The Algorithmic Sanctuary: Social-Interaction Burnout, Loneliness, and Emotional Attachment to AI Companions Among Young Adults in Indonesia

Abstract

Post-pandemic digital life has produced a new affective phenomenon: large-language-model “AI companions” that invite two-way parasocial relationships. Counter-intuitively, the most intense users in urban Indonesia appear to be not the physically isolated but socially active young adults experiencing interaction burnout, who treat AI as a refuge from the judgment costs of a collectivist culture. Yet no integrated model has tested why relational strain translates into machine attachment in a Global-South setting. Drawing on Parasocial Interaction Theory, the Computers-Are-Social-Actors paradigm, and emotional-labor theory, this cross-sectional survey of 1,200 young adults recruited through a public organization in Palembang, South Sumatera, Indonesia, measured social-interaction burnout, subjective loneliness, collectivist judgment apprehension, AI parasocial interaction, and emotional attachment to AI using validated Likert scales analysed with reliability, correlation, multiple regression, bootstrap mediation, and moderation. All scales were reliable (α = .84–.91; KMO = 0.96). The model explained 48% of variance in attachment (adjusted R² = .486, F = 284, p < .001). Parasocial interaction (β = .318), loneliness (β = .254), and burnout (β = .238) were the strongest predictors (all p < .001). Parasocial interaction partially mediated the burnout–attachment path (indirect = .160, 95% CI [.132, .191]), and judgment apprehension moderated the loneliness–attachment link (β = .127, p < .001). Findings introduce the “Algorithmic Sanctuary” account of AI companionship and inform digital-wellbeing policy in collectivist societies.