What makes you attached to social companion AI? A two-stage exploratory mixed-method study

Abstract

Social companion AI, as a generative AI application with empathy and emotional support functions, is gradually becoming a new object of human emotional attachment. This study explores the formation framework of human-SCAI attachment through a two-stage mixed-method approach. In Study 1, using reviews of two films themed around human-AI intimate relationships (Her and M3GAN) as analysis data, semantic network analysis and topic modeling were conducted to identify seven potential concepts and propose the “Interpersonal & Human-AI Relationship Attitudes → Value Evaluation → Attachment Manifestation” framework for AI attachment formation. The study found that perception of AI agent personification and interpersonal dysfunction are driving factors for intimate human-SCAI interactions. Based on social exchange theory, it was discovered that the cost-benefit exchange mechanism in the interaction process influences the formation and varied manifestations of AI attachment. Building on the conclusions of Study 1, a research model was proposed and Study 2 was conducted, involving a survey of long-term users of AI companions and structural model testing using SmartPLS. This study provides insights into understanding human-AI intimate relationships and the mechanisms of AI attachment formation in the GenAI era, while also offering insights and recommendations regarding the potential risks of human-SCAI intimate relationships.