Psychological mechanisms of perceived technological dependence on generative artificial intelligence tools: perceived knowledge, trust, anxiety, and self-efficacy

The rapid expansion of generative artificial intelligence (GenAI) and AI-generated content (AIGC) tools within higher education has fundamentally shifted how students approach creative and academic assignments. Recent survey metrics underline this profound transition, indicating that university-level adoption has evolved from sporadic experimentation into an entrenched daily practice. While global estimates from the Student Generative AI Survey and the Global AI Student Survey demonstrate that upwards of 86% to 92% of university students actively utilize these platforms in their studies, the psychological and cognitive impacts of this integration remain under-investigated—particularly within design-related disciplines. A comprehensive cross-sectional study recently published in Frontiers in Psychology sheds light on these dynamics, examining how undergraduate design students navigate the intersection of perceived knowledge, technological trust, anxiety, and eventual functional reliance.

Methodological Framework and Participant Demographics

To evaluate the complex psychological and behavioral variables tied to AIGC adoption, researchers conducted a cross-sectional survey involving 526 undergraduate students enrolled in design-related disciplines across China. The cohort encompassed a diverse array of creative majors, including visual communication, graphic and brand design, industrial and product design, interaction and UI/UX design, and environmental, interior, and landscape design. Data collection took place between June and August 2025 via course-related digital networks and student communication channels.

Out of 537 initially submitted questionnaires, 11 were excluded because respondents reported no prior exposure to AIGC tools in their creative workflows, yielding a final analytical sample of 526 participants. The demographic breakdown revealed a balanced representation across gender, with 270 male students (51.3%) and 256 female students (48.7%). The distribution across academic years spanned first-year undergraduates (30.0%), second-year students (28.5%), third-year students (22.6%), and fourth-year students (18.8%). Furthermore, participant self-assessments regarding AIGC familiarity highlighted a digitally literate population: 38.8% reported occasional use, 34.6% identified as frequent users, and 26.6% classified themselves as highly proficient.

Evaluating the Extended Technology Acceptance Model

The study utilized partial least squares structural equation modeling (PLS-SEM) via SmartPLS 4.0 to analyze a total of 10 reflective constructs measured through structured five-point Likert scales. Moving beyond the traditional Technology Acceptance Model (TAM)—which typically emphasizes perceived usefulness and perceived ease of use—the researchers integrated variables such as perceived reliability, perceived intelligence, perceived AIGC knowledge, AIGC-related anxiety, self-efficacy, trust, usage intention, and perceived technological dependence.

The statistical evaluation demonstrated strong internal consistency and convergent validity, with Cronbach’s alpha values ranging from 0.853 to 0.914, composite reliability metrics between 0.901 and 0.939, and average variance extracted (AVE) values exceeding the 0.50 threshold across all measures. Out of twelve hypothesized pathways, eleven were empirically supported by the bootstrap analyses involving 5,000 resamples.

Key Findings on Perceived Knowledge and Psychological Responses

The investigation revealed that technological perceptions heavily influence students’ subjective understanding of generative tools. Specifically, perceived reliability (beta = 0.371, p < 0.001), perceived usefulness (beta = 0.272, p < 0.001), and perceived intelligence (beta = 0.131, p = 0.005) were found to be positively associated with perceived AIGC knowledge. Interestingly, perceived ease of use did not demonstrate a statistically significant direct association with perceived knowledge (beta = 0.061, p = 0.187), suggesting that an accessible user interface alone does not necessarily translate to a deeper cognitive understanding of a tool’s underlying capabilities and limitations.

Perceived AIGC knowledge functioned as a pivotal cognitive anchor within the model. It exhibited strong positive associations with trust (beta = 0.666, p < 0.001), self-efficacy (beta = 0.460, p < 0.001), and usage intention (beta = 0.296, p < 0.001), while displaying a significant negative association with AIGC-related anxiety (beta = -0.487, p < 0.001).

Mediation analyses further elucidated these psychological pathways. Trust, anxiety, and self-efficacy each acted as significant statistical mediators in the relationship between perceived AIGC knowledge and usage intention. Among these, the indirect effect through trust was notably robust (beta = 0.194, p < 0.001), highlighting that confidence in AI-generated suggestions heavily encourages students to integrate these resources into their creative ideation and iterative refinement processes.

From Usage Intention to Perceived Technological Dependence

A central contribution of this research is its analytical shift from initial technology adoption to post-adoption functional reliance. The study evaluated whether strong usage intentions are mirrored by a perceived technological dependence—defined operationally as the self-reported incorporation of AIGC into routine design tasks, a preference for automated workflows over traditional methods, and the anticipated loss of efficiency if working without AI assistance.

The structural model confirmed a substantial positive association between usage intention and perceived technological dependence (beta = 0.644, p < 0.001), with a large effect size (f-squared = 0.708). The overall model explained nearly 59% of the variance in usage intention (R-squared = 0.591) and over 41% of the variance in perceived technological dependence (R-squared = 0.415).

Broader Implications for Design Education and Future Research

While the findings provide robust empirical insights into human-AI interaction within academic settings, the authors emphasize several methodological limitations. Because the data were gathered via a cross-sectional survey at a single point in time, the structural paths represent theoretical expectations and statistical associations rather than proven temporal sequences or causal relationships. Furthermore, reliance on self-reported measures leaves open the possibility that shared response tendencies or general academic confidence may have influenced ratings, particularly regarding perceived knowledge and trust.

Despite these caveats, the implications for design education are substantial. As generative AI tools become permanent fixtures in creative industries, educators face the dual challenge of fostering technical proficiency while safeguarding students’ independent critical thinking, aesthetic judgment, and creative authorship. The research suggests that rather than restricting AI access, curricula should incorporate critical evaluation exercises—such as comparing multiple prompt iterations, analyzing output inconsistencies, and documenting design rationales.

Future investigations will benefit from longitudinal designs, behavioral log tracking, and objective assessments of AI literacy to determine how perceived technological dependence translates into actual creative practice over the course of an academic career. By understanding the intricate balance between trust, anxiety, and knowledge, academic institutions can better prepare students to harness generative tools responsibly without sacrificing their core creative autonomy.