A theoretical paper published in Frontiers in Education argues that the relationship between AI learning empowerment and college students' psychological well-being may follow an inverted-U pattern — initially beneficial but eventually detrimental — rather than the linear positive effects assumed by much existing research.
The analysis, authored by S Y Shufang Yang of Guangdong Engineering Polytechnic's School of Finance and Commerce, does not report new empirical data. Instead, it critically examines existing theories and proposes a testable framework to explain when and why AI tools shift from enhancing to undermining student well-being.
According to the paper, the dominant paradigm in AI education research has been linear: more AI use yields better learning outcomes and more positive psychological experiences. But Yang argues this assumption sits uneasily with broader technology research, which has pointed toward curvilinear patterns.
The paper cites Dienlin and Johannes (2020), who noted that "both low and excessive use are related to decreased well-being, whereas moderate use is related to increased well-being" — though those authors also cautioned that empirical evidence for this curvilinear position remains mixed.
Yang's analysis builds on the Technology and Consumer Well-being Paradox Model developed by McLean et al. (2025), which broke through a simple "good or bad" dichotomy by acknowledging that technology can both enhance and diminish well-being over time. However, the paper identifies three unresolved tensions in that model: its origin in consumer marketing rather than education, its failure to specify a nonlinear functional form, and its underdeveloped explanation of the psychological mechanisms that turn beneficial use harmful.
Distinguishing use, empowerment, and dependency
A central contribution of the paper is its explicit separation of three constructs often conflated in the AI-education literature. AI use refers simply to the behavioural frequency of interacting with AI tools — descriptively neutral. AI empowerment is the subjective sense of enhanced capability and goal attainment that students derive from AI-assisted learning — a positive experiential state. AI dependency, by contrast, is a cognitive-motivational construct denoting the progressive delegation of core cognitive functions — reasoning, evaluation, decision-making — to AI systems, such that the learner's sense of competence becomes contingent on AI availability.
Yang proposes that these constructs relate sequentially: AI use can generate AI empowerment, and sustained empowerment without critical engagement can, under certain appraisal conditions, crystallise into AI dependency. The inverted-U proposed in the paper concerns AI empowerment, not raw usage frequency. Dependency, not usage per se, serves as the proximal antecedent of digital burnout.
Digital burnout as the mediating mechanism
The paper identifies digital burnout as a key mediator in the shift from beneficial to harmful effects. It draws on the Digital Burnout Scale developed by Erten and Özdemir (2020), a 24-item instrument comprising three sub-dimensions: digital aging, digital deprivation, and emotional exhaustion. However, Yang notes that these dimensions originate primarily from technology burnout research in work contexts, and their validity for learning contexts — particularly for understanding the psychological costs of AI-assisted learning — has not been sufficiently verified. The "digital aging" dimension, in particular, may have limited relevance to traditional-age college students, whose digital exposure is normative rather than occupationally imposed.
Evidence from recent studies
The paper points to recent empirical work that complicates the linear picture. Yu and Luo (2026), studying 902 Chinese college students, discovered a negative link between AI anxiety and mental well-being, with the connection between AI interaction experience and anxiety forming a U-shaped curve: anxiety decreased over the initial three years of AI usage but increased among veteran users. This curve indicates that — from operational anxiety in early stages to deeper concerns about data ethics, professional displacement, and autonomy erosion among experienced users.
Elsayed et al. (2026), studying 521 university students, conceptualised AI dependency as a cognitive-motivational construct distinct from usage frequency. Their network analysis revealed non-linear, conditional associations between AI dependency and psychological outcomes, with anxiety occupying a bridging position linking cognitive reliance to environmental pressures.
What this means for students
For university students who increasingly rely on generative AI tools for writing, coding, and collaborative learning, the paper raises a practical question: at what point does AI use shift from being a productivity enhancer to a cognitive crutch that undermines genuine learning and well-being?
The paper specifies academic self-efficacy as a potential boundary condition — students with higher confidence in their own learning abilities may be less susceptible to the negative effects of AI dependency. But the framework remains theoretical, and Yang calls for empirical testing of the proposed inverted-U model.
Limitations and open questions
The paper acknowledges several limitations. The evidence for inverted-U patterns in technology research derives primarily from social media use, which differs fundamentally from AI learning tools in cognitive engagement patterns — the former is more consumptive, the latter more productive. Whether the inflection point of AI learning empowerment's inverted-U would align with social media, or whether cognitive outsourcing triggers psychological costs through entirely different mechanisms, remains an open question.
Additionally, the paper notes that the general effects of digital technology on well-being are "on the negative end of the spectrum but very small," and that evidence for the inverted-U pattern is mixed. The analysis is a theoretical contribution intended to advance mechanism specification, not a settled empirical finding.
Yang's paper was published on 20 November 2026 in Frontiers in Education, in the Psychology in Education section.