Individual dependency behavior in multimodal AI-assisted knowledge acquisition: influencing factors and explainable prediction

Abstract

This study aims to identify the key factors shaping individual dependency behavior in multimodal artificial intelligence (AI)-assisted knowledge acquisition and develops an interpretable prediction model to inform strategies for optimizing knowledge internalization. A sequential mixed-methods design was employed. Open, axial and selective coding of 32 semi-structured interviews yielded five factors: task execution, information processing, individual cognition, technology experience and knowledge management. Saturation Assessment Curve (SAC) confirmed theoretical saturation. Theoretical coverage reached over 97% by the 17th interview. An expert panel of 20 domain specialists reviewed the framework and confirmed satisfactory content validity indices (S-CVI/Ave > 0.80). Subsequently, 2,002 valid survey responses were analyzed using machine learning classification. A support vector machine (SVM) was selected for its superior accuracy and F1-score, and the Shapley Additive Explanations (SHAP) algorithm was applied to quantify feature contributions. Task execution exerts the strongest influence on dependency behavior, reflecting individuals’ reliance on multimodal AI under task pressure and goal-driven conditions. Information processing and individual cognition positively contribute as secondary drivers, while technology experience and knowledge management contribute comparatively weakly. These results suggest that habitual, pressure-driven responses progressively outweigh careful, quality-oriented considerations in AI-assisted knowledge acquisition. This study extends the theoretical understanding of dependency behavior in AI-mediated knowledge acquisition. It also offers empirical evidence to guide human–AI collaboration design, AI governance frameworks and sustainable knowledge management practices.