The Psychology of Generative AI Dependence: Reward, PsychologicalNeeds, and the Emerging Architecture of Problematic Human–AIEngagement

Abstract

The rapid diffusion of Generative Artificial Intelligence (GenAI) has transformed the relationship between humans and digital technology. Unlike digital systems that primarily delivered information, mediated interpersonal communication, or provided entertainment, generative AI can respond conversationally, personalise interactions, produce immediate solutions, simulate empathy, reduce cognitive effort, and remain continuously available. These characteristics raise an emerging question: can repeated engagement with generative AI develop into psychologically dependent or addiction-like patterns, and, if so, through what mechanisms? This paper employs an integrative critical literature review and conceptual synthesis to examine problematic generative-AI use
through the combined perspectives of behavioural addiction theory, reward and reinforcement processes, the Interaction of Person–Affect–Cognition– Execution (I-PACE) model, self-determination theory, compensatory technology use, social anxiety, loneliness, anthropomorphism, cognitive offloading, and human psychological needs. The current study identifies a major gap in the current literature: emerging research frequently examines AI dependence as either a behavioural outcome, an educational problem, or a form of social attachment, but insufficiently integrates the neuropsychological processes of reinforcement with the psychological needs that make generative AI personally rewarding. To address this gap, this paper proposes the Need–Reward–
Dependence (NRD) Framework, according to which problematic GenAI engagement develops through the interaction of three processes: unmet or salient psychological needs, rapid need-contingent reinforcement delivered by AI, and repeated cognitive or emotional outsourcing that strengthens habitual reliance. The framework proposes that competence, relatedness, autonomy, reassurance, emotional regulation, cognitive ease, social safety, and control can act as psychologically rewarding affordances of generative AI. Repeated reinforcement may increase cue-reactivity, habitual checking, perceived indispensability, cognitive delegation, emotional attachment, and impaired self-regulation in susceptible users. Importantly, this paper does not claim that generative-AI addiction is currently established as a distinct psychiatric disorder. Rather, it argues that a theoretically coherent and increasingly empirically observable continuum exists from adaptive use through reliance and dependence to potentially compulsive engagement associated with functional impairment. The paper concludes that future research should distinguish productive high-frequency use from clinically meaningful problematic use and should examine the interaction between users’ psychological vulnerabilities and the reward architecture of generative-AI systems.