This theoretical paper employs an integrative literature review to examine the psychological underpinnings of human attachment to artificial intelligence systems. Grounded in attachment theory and social cognition research, this study investigates how AI-specific characteristics—including predictable responsiveness, behavioral contingency, and absence of interpersonal judgment—activate fundamental human attachment mechanisms.
Empirical evidence from human-computer interaction studies with relational agents such as PARO and Woebot demonstrates that individuals form genuine emotional bonds with responsive AI systems, perceiving them as meaningful relational partners. The research reveals a crucial paradox: while AI relationships provide immediate emotional support and psychological safety—particularly beneficial for vulnerable populations—they simultaneously pose significant mental health risks, including emotional dependency and potential deterioration of essential interpersonal competencies such as empathy, conflict resolution, and social resilience.
The paper addresses critical ethical considerations regarding emotional manipulation and informed consent, while proposing practical guidelines emphasizing digital emotional literacy as a core competency for navigating human-AI relationships. Our conceptual framework advocates for a synergistic design paradigm that balances technological innovation with the preservation of fundamental human relational capacities. Future research directions include longitudinal developmental studies, cross-cultural comparative analyses, and clinical applications in mental health support systems.