications encompass diagnostics, treatment planning, and administrative workflows. AI technologies, including machine learning and deep learning, enable computers to analyze vast datasets, identify patterns, and generate insights that mimic human intelligence, yet with enhanced speed and scalability [1, 2]. For instance, in radiology, AI assists in interpreting complex imaging such as chest X-rays and CT scans, improving the detection of conditions like pulmonary tuberculosis and intracranial hemorrhage [3, 4]. The global availability of over 150 Food and Drug Administration (FDA)- or Conformité Européene (CE)-approved AI radiology products highlights its transformative potential in healthcare [5].​ In diagnostics, AI enhances precision in detecting pathologies and enables personalized treatment planning, from oral cancers and periodontal disease to orthodontics and neurosurgery [6, 7]. Through the analysis of genetic and clinical data, recent AI advancements are enabling more precise drug discovery and personalized treatment planning, notably within oncology and pharmacology [1, 8, 9]. Additionally, AI is critical in mitigating human error by integrating automated decision-support systems, which significantly reduce misdiagnosis rates and enhance the safety and reliability of healthcare service delivery [2, 10].​ Administratively, AI optimizes workflows such as patient scheduling and report generation, improving operational efficiency [11–13], while AI-integrated telemedicine facilitates remote monitoring and triage, helping to bridge healthcare disparities [14, 15]. Furthermore, AI integrates across diverse clinical specialties, including genomics, chronic disease predictive analytics, and robotic surgery, reflecting a shift to data-driven personalized medicine [10, 16]. Advanced AI analyzing multi-modal data like genomic sequences, proteomics, and real-time physiological signals advances predictive preventive healthcare [17]. Traditional medical education, deeply anchored in didactic lectures, textbook-oriented learning, and clinical apprenticeships, has long stood as a cornerstone for fostering foundational knowledge and cultivating hands-on practical expertise. Core components like animal experiments, cadaveric dissections, and two-dimensional imaging present challenges, including limited resource accessibility, time-intensive practices, and difficulties in visualizing three-dimensional structures. Additionally, medical students are confronted with overwhelming academic workloads, necessitating efficient study methodologies to master extensive curricula [18–20]. While these conventional approaches cultivate critical thinking and clinical competence, they struggle to meet the growing demand for technological literacy and personalized learning in contemporary healthcare [19]. The structural rigidity inherent in conventional educational paradigms frequently fails to adequately support personalized learning pathways, which is a critical deficiency systematically resolved by AI-powered adaptive learning platforms that implement real-time performance analytics-driven content optimization [21–23]. In medical education, AI-enhanced tools, including large language models like ChatGPT, facilitate personalized learning pathways by integrating adaptive knowledge delivery, instantaneous feedback mechanisms, and competency-based assessments, thereby addressing heterogeneous educational demands and temporal limitations inherent in clinical training [23–25]. The integration of virtual reality (VR) in anatomy education exemplifies how emerging technologies mitigate traditional limitations, providing immersive three-dimensional visualizations and interactive simulations that enhance spatial understanding [26]. AI holds immense promise in addressing these challenges by augmenting both clinical practice and medical training. Educational innovations, such as virtual patient simulations and AI-driven tutoring systems, provide immersive, scalable training platforms that bridge gaps in traditional methods [27, 28]. For future practitioners, early curricular exposure to AI fosters essential competencies in data analysis, ethical decision-making and human-AI collaboration, aligning with global initiatives such as Saudi Arabia’s Vision 2030 to cultivate technologically adept next-generation clinical professionals [22, 29]. Cross-sectional studies, including those evaluating Romanian medical students’ attitudes toward AI-driven telemedicine, indicate that acceptance is contingent upon factors such as technological reliability and prior exposure to AI tools [30, 31]. However, ethical considerations, such as data privacy, algorithmic bias, and patient consent, demand rigorous oversight to ensure equitable and transparent AI deployment [32, 33]. Variable efficacy reported in systematic reviews of tools like VR for anatomy education underscores the need for evidence-based, objective-specific implementation [26, 34]. Notwithstanding these innovations, resistance to technological adoption persists, driven in part by inadequate training, apprehensions on over-reliance on automated systems, and concerns regarding AI-induced employment reduction [1, 32, 35]. Concurrently, global disparities in AI integration within medical curricula persist conspicuously, with developing nations demonstrating a notable lag behind their developed counterparts in technological implementation [36]. Methods Questionnaire design The questionnaire was developed with reference to existing literature and structured into five main sections. These sections include: (1) demographic information (Items 1–4), collecting respondents’ gender, major, grade and school; (2) familiarity and usage of AI tools (Items 5–9); (3) perceived usefulness (Items 10–14); (4) perceived ease of use (Items 15–18); (5) ethical concerns and attitudes toward AI in education (Items 19–24) (Table 1 and Supplementary Table 1). Pilot testing was conducted to validate item clarity and relevance, with revisions made based on feedback. The finalized 24-item tool employs a hybrid response framework: 20 Likert-scale items (1 = Strongly Disagree to 5 = Strongly Agree) for attitudinal measurement, complemented by categorical demographic queries (Questions 1–4). Anonymity was guaranteed, and participation was entirely voluntary. Table 1Artificial intelligence in medical education: summary of questionnaire1What is your gender?A: MaleB: Female2What is your major?A: Medical relatedB: Other majors3What is your grade level?A: Grade 1A: Grade 2A: Grade 3A: Grade 4A: Grade 54Which school are you from?A: Jining Medical UniversityB: Binzhou Medical UniversityC: Shandong First Medical UniversityD: Shandong Second Medical UniversityE: Shandong University of Traditional Chinese MedicineFrom Question 5 to 24: A: Strongly Disagree; B: Disagree; C: Neutral; D: Agree; E: Strongly Agree5I am familiar with the fundamental concepts of artificial intelligence, such as machine learning and deep learning.6I have used AI tools to search for information or learn (such as DeepSeek, Doubao, ChatGPT, etc.)7I have used generative AI tools to submit assignments/final papers, etc.8I can distinguish the technical differences between generative AI (such as ChatGPT) and decision based AI (such as recommendation algorithms).9I am familiar with the application of AI in my field of study.10I agree that AI has improved my task completion efficiency.11I agree that AI can help me achieve better learning, academic, professional outcomes.12Compared to traditional resources, I believe that the information provided by AI is more accurate.13I believe that AI tools can achieve personalized learning experiences.14I believe that AI will play a crucial role in future education and career development, such as simulating clinical scenarios.15I agree AI tools are easy to learn and use.16I am able to independently solve problems encountered when using AI tools, without the help of technical support.17I agree AI interface design is intuitive and easy to operate.18The outputs generated by AI tools are clear and easy for me to understand and use.19I am concerned about the risk of privacy breaches when using AI tools.20I am worried that relying on AI will weaken critical thinking skills.21I am concerned that the use of AI generated content for academic achievement writing affects academic integrity.22I am concerned that AI may solidify biases or spread misinformation.23I am concerned that the development of AI may lead to job substitution issues.24I agree the ethical standards for AI applications in the current education field are not yet perfect. Table 1Artificial intelligence in medical education: summary of questionnaire1What is your gender?A: MaleB: Female2What is your major?A: Medical relatedB: Other majors3What is your grade level?A: Grade 1A: Grade 2A: Grade 3A: Grade 4A: Grade 54Which school are you from?A: Jining Medical UniversityB: Binzhou Medical UniversityC: Shandong First Medical UniversityD: Shandong Second Medical UniversityE: Shandong University of Traditional Chinese MedicineFrom Question 5 to 24: A: Strongly Disagree; B: Disagree; C: Neutral; D: Agree; E: Strongly Agree5I am familiar with the fundamental concepts of artificial intelligence, such as machine learning and deep learning.6I have used AI tools to search for information or learn (such as DeepSeek, Doubao, ChatGPT, etc.)7I have used generative AI tools to submit assignments/final papers, etc.8I can distinguish the technical differences between generative AI (such as ChatGPT) and decision based AI (such as recommendation algorithms).9I am familiar with the application of AI in my field of study.10I agree that AI has improved my task completion efficiency.11I agree that AI can help me achieve better learning, academic, professional outcomes.12Compared to traditional resources, I believe that the information provided by AI is more accurate.13I believe that AI tools can achieve personalized learning experiences.14I believe that AI will play a crucial role in future education and career development, such as simulating clinical scenarios.15I agree AI tools are easy to learn and use.16I am able to independently solve problems encountered when using AI tools, without the help of technical support.17I agree AI interface design is intuitive and easy to operate.18The outputs generated by AI tools are clear and easy for me to understand and use.19I am concerned about the risk of privacy breaches when using AI tools.20I am worried that relying on AI will weaken critical thinking skills.21I am concerned that the use of AI generated content for academic achievement writing affects academic integrity.22I am concerned that AI may solidify biases or spread misinformation.23I am concerned that the development of AI may lead to job substitution issues.24I agree the ethical standards for AI applications in the current education field are not yet perfect. Artificial intelligence in medical education: summary of questionnaire Inclusion criteria This anonymous cross-sectional study was conducted from May to June 2025, enrolling first- to fifth-year undergraduate medical students from five medical universities across Shandong Province (Jining Medical University, Binzhou Medical University, Shandong First Medical University, Shandong Second Medical University and Shandong University of Traditional Chinese Medicine). Including learners at various stages allowed the research to cover the continuum of medical education comprehensively, spanning fundamental medical courses, clinical training, and internship programs. Exclusion criteria To maintain focus on undergraduate medical education, we excluded: (1) non-medical program enrollees trainees; (2) participants with negative participation indicators, such as providing identical responses to over 70% of questions. This dual-filter approach ensured cohort homogeneity and data reliability through eligibility confirmation and response validation. Data collection tool Initial recruitment yielded 850 respondents, with 788 valid responses retained after rigorous quality control. Data collection employed structured questionnaires administered through the Wenjuanxing online survey mini-program, distributed via the WeChat platform. This digital approach ensured standardized questionnaire delivery and facilitated efficient data aggregation while maintaining methodological rigor. Statistical analyses Data processing and statistical evaluation were conducted using GraphPad Prism. Descriptive statistics characterized categorical variables as frequencies and percentages; while continuous variables were summarized using means ± standard deviations (SD). Given that the sample size of each group was greater than 50, the Likert-scale data were treated as continuous variables for analysis. For comparative analyses, data normality was verified using the Shapiro–Wilk test. Homoscedasticity was assessed with Bartlett’s test for three or more groups and with an F test for two-group comparisons. Independent t-tests were used to examine continuous variables between two groups, whereas one-way ANOVA with post-hoc Tukey tests was applied for multi-group comparisons (≥ 3 groups), with a significance threshold of *p < 0.05, **p < 0.01. Scale reliability was assessed via Cronbach’s α coefficients. Results Demographic data This multi-center cross-sectional study was conducted across five medical universities in Shandong Province, China, obtaining a total of 788 (92.71%) valid and 62 (7.29%) invalid anonymous responses (Fig. 1A), with subsequent analyses using only the valid responses. Among them, 464 (58.88%) were female and 324 (41.12%) were male (Fig. 1B, Supplementary Table 2). Approximately 45.30% (357) of responses came from Jining Medical University versus 54.70% (431) collectively from the other four universities (Fig. 1C). We specifically examined learners across distinct grades of the undergraduate medical education continuum in China: first-year (273, 34.64%) and second-year (190, 24.11%) students engaged in foundational preclinical sciences; third-year (149, 18.91%) students progressing to integrated organ-system curricula; fourth-year (94, 11.93%) participants immersed in core clinical clerkships; and fifth-year (82, 10.41%) interns completing advanced clinical rotations prior to graduation (Fig. 1D). Considering the strong inherent correlation between undergraduate academic performance and age in the general population, this study focused exclusively on analyzing grades while excluding age as a variable of investigation. Fig. 1Demographic information. A Number/proportion of valid and invalid questionnaires obtained. B Gender distribution of study participants. C Gender distribution of study participants, School 1–5 represent Jining Medical University, Binzhou Medical University, Shandong First Medical University, Shandong Second Medical University and Shandong University of Traditional Chinese Medicine, respectively. D Grade distribution of study participants Demographic information. A Number/proportion of valid and invalid questionnaires obtained. B Gender distribution of study participants. C Gender distribution of study participants, School 1–5 represent Jining Medical University, Binzhou Medical University, Shandong First Medical University, Shandong Second Medical University and Shandong University of Traditional Chinese Medicine, respectively. D Grade distribution of study participants Summary of overall responses to the questionnaire survey Basing on the valid responses, we analyzed the overall familiarity and usage of AI tools, PE, PEU, ethical concerns and attitudes toward AI in education (Fig. 2). Nearly half (47.33%) of participants report familiarity with fundamental AI concepts, while only negligible minorities (7.49%) claim no awareness of AI whatsoever. Application rates are notably high: 91.24% frequently utilize AI tools for research and learning purposes, with a mere 2.03% reporting no such usage. Furthermore, an overwhelming 79.95% engage AI for completing assignments and writing final papers; conversely, only 4.19% abstain entirely from using AI in these academic tasks. However, a critical knowledge gap persists, as only 65.99% demonstrate even a basic ability to distinguish between Generative AI (GenAI) and Decision-focused AI. Approximately 75.25% agree they understand AI’s applications within their specific field of study (Table 2).Table 2Responses toward “familiarity and use of AI Tools”Strongly DisagreeDisagreeNeutralAgreeStrongly AgreeN%N%N%N%N%Q5.597.4935645.1822528.558811.17607.61Q6.162.03536.7323930.3333842.8914218.02Q7.334.1912515.8631539.9719124.2412415.74Q8.8210.4118623.6032841.6216721.19253.17Q9.12215.48739.2617221.8334343.53789.90 Table 2Responses toward “familiarity and use of AI Tools”Strongly DisagreeDisagreeNeutralAgreeStrongly AgreeN%N%N%N%N%Q5.597.4935645.1822528.558811.17607.61Q6.162.03536.7323930.3333842.8914218.02Q7.334.1912515.8631539.9719124.2412415.74Q8.8210.4118623.6032841.6216721.19253.17Q9.12215.48739.2617221.8334343.53789.90 Responses toward “familiarity and use of AI Tools” Fig. 2Summary of overall responses to the questionnaire survey. This figure presents the medical students’ responses to questionnaire items assessing their familiarity, usage, attitudes, ethical and data security concerns regarding AI in medical education. Q stands for Question, and the detailed questions can be found in Table 1 and Supplementary Table 1 Summary of overall responses to the questionnaire survey. This figure presents the medical students’ responses to questionnaire items assessing their familiarity, usage, attitudes, ethical and data security concerns regarding AI in medical education. Q stands for Question, and the detailed questions can be found in Table 1 and Supplementary Table 1 The quantitative assessment of PU and PEU was conducted using Likert scales. Prior to statistical analysis, reliability was established through Cronbach’s α coefficients, which exceeded the recommended threshold of 0.7 for both constructs. Specifically, the internal consistency of the PU scale was α = 0.819, while the PEU scale achieved α = 0.824, indicating strong inter-item correlations and robust measurement reliability. PU attained a mean score of 3.60 ± 0.85, with 78.30% of responses exceeding the neutral threshold of 3, signifying favorable attitudes regarding AI’s utility. PEU demonstrated a higher mean of 3.66 ± 0.76, accompanied by 77.79% of scores above 3. This convergence of results, with clear majorities surpassing neutrality for both constructs, collectively indicates a generally positive perception among respondents concerning AI’s usefulness and user-friendliness (Table 3). Table 3The cronbach’s α coefficients and scale scores for ai’s PU and PEUαMeanSD95% CIPU (Q10 - Q14)0.8193.600.853.54–3.66PEU (Q15 - Q18)0.8243.660.763.63–3.68 Table 3The cronbach’s α coefficients and scale scores for ai’s PU and PEUαMeanSD95% CIPU (Q10 - Q14)0.8193.600.853.54–3.66PEU (Q15 - Q18)0.8243.660.763.63–3.68 The cronbach’s α coefficients and scale scores for ai’s PU and PEU Significant ethical and risk concerns emerged prominently. Privacy breaches during AI tool usage worry 48.48% of participants (Agree to Strongly Agree, hereafter else other indicated). A dominant 61.93% express apprehension about AI overreliance potentially eroding critical thinking faculties. Academic integrity is a paramount concern, with 55.84% highlighting anxieties regarding AI-generated content’s impact. Concerns about AI perpetuating biases or disseminating misinformation resonate strongly with 54.44% of respondents. Potential job displacement due to AI advancement concerns 53.68%, while a striking 59.26% perceive current ethical frameworks governing AI application as inadequate (Table 4). This collective apprehension underscores a profound tension: despite widespread adoption and largely positive perceptions of AI’s usability and usefulness, substantial ethical misgivings and perceived risks regarding its societal and academic impact prevail among the student body. The coexistence of high utilization and deep-seated concerns presents a complex landscape for integrating AI responsibly within the academic environment. Table 4Responses toward “ethical and data security concerns”Strongly DisagreeDisagreeNeutralAgreeStrongly Agree N % N % N % N % N %Q19.131.65334.1936045.6928235.7910012.69Q20.50.63587.3623730.0835144.5413717.39Q21.50.63192.4132441.1233542.5110513.32Q22.20.25293.6832841.6233242.139712.31Q23.81.02475.9631039.3431640.1010713.58Q24.40.51151.9030238.3235645.1811114.09 Table 4Responses toward “ethical and data security concerns”Strongly DisagreeDisagreeNeutralAgreeStrongly Agree N % N % N % N % N %Q19.131.65334.1936045.6928235.7910012.69Q20.50.63587.3623730.0835144.5413717.39Q21.50.63192.4132441.1233542.5110513.32Q22.20.25293.6832841.6233242.139712.31Q23.81.02475.9631039.3431640.1010713.58Q24.40.51151.9030238.3235645.1811114.09 Responses toward “ethical and data security concerns” Factors influencing familiarity with and usage of AI tools Analysis of variance revealed distinct demographic influences on AI tool familiarity and usage. Male participants demonstrated higher proficiency than female participants, suggesting a gender-based divergence in technology engagement (Table 5). In contrast, institutional affiliation showed negligible effect, as ANOVA indicated no significant school-level differences despite mean variations (Table 6). Academic progression, however, exerted a substantial influence. The ANOVA yielded a highly significant F-value of 26.754 (p < 0.001), with mean scores escalating from 2.93 to 3.49 (SDs: 0.60 to 0.66) across grade levels (Table 7), indicating progressively stronger AI competency with academic advancement. These findings collectively demonstrate that while institutional environment does not significantly predict AI familiarity, both gender stratification and educational maturation serve as critical determinants. Such insights warrant consideration in developing differentiated pedagogical approaches and resource allocation strategies. Table 5Gender differences across all dimensions related to AICategoryMale (n = 324)Mean ± SDFemale (n = 464)Mean ± SDt valuep valueFamiliarity and usage (Q5 - Q9)3.21 ± 0.703.10 ± 0.632.3190.020*PU (Q10 - Q14)3.62 ± 0.673.59 ± 0.590.5150.607PEU (Q15 - Q18)3.66 ± 0.643.65 ± 0.570.2090.834Ethical concerns and attitudes (Q19 - Q24)3.63 ± 0.603.65 ± 0.500.5100.611 Table 5Gender differences across all dimensions related to AICategoryMale (n = 324)Mean ± SDFemale (n = 464)Mean ± SDt valuep valueFamiliarity and usage (Q5 - Q9)3.21 ± 0.703.10 ± 0.632.3190.020*PU (Q10 - Q14)3.62 ± 0.673.59 ± 0.590.5150.607PEU (Q15 - Q18)3.66 ± 0.643.65 ± 0.570.2090.834Ethical concerns and attitudes (Q19 - Q24)3.63 ± 0.603.65 ± 0.500.5100.611 Gender differences across all dimensions related to AI Table 6School differences across all dimensions related to AICategory1 (n = 357)Mean ± SD2 (n = 191)Mean ± SD3 (n = 77)Mean ± SD4 (n = 89)Mean ± SD5 (n = 74)Mean ± SDF valuep valueFamiliarity and usage (Q5 - Q9)3.11 ± 0.663.15 ± 0.653.28 ± 0.633.14 ± 0.703.17 ± 0.691.0950.358PU (Q10 - Q14)3.61 ± 0.633.63 ± 0603.71 ± 0.633.52 ± 0.743.50 ± 0.541.5650.182PEU (Q15 - Q18)3.68 ± 0.603.68 ± 0.553.69 ± 0.653.53 ± 0.643.57 ± 0.591.6120.169Ethical concerns and attitudes (Q19 - Q24)3.62 ± 0.563.67 ± 0.573.68 ± 0.573.63 ± 0.533.62 ± 0.660.4040.806 Table 6School differences across all dimensions related to AICategory1 (n = 357)Mean ± SD2 (n = 191)Mean ± SD3 (n = 77)Mean ± SD4 (n = 89)Mean ± SD5 (n = 74)Mean ± SDF valuep valueFamiliarity and usage (Q5 - Q9)3.11 ± 0.663.15 ± 0.653.28 ± 0.633.14 ± 0.703.17 ± 0.691.0950.358PU (Q10 - Q14)3.61 ± 0.633.63 ± 0603.71 ± 0.633.52 ± 0.743.50 ± 0.541.5650.182PEU (Q15 - Q18)3.68 ± 0.603.68 ± 0.553.69 ± 0.653.53 ± 0.643.57 ± 0.591.6120.169Ethical concerns and attitudes (Q19 - Q24)3.62 ± 0.563.67 ± 0.573.68 ± 0.573.63 ± 0.533.62 ± 0.660.4040.806 School differences across all dimensions related to AI Table 7Grade differences across all dimensions related to AICategory1 (n = 273)Mean ± SD2 (n = 190)Mean ± SD3 (n = 149)Mean ± SD4 (n = 94)Mean ± SD5 (n = 82)Mean ± SDF valuep valueFamiliarity and usage (Q5 - Q9)2.93 ± 0.603.00 ± 0.603.33 ± 0.643.48 ± 0.663.49 ± 0.6326.754< 0.001**PU (Q10 - Q14)3.61 ± 0.663.60 ± 0.583.59 ± 0.663.66 ± 0.633.54 ± 0.600.4390.781PEU (Q15 - Q18)3.70 ± 0.613.64 ± 0.593.60 ± 0.613.68 ± 0.663.60 ± 0.490.9790.418Ethical concerns and attitudes (Q19 - Q24)3.44 ± 0.553.58 ± 0.523.78 ± 0.503.85 ± 0.473.93 ± 0.4724.651< 0.001** Table 7Grade differences across all dimensions related to AICategory1 (n = 273)Mean ± SD2 (n = 190)Mean ± SD3 (n = 149)Mean ± SD4 (n = 94)Mean ± SD5 (n = 82)Mean ± SDF valuep valueFamiliarity and usage (Q5 - Q9)2.93 ± 0.603.00 ± 0.603.33 ± 0.643.48 ± 0.663.49 ± 0.6326.754< 0.001**PU (Q10 - Q14)3.61 ± 0.663.60 ± 0.583.59 ± 0.663.66 ± 0.633.54 ± 0.600.4390.781PEU (Q15 - Q18)3.70 ± 0.613.64 ± 0.593.60 ± 0.613.68 ± 0.663.60 ± 0.490.9790.418Ethical concerns and attitudes (Q19 - Q24)3.44 ± 0.553.58 ± 0.523.78 ± 0.503.85 ± 0.473.93 ± 0.4724.651< 0.001** Grade differences across all dimensions related to AI Factors influencing PU and PEU Based on the analysis of combined PU and PEU scores, no significant differences emerged across gender, grade, or school affiliation. Gender comparisons revealed closely aligned means for both PU and PEU (Table 5). These statistically non-significant results indicate gender does not meaningfully influence either construct. The consistently high p-values confirm grade progression does not significantly affect outcomes. School affiliation likewise demonstrated non-significant effects for both PU and PEU (Table 6). Similarly, grade-level analysis showed minimal variation in PU and PEU scores (Table 7). This comprehensive analysis demonstrates robust consistency in PU and PEU perceptions across all examined demographic variables within the studied population. Factors influencing ethical concerns and attitudes toward AI in education Based on the analysis of combined privacy and ethics concerns, significant differences were observed only for grade level. As presented in the results, no significant difference was found between genders, as indicated by the T-test, in which mean scores were closely aligned (Table 5). Similarly, the one-way ANOVA for school affiliation yielded no significant effect. Mean scores across schools were highly consistent, ranging narrowly from 3.62 to 3.68 (SDs: 0.53 to 0.66) (Table 6). In contrast, grade level demonstrated a highly significant impact (F-value: 24.651; p < 0.001) (Table 7). Mean concern scores varied substantially across grades, from 3.44 to 3.93 (SDs: 0.47 to 0.55), indicating that students’ privacy and ethics concerns regarding AI tools increase meaningfully with academic progression. These findings underscore grade as the primary differentiating factor in this study. Discussion Inter-institutional homogeneity in AI attitudes The absence of significant attitudinal differences toward AI among medical students across five Shandong institutions (p > 0.05 across measured dimensions) highlights a homogenized educational landscape shaped by China’s standardized curricula and centralized policy frameworks. Systemic adherence to national guidelines has neutralized institutional disparities in resources or pedagogy. Foundational AI literacy modules integrated into core curricula, alongside uniform infrastructure initiatives, ensured consistent exposure. The universal adoption of AI tools (such as DeepSeek, ChatGPT for literature analysis/writing) and centralized platforms (such as Chaoxing Learning Platform and Chinese University MOOC) has normalized engagement, and bridged potential gaps in technological access through policy-mandated resource standardization. China’s medical education framework demonstrates how policy coherence and infrastructural parity cultivate uniform technological literacy, even across diverse institutions. These findings align with research on centralized educational systems, where policy-driven standardization typically overrides institutional idiosyncrasies in technology implementation, fostering uniform technological literacy among learners [37]. However, the generalizability of this finding may be influenced by the uneven sample distribution across institutions and the use of convenience sampling. Gender perceptions of AI Gender influences attitudes toward AI in healthcare education and practice, though findings vary. Most studies indicate that male students exhibit higher enthusiasm and engagement with AI technologies, demonstrating more confidence in their diagnostic capabilities and learning assistance [1, 19, 24, 28]. In contrast, female students often express greater concerns about privacy, data security, and ethical implications of AI [22, 28, 38], while showing more positive attitudes toward AI integration in education and clinical practice [22, 38]. Additionally, female respondents tend to emphasize the need for ethical guidelines and curricular integration of AI [1, 38]. Moreover, Iranian dental studies highlighted nuanced gender dynamics: females prioritized AI integration into curricula, while males expressed stronger trust in AI’s diagnostic capabilities [1]. However, some studies indicated that gender showed no statistically significant impact. Multivariate models revealed inconsistent gender differences, while satisfaction with ChatGPT’s feedback demonstrated no gender-based variation [30, 39]. The significant variations in satisfaction were primarily associated with whether ChatGPT was used during class, rather than gender-related factors [19, 39]. Statistical analysis of our survey data revealed that while male respondents reported slightly higher levels of familiarity with and usage of AI technologies (mean 3.21 vs. 3.10, p = 0.020), no significant gender-based differences were found in attitudes, perceived correctness, practical utility, or ethical concerns across the measured parameters. This suggests that while a baseline gender gap in exposure may exist, it does not translate into significantly divergent perceptions in our cohort. These findings should be interpreted considering potential self-report biases and the lack of multivariable adjustment for confounding factors. The significant impact of academic year on AI attitudes, usage, and concerns The academic year serves as a powerful determinant of medical students’ acceptance, perceived utility, and concerns regarding AI. A pronounced progression in attitudes is evident, with senior students (Years 4–5) consistently exhibiting significantly higher levels of AI acceptance and adoption intent [28, 30]. This heightened receptivity among advanced learners may primarily attribute to cumulative clinical exposure. Within clinical environments, particularly during rotations such as radiology, seniors directly observe and engage with practical AI applications like diagnostic algorithms and telemedicine tools. This experiential learning fosters a concrete understanding of AI’s capacity to augment clinical efficiency and support complex decision-making, aligning with their advanced curricula’s emphasis on evidence-based practice [28, 30]. Our data strongly support this, showing a significant increase in AI familiarity and usage scores with academic year (F = 26.754, p < 0.001). However, this contradicts the findings reported by ALruwail BF et al., who demonstrated that the level of AI practice among interns was significantly lower than that among students in the early stages of medical education [22]. The discrepancy may be attributed to the transitional nature of internship, where clinical responsibilities and time constraints could potentially limit interns’ exposure to and engagement with AI tools. Conversely, junior students (Years 1–2), who are predominantly immersed in foundational biomedical sciences, often perceive AI as an abstract concept that is less immediately relevant to their core learning objectives. They tend to view AI primarily as a theoretical learning aid, such as adaptive quizzes, rather than a tool with direct clinical applications [28]. Their limited exposure to real-world healthcare scenarios hinders appreciation of AI’s clinical utility. This is evidenced by the highest scores for ethical concerns among fifth-year students (3.93 ± 0.47). However, a critical nuance emerges with interns and those entering clinical practice. While their usage intent remains high due to practical necessity, this group paradoxically exhibits the most pronounced ethical concerns. This reflects heightened awareness of AI’s tangible implications for patient care, professional responsibility, and clinical judgment gained through direct patient interaction and responsibility. Thus, while clinical exposure drives acceptance and perceived usefulness, it simultaneously amplifies awareness of potential risks, leading to a more complex, cautiously optimistic stance among those closest to independent practice [40]. This consistent trend across diverse geographical contexts underscores the imperative for staged curricular integration, progressively introducing AI concepts and applications aligned with students’ evolving clinical competencies and experiences [38]. Pervasive ethical and data security concerns Ethical reservations and data security apprehensions constitute major, pervasive barriers to AI adoption among medical students [1, 19, 22, 28, 30, 34, 41]. In our study, these concerns were notably high, with over 48% of participants agreeing or strongly agreeing with statements about privacy risks, erosion of critical thinking (61.93%), and threats to academic integrity (55.84%). Data privacy emerged as the paramount concern, driven by well-publicized breaches and fears of unauthorized data monetization, particularly concerning tools requiring sensitive personal health information. Closely intertwined are profound anxieties regarding algorithmic accountability and bias. Biased training data has been shown to lead to misdiagnoses, eroding trust in AI’s objectivity and fairness. The “black box” nature of many algorithms, lacking transparency in decision-making processes, further amplifies skepticism and hesitation [42, 43]. The advent and growing prominence of GenAI significantly amplifies pre-existing ethical concerns by introducing unique risks. Particular apprehension centers on its capacity for fraudulent data generation and misinformation propagation. GenAI models, capable of synthesizing highly plausible text, images, or simulated patient data, may inadvertently yield vast quantities of spurious outputs: factually inaccurate, clinically misleading, or entirely fabricated [44, 45]. This capability poses severe threats within medicine. Integrating erroneous synthetic data into educational resources risks misinforming learners; its infiltration into clinical decision support systems or research datasets could precipitate catastrophic diagnostic errors, flawed research conclusions, and patient harm. GenAI’s propensity to fabricate citations, fabricate nonexistent clinical evidence, or generate convincing yet inaccurate patient simulations represents a distinct vector for epistemological degradation in medical education and practice [46, 47]. Furthermore, its ability to create sophisticated “deepfakes” exacerbates concerns regarding data provenance and authenticity, undermining confidence in digital information essential for evidence-based clinical reasoning. Study limitations This study exhibits several methodological limitations. First, the use of convenience sampling and self-reported data from a single province limits the generalizability of the findings and introduces potential for self-report bias. Second, the cross-sectional design and lack of multivariable adjustment preclude causal inferences and carry a risk of residual confounding. Third, the imbalance in participant numbers across the five institutions may affect the robustness of inter-institutional comparisons. Fourth, we failed to adequately incorporate and analyze two variables that may influence technological attitudes. On the one hand, we neglected to incorporate geographic origin of the students (specifically the urban-rural divide) as a variable factor. While internet access is increasingly effective across both urban and rural China, regional factors still significantly shape the quality of technological infrastructure, the depth of exposure to AI training resources, and the nature of AI-driven educational experiences. As reported in some studies, regional disparities impact AI usage and attitudes, with urban-rural digital divides in developing nations affecting exposure, while developed regions focus on ethics and privacy [13, 22, 30]. On the other hand, parental education levels, a factor strongly correlated with household digital literacy and intergenerational transmission of technological attitudes, were not measured. Prior research indicates that parents with higher educational backgrounds tend to foster greater comfort with emerging technologies, potentially influencing their children’s acceptance of AI [28]. Conversely, students from families with lower parental education might exhibit more caution or resistance to AI due to limited home-based technological support or generational gaps in digital fluency. Ignoring this variable means the study cannot disentangle whether observed attitudes toward AI stem from students’ personal experiences or familial influences, creating an incomplete picture of the factors shaping their perceptions. Conclusion This cross-sectional study, conducted across five medical universities in Shandong Province, China, investigates medical students’ attitudes toward AI integration in medical education. It reveals that while participants report high familiarity with AI and endorse its PU and PEU, significant ethical and privacy concerns persist as key barriers to broader acceptance. These findings underscore the necessity of modernizing medical curricula, and establishing transparent governance frameworks specifically designed to address these ethical challenges. Implementing such measures would facilitate the responsible integration of AI into healthcare education and future clinical practice. Supplementary Information Supplementary Material 1. Supplementary Table 1: Artificial Intelligence in Medical Education: Questionnaire. Supplementary Table 2: Detail participants gender distribution by school and grade. Supplementary Material 1. Supplementary Table 1: Artificial Intelligence in Medical Education: Questionnaire. Supplementary Table 2: Detail participants gender distribution by school and grade.