Why it’s still open
Survey scales for ‘AI addiction’ borrow substance-use criteria and rarely show impairment; critics argue frequent use is being pathologized. Yet developers refuse to work without AI, companion users describe withdrawal-like distress during outages, and daily use correlates with depression. How to tell a useful tool used often from a pull that works against the user is unresolved.
What would count as an answer
Measures of functional impairment and of loss of control, validated against behavior, that distinguish dependence from valuable frequent use.
Papers that tackle it
- Ciudad-Fernandez et al 2025 - Questioning ChatGPT addiction — The case against the addiction label.
- Perlis et al 2026 - Generative AI use and depressive symptoms — Daily use and depression, stronger for personal use.
- Shen et al 2026 - The AI genie phenomenon — Three types of self-described chatbot addiction.
- METR 2026 - Developer productivity update — Refusing to work without AI.
- Fang et al 2025 - Psychosocial effects of chatbot use — Problematic use rises with heavier use.
Methods in play
- Cross-sectional surveys
- Forum and Reddit text analysis
- Productivity field experiments
- Randomized controlled trials
Who’s closest
Placed by the editor, v1 — corrections welcome.
Related questions
Know a paper that belongs here, or think this is framed wrongly? Suggest a paper or a correction (forms on the Atlas home).