Using Machine-Learning and Network Analysis to Investigate the Risk Factors of AI Dependence: The Crucial Role of Escape and Social Motivation

Yufan Chen, Xiaoyin Miao, Zeyang Yang

People have become accustomed to studying or working with the guidance of artificial intelligence (AI) in recent years. Studies have begun investigating the risk factors of AI dependence, though most have used hypothesis-testing methods. The present study aimed to investigate predictors of AI dependence using machine-learning and network analysis, which are data-driven approaches. The included risk factors were Big Five personality traits, self-efficacy, depression, social anxiety, adverse childhood experiences, and AI use motivation, selected based on theories and empirical studies. Participants consisted of 1258 university students (942 females and 316 males) with a mean age of 22.11 years (SD = 2.69). Four machine-learning algorithms were tested, including Elastic Net, Random Forest, XGBoost, and LightGBM. Machine-learning results indicate that escape and social motivation for AI use, along with social anxiety, were the main predictors of AI dependence. Network analysis results show that escape and social motivation were the most central nodes, with the highest Expected Influence (EI) indices. This study indicates that when addressing mental health problems related to AI dependence, it is more effective to focus on emotional isolation and social interaction challenges rather than simply cutting down on AI use.