The relationship between college students’ AI technology dependence and learning burnout: a chain mediation analysis of technology acceptance and AI self-efficacy

With the pervasive integration of generative artificial intelligence (GenAI) into higher education, a critical concern has emerged: the non-adaptive use of AI learning tools by college students, which poses a significant risk to their learning psychology and academic adaptation. A recent study, published in Frontiers in Psychology, has delved into this complex relationship, revealing a significant chain mediation mechanism linking AI technology dependence to learning burnout through technology acceptance and AI self-efficacy.

Key Findings Emerge on AI Dependence and Student Burnout

The research, conducted by Gu, He, and Zhang, highlights that AI technology dependence is not merely an issue of excessive usage but a psychological phenomenon that can directly contribute to learning burnout. The study’s findings indicate a significant positive correlation between AI technology dependence and learning burnout. Furthermore, the research meticulously unpacks the intricate pathways through which this dependence impacts students. Technology acceptance, representing a student’s rational evaluation and positive attitude towards AI tools, and AI self-efficacy, reflecting a student’s confidence in their ability to use AI effectively and independently, both emerge as crucial mediating factors.

The study establishes that technology acceptance plays a significant independent mediating role. When students develop a non-adaptive dependence on AI, their rational acceptance and perceived usefulness of these tools tend to decline. This diminished acceptance, in turn, exacerbates feelings of burnout. Similarly, AI self-efficacy also acts as a significant independent mediator. As students become overly reliant on AI, their opportunities for independent problem-solving and skill development diminish, leading to a reduction in their confidence and perceived ability to use AI effectively, thus contributing to burnout.

Crucially, the research uncovers a significant chain mediation path, suggesting a sequential process where AI technology dependence first erodes technology acceptance, which then further weakens AI self-efficacy. This dual depletion of psychological resources ultimately amplifies learning burnout. This finding provides a more nuanced understanding of how external technology behaviors translate into internal psychological states and academic outcomes.

Background: The AI Revolution in Higher Education

The rapid evolution and widespread adoption of generative AI technologies like ChatGPT, Gemini, and sophisticated academic auxiliary tools have fundamentally reshaped the landscape of higher education. These tools offer unprecedented capabilities in natural language processing, information synthesis, real-time feedback, and personalized content generation, making them indispensable assets for students navigating complex academic tasks. From classroom learning and assignments to in-depth literature research, AI has become deeply embedded in the daily routines of college students.

However, this technological empowerment has also given rise to new challenges. Non-adaptive usage patterns, including overuse, cognitive offloading (delegating cognitive tasks to AI), passive task completion, and excessive emotional attachment to AI tools, are becoming increasingly prevalent. This phenomenon, termed "AI technology dependence," is characterized not only by high frequency of use but also by a suite of psychological changes, such as a decline in independent thinking, reduced metacognitive engagement, and a diminished sense of academic accomplishment. These changes can significantly impact students’ motivation, emotional well-being, and overall academic adaptation.

Learning burnout, a state of emotional, physical, and mental exhaustion caused by prolonged academic stress, repetitive learning experiences, and a low sense of accomplishment, is a well-established predictor of negative academic outcomes, including withdrawal, procrastination, mental health issues, and even academic interruption. While existing research has largely established a positive correlation between excessive technology use and digital addiction with learning burnout, the specific nuances of AI learning have remained less explored. A key deficiency identified by the researchers is the failure to differentiate between empowering and non-adaptive AI use, and a limited understanding of the complete psychological pathway linking AI dependence to burnout.

Theoretical Framework: TAM and SET Intersect

To address these gaps, the study draws upon two robust theoretical frameworks: the Technology Acceptance Model (TAM) and Self-Efficacy Theory (SET). TAM posits that an individual’s acceptance of a technology is primarily driven by perceived usefulness and perceived ease of use, which influence their attitude, intention, and actual use. Rational and positive technology acceptance is seen as a facilitator of effective and sustainable technology use, whereas passive or irrational acceptance can lead to negative consequences. In the context of AI learning, technology acceptance reflects a student’s stable, healthy, and rational cognitive evaluation of AI tools, serving as a crucial cognitive bridge between external technology engagement and internal psychological states.

SET, on the other hand, emphasizes the role of an individual’s beliefs about their capabilities in a specific domain. AI self-efficacy, as a domain-specific belief, refers to a student’s confidence in their ability to effectively, reasonably, and independently utilize AI tools to achieve learning goals. Students with high AI self-efficacy are more adept at regulating their behavior, managing difficulties, and experiencing less anxiety. Conversely, those with low AI self-efficacy may become overly reliant on external tools, avoid challenging tasks, and experience greater emotional distress when encountering obstacles.

The researchers argue that TAM and SET can be integrated into a hierarchical framework. Technology acceptance, as conceptualized in TAM, represents a more external cognitive appraisal of the technology. AI self-efficacy, rooted in SET, reflects a deeper, competence-based belief that emerges from sustained interaction and successful experiences. The proposed chain mediation model posits that this sequential order is critical: students’ initial cognitive evaluations of AI tools (technology acceptance) lay the groundwork for the development of their confidence in using AI (AI self-efficacy).

Methodology: A Data-Driven Investigation

The study employed a cross-sectional online questionnaire survey administered to 466 undergraduate students from five ordinary universities in China between September and October 2025. Participants, who had used at least one AI learning tool, provided informed consent. The data collection involved validated scales to measure:

• AI Technology Dependence: Assessed using the conversational AI dependence scale, measuring uncontrollability, withdrawal symptoms, mood modification, and negative impact. The scale yielded a Cronbach’s alpha of 0.735.
• Technology Acceptance: Measured by adapted dimensions from the accept artificial intelligence-driven technology scale, focusing on perceived usefulness and perceived ease of use. The scale achieved a Cronbach’s alpha of 0.905.
• AI Self-Efficacy: Evaluated using the AI self-efficacy scale, covering assistance, anthropomorphic interaction, comfort with AI, and technological skills. This scale demonstrated a high Cronbach’s alpha of 0.977.
• Learning Burnout: Assessed using the widely cited College Students’ Learning Burnout Scale, comprising low mood, inappropriate behavior, and low sense of accomplishment. The scale reported a Cronbach’s alpha of 0.914.

Data analysis involved descriptive statistics, Pearson correlation analysis, and structural equation modeling (SEM) using SPSS 26.0 and AMOS 27.0. The bias-corrected bootstrap method with 5,000 resamples was utilized to test mediation and chain mediation effects, with 95% confidence intervals used to determine significance.

Results: Unpacking the Mediating Pathways

The statistical analyses confirmed the hypothesized relationships. A common method bias test using Harman’s single-factor analysis indicated no serious bias, with the first common factor explaining less than 40% of the variance. Descriptive statistics and correlation analysis revealed that AI technology dependence was significantly negatively correlated with technology acceptance (r = -0.362, p < 0.001) and AI self-efficacy (r = -0.334, p < 0.001), and significantly positively correlated with learning burnout (r = 0.241, p < 0.001). Technology acceptance and AI self-efficacy were positively correlated with each other (r = 0.713, p < 0.001) and negatively correlated with learning burnout (r = -0.597 and r = -0.622, respectively, both p < 0.001).

The mediation effect test, utilizing the bootstrap method, yielded significant results for all proposed indirect paths:

• Independent Mediation Path 1 (AI Technology Dependence → Technology Acceptance → Learning Burnout): This path showed a significant mediating effect (effect size = 0.171, 95% CI [0.099, 0.247]).
• Independent Mediation Path 2 (AI Technology Dependence → AI Self-Efficacy → Learning Burnout): This path also demonstrated a significant mediating effect (effect size = 0.053, 95% CI [0.014, 0.096]).
• Chain Mediation Path (AI Technology Dependence → Technology Acceptance → AI Self-Efficacy → Learning Burnout): This core finding revealed a significant chain mediating effect (effect size = 0.149, 95% CI [0.096, 0.205]).

The total indirect effect accounted for a substantial portion of the total effect (0.373 out of 0.383), indicating that the relationship between AI technology dependence and learning burnout is primarily driven by these indirect pathways. The direct effect of AI technology dependence on learning burnout was not significant after accounting for the mediators (effect size = 0.010, 95% CI [-0.102, 0.122]), suggesting that the influence is almost entirely indirect. The structural equation model exhibited excellent fit indices (e.g., GFI = 0.927, CFI = 0.946, RMSEA = 0.054), supporting the proposed chain mediation model.

Discussion: Towards a Healthier Human-Computer Collaborative Learning Model

The study’s findings underscore the complex psychological mechanisms underlying the relationship between AI technology dependence and learning burnout among college students. The direct positive association between AI dependence and burnout aligns with existing theories like the Conservation of Resources Theory, suggesting that excessive reliance on AI depletes students’ cognitive resources and leads to exhaustion.

The independent mediating roles of technology acceptance and AI self-efficacy highlight two critical intervention points. Improving students’ rational acceptance of AI—viewing it as a tool rather than a crutch—and bolstering their confidence in using AI effectively are crucial for mitigating burnout. The chain mediation finding is particularly significant, illustrating a step-by-step degradation of psychological resources. AI dependence first diminishes students’ ability to rationally assess and accept AI tools, which, in turn, undermines their confidence in their own AI-related skills. This dual erosion of cognitive and efficacy-based resources culminates in increased learning burnout.

Practical Implications for Higher Education Institutions

The research offers concrete, actionable insights for universities aiming to foster adaptive AI utilization and mitigate learning burnout:

1. Strengthen AI Literacy Education: Institutions should systematically embed AI literacy modules into curricula. These modules should cover evidence-based AI tool operation, the critical distinction between AI assistance and independent learning, methods for critically evaluating AI-generated content, and academic integrity guidelines. Educators can design scaffolded tasks that encourage AI use for preliminary research while demanding independent critical thinking and verification. Workshops and peer-sharing sessions can further educate students on the risks of over-reliance.

2. Enhance Perceived Usefulness and Ease of Use: To improve rational technology acceptance, universities should optimize the deployment of user-friendly AI learning platforms and provide structured training sessions to reduce operational barriers. In pedagogical practices, instructors should explicitly emphasize how AI can enhance efficiency without compromising deep learning, thereby fostering balanced perceptions of AI’s affordances. This can reduce compulsive usage and emotional over-reliance, acting as a buffer against burnout.

3. Protect and Enhance AI Self-Efficacy: To interrupt the burnout transmission chain, educators should focus on process-oriented assessments, provide incremental success experiences, and implement metacognitive scaffolding. Formative evaluations that reward effort and progress, alongside tiered learning tasks designed for individual proficiency, can build confidence. Metacognitive reflection protocols can help students monitor their AI usage, identify dependency patterns, and proactively adjust their behavior, thus sustaining motivational resources and reducing vulnerability to burnout.

Limitations and Future Research Directions

The study acknowledges several limitations. The cross-sectional design prevents definitive causal inferences, as reverse causation—where burnout might lead to increased AI dependence—cannot be entirely ruled out. The sample, drawn through convenience sampling from specific universities in China, with an unbalanced gender ratio, may limit the generalizability of the findings to broader student populations across different educational levels, disciplines, and regions.

Future research should adopt longitudinal designs to track the dynamic changes in these variables and establish clearer causal relationships. Expanding the sample scope through stratified sampling and incorporating moderating variables such as discipline type, learning motivation, and AI literacy would provide a more comprehensive understanding of the proposed model. Combining qualitative methods, such as in-depth interviews, with quantitative data can offer richer insights into the lived experiences of students navigating AI in their academic journeys. Furthermore, exploring alternative theoretical models and comparing them with the proposed chain mediation structure would strengthen the robustness of the findings.

In conclusion, this study provides a vital empirical foundation for understanding the psychological ramifications of AI technology dependence in higher education. By illuminating the mediating roles of technology acceptance and AI self-efficacy, it offers a clear roadmap for universities to cultivate a healthier, more sustainable human-computer collaborative learning environment.