Tokenaholics: The New Silent Addiction of the Artificial Intelligence Era
The expansion of generative artificial intelligence makes it necessary to distinguish computational consumption, adoption, and problematic dependence. This article proposes Tokenaholics as an exploratory concept without diagnostic equivalence and combines an interdisciplinary critical review with secondary analysis of public aggregates from 2025 and 2026. It examines Anthropic usage profiles by country and occupational group, Vercel daily token and spending shares, and GitHub indicators. K-means using five variables selects a two-group partition among 121 countries, with limited separation (silhouette of 0.308). The GitHub– Anthropic comparison matches eight of the ten countries in the GitHub table. Computer and mathematical occupations have published shares of 23.80% in chat and 28.17% in the API in May 2026; these measures retain each channel’s denominator. Temporal evaluation of Ridge and three baselines does not support stable predictive superiority. Budget scenarios for 2027 combine token quantity and effective price under a constant mix. The findings describe geographic heterogeneity, differences between channels, and distinct distributions of tokens and expenditure. These results characterize the sociotechnical context without operationalizing or validating Tokenaholics, estimating the prevalence of dependence, or identifying causality. The contribution to computer science is to connect a bounded conceptual definition with reproducible analytical procedures and a future agenda for evaluating regulation, verification, and functional impairment.
Authors
- Murillo Edson de Carvalho Souza (ORCID: https://orcid.org/0000-0002-3827-1952)
Institutions
- Universidade de Brasília (BR)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24247867
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- preprint