Kai Qiao, X Rebecca Sheng, Zhenyu Li
The rapid advancement of artificial intelligence (AI), especially large language models, has made human-computer interaction more dynamic and personalized, thereby accelerating the development of AI-driven virtual humans (AI-Avatars). Based on the Computers Are Social Actors (CASA) paradigm, this study employs text mining and experimental methods to compare user interactions in live streaming contexts with Human-Avatars and AI-Avatars. The findings indicate that curiosity about “AI” acts as a primary factor in user engagement. In AI-Avatar contexts, users’ danmaku are more diverse and information-seeking, whereas in Human-Avatar contexts, they are more uniform and socially oriented. Despite these differences, users show similar interaction patterns across both types, highlighting the functional viability of AI-Avatars. However, limited social intelligence prevents AI-Avatars from building interaction ritual chains and emotional resonance, limiting long-term user engagement. These findings enrich our understanding of the psychological and social dynamics underlying human interactions with AI agents in mediated environments.