Abstract
Individuals resort to human-AI relationships to fulfill psychological, social, and empathetic needs. Nevertheless, dilemmas arise from these interdependent relationships which, although destined to enhance users' well-being, can also be detrimental to them. By adopting a mixed-method approach (netnography, n = 1131 posts; survey, n = 610), through the lens of Interdependence Theory, this study investigates characteristics of interdependent human-AI relationships and how users' vulnerability emerges within those relationships, ultimately affecting their well-being. Results suggest that users experience dilemmas within relationships with AI companions. Specifically, AI companion's warmth and availability create a “positive” dependence, softening vulnerability and positively influencing well-being. However, the AI emotional abilities and third-party company control increase perceptions of relational power imbalance, thus positively influencing vulnerability and negatively influencing well-being. We provide practical applications centered on app developers integrating protective app features to increase consumers' self-confidence, coping strategies, self-worth and benevolence, as well as adopting a collaborative model with consumers.