AI is being increasingly used in the education sector (Table 1), and its applications can be mainly categorized into two types: learning and entertainment. The impact of these two types of applications on students’ creativity has drawn significant attention, and existing research has conducted numerous explorations (2). AI used for learning refers to the application of AI technology to assist students in acquiring knowledge, enhancing skills, promoting understanding, and reinforcing learning content (13). It encompasses various aspects such as intelligent tutoring, providing learning resources, and facilitating teaching, aiming to optimize the learning process and improve learning outcomes. Existing research in this area mainly focuses on intelligent tutoring, learning resource provision, and teaching assistance (14, 15). Regarding the impact on students’ creativity, on one hand, the abundant resources provided by AI can broaden students’ knowledge horizons and stimulate innovative thinking (16). On the other hand, excessive reliance on AI may lead to a decline in students’ independent thinking ability and thus inhibit the development of creativity (17). AI used for entertainment refers to the use of AI technology to create entertaining experiences for students, integrating educational elements into the entertainment process to achieve learning through entertainment. Existing research in this area mainly focuses on gamified learning and virtual experiences (18, 19). As for the impact on students’ creativity, on the positive side, it can stimulate students’ curiosity and exploration desire, allowing them to learn and create in an interesting environment (2). On the negative side, if the entertainment component is overly emphasized, it may distract students from learning and prevent them from focusing on knowledge acquisition and creativity development (16).

Recent studies have increasingly emphasized the role of frontier technologies such as generative AI, virtual reality (VR), and blockchain in education and creativity research. Generative AI has been shown to facilitate personalized learning and foster creative thinking by providing adaptive feedback and content generation (20). VR has been applied to create immersive learning environments that enhance experiential training, while blockchain has been explored for its potential to ensure secure data management and transparent credentialing in higher education (21, 22). However, despite these promising developments, very limited research has systematically examined the application of these technologies within the context of Traditional Chinese Medicine (TCM) education. This gap highlights the originality and necessity of the present study, which aims to explore how AI-enabled approaches can contribute to creativity development in specialized domains such as TCM.

Social Cognitive Theory (SCT) emphasizes the dynamic interplay between individual behavior, environmental influences, and personal factors (23). It posits that learning occurs through observation, imitation, and modeling, with individuals developing capabilities by observing others’ behaviors and outcomes (24). SCT highlights the importance of self-efficacy, the belief in one’s ability to succeed in specific tasks, which influences motivation and behavior (25). In the context of this study, SCT provides a robust framework to understand how AI use associated with student creativity through digital competencies. AI tools, when used for learning or entertainment, create an environment where students observe and model behaviors related to digital engagement (26, 27). This exposure enhances their digital competencies, which in turn fosters creativity. Additionally, students’ attitudes toward AI reflect their self-efficacy beliefs, influencing how effectively they engage with AI tools. Thus, SCT explains the observed relationships between AI use, digital competencies, and creativity by emphasizing the role of environmental influences (AI tools), personal factors (attitudes), and behavioral outcomes (creativity). Beyond serving as a general explanatory framework, SCT in this study is extended to conceptualize how different types of AI use function as distinct environmental stimuli, how digital competencies operate as a core cognitive capability, and how attitudes toward AI shape the effectiveness of these stimuli. In doing so, this study advances SCT by specifying a technology-centered mechanism through which environmental factors are transformed into creative outcomes in AI-supported educational contexts.

Student digital competencies refer to the set of skills, knowledge, and attitudes that enable students to effectively use digital technologies for learning, communication, problem solving, and creative expression in various contexts (28). Traditional tools emphasize operational proficiency (e.g., mastering microscope adjustments), while AI tools prioritize digital literacy in the intelligent era, such as understanding how AI processes data, evaluating model accuracy, and optimizing algorithm parameters (Table 2). Thus, investigate the relationship between AI and digital competencies is particularly important.

When student use AI for learning purposes, it may positively associate with student digital competencies. The reasons maybe that: First, AI tools can provide students with personalized learning pathways and resources, helping them acquire knowledge and skills more efficiently (29, 30). For example, AI driven learning platforms can tailor content based on individual progress and abilities, enhancing students’ proficiency in using digital tools and learning outcomes. Additionally, AI’s intelligent feedback mechanisms help student correct mistakes in real time, strengthening their self-directed learning and problem solving abilities (31), which are key components of digital competencies. Moreover, AI can create enriched and interactive learning environments through virtual labs and online collaboration tools, further boosting students’ digital literacy. These factors collectively make AI a powerful driver for enhancing student digital competencies in learning contexts. Based on above discussions, this paper proposed the following hypothesis:

Another important purpose student use AI is for entertainment purposes (32). The logic behind AI used for entertainment positively associates with student digital competencies is mainly realized through several variables. First, AI driven entertainment content (such as intelligent games and virtual reality experiences) typically requires students to possess certain digital operation skills, such as using complex software or hardware (33). This interactivity prompts students to continuously learn and adapt to new digital tools and technologies while being entertained, thereby improving their digital competencies. Second, creative elements in AI entertainment applications (such as programming games or digital art creation tools) can stimulate students’ creativity and digital literacy (34). Through these tools, students not only enjoy entertainment but also learn and apply digital skills in practice. Furthermore, AI in entertainment can help students better understand and utilize digital resources through recommendation systems and personalized experiences (35), further enhancing their digital competencies. This subtle influence in entertainment makes AI an important pathway for improving student digital competencies. Based on above discussions, this paper proposed the following hypothesis:

Student creativity refers to the ability of students to generate novel, original, and valuable ideas or solutions in various contexts (36). It involves the capacity to think divergently, explore multiple possibilities, and apply imaginative thinking to solve problems or create new products. Digital competencies may positively influence student creativity through enhanced information processing and collaborative abilities. First, digital competencies enable students to access, filter, and integrate information more efficiently, providing a rich pool of resources for creative thinking (10). For example, students can leverage online resources to gather cross disciplinary knowledge, which can spark new ideas and solutions. Second, digital tools support collaborative learning, allowing students to interact with diverse peers in virtual teams. This multicultural exchange fosters creative collisions and innovation. Additionally, digital competencies facilitate the expression and presentation of creative ideas, such as using multimedia tools to visualize abstract concepts (37), thereby enhancing the realization and dissemination of creativity. Therefore, digital competencies positively associate with student creativity through variables like information integration, collaborative interaction, and creative expression. Based on above discussions, this paper proposed the following hypothesis:

Attitude toward AI is defined as an individual’s cognitive, emotional, and behavioral tendency toward artificial intelligence (12). This paper posits that the more positive students’ attitudes toward artificial intelligence (AI), the weaker the effect of AI on enhancing students’ digital competencies during the learning process. This negative moderating effect may be based on the following points: When students hold a highly positive attitude toward AI, they may become overly reliant on the instant feedback and solutions provided by AI (38), thereby reducing opportunities for independent thinking and problem solving (39, 40). This dependency may lead to a lack of independent analysis and problem solving skills when students face complex issues, thus weakening the enhancement of their digital competencies. For example, students may habitually rely on AI’s suggestions rather than actively exploring and learning new skills and knowledge. A positive attitude may also shift students’ learning motivation more toward the process of using AI rather than the learning content itself (41, 42). This shift in motivation may cause students to focus more on how to use AI during the learning process, rather than on how to enhance their digital competencies through AI. For example, students may enjoy the process of interacting with AI more than deeply understanding and applying the knowledge and skills provided by AI. Based on above discussions, this paper proposed the following hypothesis:

This paper proposes that students’ attitudes toward artificial intelligence (AI) play a positive moderating role between AI used for entertainment and student digital competencies. This means that when students hold positive attitudes toward AI, its application in entertainment contexts may more effectively promote the development of students’ digital competencies. The rationale behind this logic is that a positive attitude can enhance students’ acceptance and willingness to use AI tools (41), especially in entertainment contexts where AI applications are often more engaging and enjoyable. For example, through AI driven games or interactive entertainment applications, students may unconsciously improve their digital skills (43, 44), such as data analysis, problem solving, and logical thinking, while enjoying entertainment. Moreover, a positive attitude may also encourage students to actively explore more functionalities of AI (33), thereby further enhancing their digital literacy. Therefore, students’ positive attitudes toward AI may amplify the association between AI and digital competencies in entertainment contexts, making it an effective learning tool. Based on above discussions, this paper proposed the following hypothesis:

Based on the above discussion, the proposed research model is presented in Figure 1.

In this paper, a questionnaire was used to collect data. Each measurement item was drawn from previous literature (see Table 3). AI used for learning (AL) included five items and was taken from (45). AI used for entertainment (AE) included four items and was taken from Ali-Hassan, Nevo and Wade (46). Attitudes toward AI (AT) was measured with four items derived from Shao, Nah, Makady and McNealy (47). Digital competencies (DC) was measured with four items derived from David, Zinica, Barbuta-Misu, Savga and Virlanuta (48). Student creativity (SC) included four items and was taken from Jia, Luo, Fang and Liao (49). All of the above items were scored on a seven-point Likert scale ranging from 1 (strongly disagree/unlikely) to 7 (strongly agree/likely). In this study, all measurement scales were either adopted or adapted from well-established instruments in prior literature. For scales originally developed in non-educational or organizational contexts (e.g., AI used for entertainment), an adaptation process was conducted to ensure their suitability for the TCM undergraduate learning environment. Specifically, all items were carefully reworded to fit the educational context by replacing work-related expressions (e.g., “job” or “work”) with study-related expressions (e.g., “learning” or “study”). To enhance content validity, the adapted questionnaire was reviewed by two scholars specializing in educational technology and one expert in Traditional Chinese Medicine (TCM) education, who evaluated the clarity, relevance, and contextual appropriateness of each item. In addition, a pilot test was conducted with 30 TCM undergraduate students to examine item comprehension and wording suitability. Minor revisions were made based on their feedback. Furthermore, the validity of all adapted scales was empirically supported by the results of the measurement model. All constructs demonstrated high factor loadings, strong internal consistency reliability, and satisfactory convergent and discriminant validity, indicating that the adapted instruments were appropriate for the TCM educational context. This adaptation approach is consistent with prior studies that have successfully transferred technology-use and digital behavior scales from organizational settings to educational contexts (50), supporting the feasibility of contextual scale adaptation.

Considering that generative artificial intelligence is currently the most widely used type of AI among college students in TCM schools, the AI investigated in this paper primarily refers to generative AI. This study conducted data collection through an online questionnaire to efficiently obtain rich data. The samples were sourced from undergraduate students majoring in TCM at a university of shandong traditional Chinese medicine. TCM programs in China share similar curricula. After designing the questionnaire on the platform, we asked the counselors to invite undergraduate students from the college to complete it. The questionnaire was distributed through Soujump, an online survey platform, which facilitated easy access and completion by the respondents. This platform is widely used by previous scholars (51). In addition, the platform only allowed one questionnaire to be completed by each IP address, which prevented duplicate questionnaires from being completed by the same respondent. After the twoweek data collection process, this study collected 271 completed questionnaires. The final valid sample consisted of 271 undergraduate students majoring in Traditional Chinese Medicine. To further evaluate whether this sample size was sufficient for the complexity of the moderated-mediation model, a post hoc power analysis was conducted using G*Power 3.1. With N = 271, α = 0.05, and five predictors, the achieved power to detect a medium effect size (f² = 0.15) was approximately 1.00. This indicates that the sample size was more than adequate to detect medium effects, although very small effects may not be reliably captured.

The descriptive statistics of respondent characteristics are presented as follows (Table 4). Based on the demographic data from questionnaire, the gender distribution revealed that males comprised 32.1% of the respondents, while females accounted for 67.9%. Regarding their educational backgrounds, freshmen made up 33.58% of the sample, followed by a minimal 0.74% representation from sophomores. Juniors constituted the largest group at 64.21%, with both seniors and those pursuing master’s degrees or higher each contributing another 0.74%. When examining the participants’ experience with artificial intelligence tools, it was found that 33.58% had less than 1 year of experience, 32.84% had between one and 2 years, 15.87% had between 3 and 5 years, and 14.76% had more than 5 years of experience. In terms of the purposes for using artificial intelligence, a significant 77.12% of respondents utilized AI for learning, while 20.66% used it for entertainment, and a smaller portion of 2.21% cited other purposes.

This study assessed the reliability and validity of the measurement model. Specifically, as demonstrated in Table 5, Cronbach’s alpha coefficients for the constructs fell within a range of 0.887 to 0.924, and composite reliability values ranged from 0.922 to 0.946, both exceeding the 0.7 threshold, thus indicating solid reliability. In terms of validity, the study evaluated both convergent and discriminant validity. The average variance extracted (AVE) for all constructs exceeded 0.7, and all factor loadings listed in Table 6 surpassed the 0.7 threshold, indicating strong convergent validity. Additionally, the square root of each construct’s AVE, as shown in Table 5, exceeded its correlations with other constructs.

To evaluate the overall model fit, multiple fit indices were examined, including SRMR, NFI, d_ULS, and d_G. As shown in Table 7, the SRMR value of 0.060 is below the recommended threshold of 0.08, indicating a good approximate model fit. The NFI value of 0.838 exceeds the acceptable cut-off of 0.70, suggesting an adequate incremental fit. Moreover, both d_ULS and d_G are lower than their corresponding 95% bootstrap quantiles, further supporting the satisfactory fit of the proposed model. Taken together, these results demonstrate that the overall model fit is acceptable and suitable for hypothesis testing as shown in Figure 2. Specifically, the results showed that AI used for learning positively associated with digital competences (β = 0.348, T = 4.307), indicating H1 is supported. AI used for entertainment positively associated with digital competences (β = 0.303, T = 5.140), indicating H2 is supported. Digital competence (β = 0.844, T = 37.716) positively associated with student creativity, thus supporting H3.

To test the mediating effect, hypothesis testing was carried out using PROCESS in SPSS with the bias-corrected method and the percentile method at a 95% confidence interval (52). In this approach, the mediating effect of the mediator variable was supported when the confidence interval for the indirect effect does not include zero. A distinction was also made between full and partial mediation depending on whether the confidence interval for the direct effect includes zero. As Table 8 shown, digital competences completely mediated the relationship between AI used for learning and student creativity. Besides, digital competences partially mediated the relationship between AI used for entertainment and student creativity. These results indicated that AI used for learning and AI used for entertainment directly affect student creativity and indirectly through digital competences.

As shown in Table 9, this study conducted multilevel regression analyses of the moderating effects of attitudes toward AI. The results showed that attitudes toward AI negatively moderated the relationship between AI used for learning and digital competences (β = −0.135, T = 2.527) as Figure 3 showing, implying that H9a was supported. This moderating effect is further illustrated in Figure 4. In addition, the results showed that attitudes toward AI positively moderated the relationship between AI used for entertainment and digital competences (β = 0.213, T = 3.236) as Figure 3 showing, implying that H9b was supported. This moderating effect is further illustrated in Figure 5.

The IPMA results (see Figure 6) provide nuanced insights into how the four predictors, AI used for learning (AL), AI used for entertainment (AE), attitudes toward AI (AT), and digital competencies (DC), contribute to student creativity (SC) in terms of both importance (total effects) and performance.

Overall, AT and DC emerge as the most influential variables with relatively high importance values. Attitudes toward AI (AT) demonstrates the highest total effect, indicating that students’ positive perceptions and acceptance of AI play a key role in enhancing creativity. At the same time, AT also shows relatively high performance, suggesting that students generally exhibit favorable attitudes toward AI, and further strengthening this dimension may generate additional creative benefits.

Digital competencies (DC) also shows a strong importance level, ranking second among all predictors. However, its performance is only moderate relative to AT, implying that while digital competence is a critical driver of creativity, there remains substantial room for improvement. Strengthening students’ digital skills may therefore produce meaningful gains in creative outcomes.

In contrast, AI used for learning (AL) displays moderate performance but lower importance, suggesting that while students frequently use AI for academic purposes, its direct impact on creativity is less substantial than attitudinal or skill-related factors. This indicates that general learning-related AI use may not automatically translate into higher creativity without supportive cognitive or motivational mechanisms.

Finally, AI used for entertainment (AE) exhibits both lower importance and lower performance, making it the least influential predictor of student creativity. This suggests that entertainment-oriented AI use contributes minimally to creativity enhancement and should not be prioritized in instructional design or student development strategies.

Taken together, the IPMA highlights AT and DC as priority development areas, with attitudes offering the strongest leverage and digital competencies representing a high-impact but underdeveloped dimension. Conversely, AL and AE provide limited value for improving student creativity from an importance–performance perspective.

A potential concern in observational studies is endogeneity, as more creative students may also be more inclined to adopt AI tools, making reverse causality plausible. To address this concern, we applied the Gaussian copula approach (53) to test whether AI used for learning (AL) and AI used for entertainment (AE) are endogenous. Specifically, copula terms were generated for AL and AE by transforming them into normal scores based on their empirical distribution, and these terms were included in the regression models. The results showed that none of the copula terms were statistically significant (p > 0.10), suggesting no evidence of endogeneity in the current study. Nevertheless, given the cross-sectional nature of the data, reverse causality cannot be fully ruled out. Future research should employ longitudinal or experimental designs, or instrumental variables, to further mitigate endogeneity concerns.

To further ensure the robustness of our findings, we conducted three additional analyses. First, we included all control variables (Model 1), and the main associations, indirect effects via digital competencies, and moderating effects remained consistent with the baseline model. Second, we selectively removed control variables that had a significant effect on creativity (Model 2), and third, we removed those that were not significant (Model 3). In both cases, the results were consistent with the original theoretical model. These analyses indicate that the findings are highly robust and not driven by the inclusion of specific moderating or control variables.

The present study explores the relationship between AI use, digital competencies, and student creativity among TCM students in China. Our findings reveal several important insights that contribute to the existing literature on AI’s role in education and creativity development. Firstly, we found that AI used for learning (AL) and AI used for entertainment (AE) both positively associated with digital competencies (DC) among TCM students. Specifically, AL positively positively associated with DC (β = 0.349, T = 4.281), and AE also positively associated with DC (β = 0.349, T = 5.120). This result is consistent with previous research that suggests AI can enhance digital skills through various applications in both academic and recreational contexts (43, 54). However, our study extends this understanding by demonstrating that even in a specialized field like TCM, AI’s dual role in learning and entertainment can effectively promote digital competencies. This finding highlights the versatility of AI as a tool for skill development across diverse educational domains.

Secondly, digital competencies were found to be positively associated with student creativity (β = 0.685, T = 13.155) within the context of TCM undergraduate education. This result aligns with the broader literature on the importance of digital skills in fostering creativity (37, 55). However, our study innovatively demonstrates that in the context of TCM education, digital competencies play a crucial mediating role between AI use and creativity. This suggests that the development of digital competencies through AI use is not just a standalone outcome but also a pathway to enhancing creativity in specialized fields. This mediating effect highlights the importance of integrating digital literacy into the curriculum of traditional disciplines to unlock creative potential. Moreover, our study identified differential mediating effects of digital competencies on the relationship between AI use and student creativity. Digital competencies completely mediated the relationship between AL and student creativity, while partially mediating the relationship between AE and student creativity. This nuanced finding differs from previous studies that often assume a uniform mediating effect across different types of AI use (56). Our results suggest that the associations between AI use and creativity via digital competencies vary depending on whether AI is used for learning or entertainment. This highlights the need for a more nuanced understanding of AI’s role in education, recognizing that different AI applications may have distinct pathways to influencing student outcomes.

Finally, the study uncovered the moderating role of attitudes toward AI in the relationship between AI use and digital competencies within the context of TCM education. Specifically, positive attitudes toward AI enhanced the association between AE and digital competencies, while negative attitudes dampened the association between AL and digital competencies. This finding contrasts with some existing research that assumes a consistently positive influence of positive attitudes on AI adoption (5, 57, 58). Our results suggest that the relationship between AI use and digital competencies is more complex and contingent on the specific type of AI application and the user’s attitude. This highlights the importance of addressing students’ attitudes toward AI in educational interventions to maximize the benefits of AI tools.

This study makes several important theoretical contributions by extending Social Cognitive Theory (SCT) and enriching the literature on AI in education and student creativity. First, this study extends SCT into AI-supported educational contexts by conceptualizing AI use as a differentiated environmental stimulus rather than a uniform technological factor. While SCT traditionally treats the environment as a general external influence, our findings show that different forms of AI use (learning-oriented versus entertainment-oriented) function in fundamentally different ways. This distinction demonstrates that the “environment” in SCT should not be viewed as homogeneous in technology-rich contexts, but instead as a set of heterogeneous digital affordances with distinct cognitive and motivational implications. By doing so, this study refines SCT by introducing a more granular understanding of technological environments.

Second, this study identifies digital competencies as a core mediating mechanism that transforms AI use into creative outcomes. Prior SCT-based research often emphasizes self-efficacy, motivation, or behavioral engagement as central cognitive mediators (5, 59). Our findings complement and extend this tradition by showing that in AI-driven learning environments, digital competencies represent a fundamental cognitive capability that enables individuals to interpret, evaluate, and productively utilize AI outputs. The complete mediation between AI used for learning and student creativity and the partial mediation between AI used for entertainment and creativity indicate that creativity does not emerge directly from AI use itself, but from students’ ability to cognitively appropriate AI through digital skills. This insight advances SCT by specifying a technology-centered cognitive pathway that has been largely overlooked in prior research.

Third, this study advances SCT by revealing the dual moderating role of attitudes toward AI. Existing studies generally assume that positive attitudes uniformly strengthen technology use outcomes (2, 32). However, our findings demonstrate a more nuanced pattern: positive attitudes enhance the effect of entertainment-oriented AI use on digital competencies, while weakening the effect of learning-oriented AI use. This dual moderation suggests that attitudes toward AI do not simply amplify environmental effects, but reshape how different technological environments are cognitively processed. This challenges the linear assumption embedded in much of the SCT-based technology literature and highlights the conditional and context-sensitive nature of personal beliefs in AI-enabled learning.

To better align practical implications with the empirical findings, this section is organized into three subsections corresponding to educators, curriculum designers, and policymakers. Each implication is directly grounded in the key results of this study, including the mediating role of digital competencies, the dual moderating role of attitudes toward AI, and the IPMA results that identify digital competencies and attitudes as the most influential factors for enhancing student creativity.

This study provides insights into the association between AI and digital competencies and student creativity among TCM students, but it also has limitations. First, the generalizability of the findings should be interpreted with caution. Since the sample consists solely of undergraduate students majoring in Traditional Chinese Medicine at a single university in one region of China, the results cannot be directly generalized to other academic disciplines, educational levels, institutional types, or cultural contexts. Future studies should replicate this model across different majors, universities, and countries to examine its broader applicability. Second, the study relied on self-reported questionnaires, which may not fully capture the complexity of constructs like digital competencies and creativity. Incorporating qualitative interviews, performance-based assessments, or longitudinal observations could provide a more comprehensive understanding of these relationships. Third, a key limitation of this study is its cross-sectional design, which does not allow for causal inference. Accordingly, all findings should be interpreted as correlational rather than causal. Future research employing longitudinal or experimental designs is needed to establish causal relationships. Another limitation is that the study was not pre-registered, which may reduce the transparency and reproducibility of the research process. Future studies should adopt pre-registration practices, particularly when employing complex moderated-mediation models, to strengthen research rigor. Nevertheless, no a priori power analysis was conducted, which is a limitation. Future studies should incorporate pre-registration and a priori power analysis at the design stage to further strengthen methodological rigor. Additionally, the study focused on attitudes toward AI as a moderator but did not explore other potential moderating factors, such as individual differences, educational environment, or sociocultural influences (60). Future research should consider these factors to provide a more nuanced understanding of how AI impacts educational outcomes. Finally, AI may overthink simple problems into complex models, causing students to rely on technology rather than think independently (e.g., in TCM diagnosis, students directly apply AI recommendations and skip the core thinking training of syndrome differentiation and treatment). Future studies should address these aspects to better understand the ecological dynamics of AI integration in education. Future research should further explore how frontier technologies can be integrated to modernize TCM education. For example, future studies could investigate how generative AI may be leveraged to design intelligent diagnostic simulations that foster adaptive learning pathways and stimulate creative problem solving. Similarly, VR could be examined for its potential to build immersive anatomy or herbal medicine laboratories that enhance experiential and practice-oriented learning. In addition, blockchain warrants exploration for its role in securing the management of clinical training data and ensuring transparent credentialing of student competencies. Future research should integrate the “context–structure–mechanism–outcome” evolutionary paradigm of complex systems (61) to explore how emerging technologies (e.g., AI, VR, and blockchain) and human–AI collaboration contexts dynamically reshape cognitive structures and mechanisms, thereby driving nonlinear transformations in creativity and the modernization of TCM education.