Are students dependent on AI in writing courses? Analyzing factors influencing dependence on generative AI through the I-PACE model.

Background: The rapid penetration of artificial intelligence (AI) in the field of education has brought about an improvement in learning efficiency, but it has also triggered serious concerns. The harms of students' excessive dependence on AI to complete learning tasks, the weakening of critical thinking, and the degradation of self-learning ability are gradually emerging. This forces researchers to deeply understand the intrinsic mechanism of users' psychological dependence on AI. The theoretical framework of problematic behavior has been fully developed and can effectively explain and predict users' emotional connection, cognition, trust, and degree of dependence when using AI. Method: This study adopts a mixed-method design. For the quantitative phase, survey responses were collected from two universities in China. This component aims to validate the influence of AI dependence in the context of academic writing. For the qualitative phase, in-depth interviews were conducted with eight respondents. Quantitative data were analyzed using structural equation modeling, while qualitative data were processed through thematic analysis. This study explores Chinese college students' motivations for excessive AI use and their perceptions of AI dependence. Results: Quantitative research indicates that academic stress, AI literacy, and perceived trust have a significant impact on AI dependence, while perceived usefulness can mediate the relationship between academic self-efficacy and academic stress. Perceived trust plays a mediating role between social influence and AI dependence. Social influence predicts AI dependence behavior through perceived trust. Qualitative research indicates that although students are aware of the risks of AI, factors such as academic stress, peer influence, the efficiency of tools, and policy ambiguity are significant incentives for the strategic use of AI; there exists a strong psychological connection with AI dependence. Discussion: The study extends the I-PACE model to generative AI academic writing contexts, revealing the specific social impact and the trust transmission mechanism. The study also emphasized that academic stress and AI literacy are important influencing factors. For educational practice, both protective factors and risk factors should be taken into account: efforts should be made to enhance students' AI literacy, while also paying attention to alleviating excessive academic stress on students, in order to reduce the excessive dependence of college students on generative AI. This study emphasizes that responsible use of AI should be promoted in the process of introducing AI into teaching.