Addicted, attached, or just delegating? A scoping review on “problematic artificial intelligence use”

Background: The rapid diffusion of generative and conversational AI has raised concerns about problematic AI use and AI dependence. This led to a proliferation of studies addressing the problem, despite the lack of a common framework. This scoping review maps: (RQ1) definitions, (RQ2) measurement approaches, (RQ3) correlates and outcomes, and (RQ4) preliminary evidence across operationalization. Methods: Following PRISMA-ScR guidelines, we searched Web of Science and Scopus using predefined strings on artificial intelligence and problematic use. Thirty-seven empirical peer-reviewed studies were included. Results: Findings highlight inconsistent terminology and considerable heterogeneity in how problematic AI use is conceptualized, exacerbated by a frequent gap between constructs and operationalization that limits interpretation of outcomes. After recoding measures by their substantive operationalization, we analyzed evidence for three main strands: (1) behavioral addiction and/or compulsive use, consistently associated with depression, loneliness, social anxiety, escapism, flow state and low self-esteem/self-efficacy, where younger age and male gender emerge as risk factors; (2) cognitive (over)reliance, linked to performance expectations, academic stress/frustration of needs and literacy/trust in AI, with converging evidence of an erosion of downstream skills (and a decline in performance when AI is unavailable); and (3) Psychological and emotional dependence, associated with loneliness, anxious attachment, anthropomorphizing, and the perception of warmth/emotional intelligence, reliability and availability of AI. Conclusion: The field is fragmented and would benefit from clearer construct specification, AI-specific validated scales capturing all features of the phenomenon, and more longitudinal and experimental designs to clarify causal mechanisms and support safer system design.