Paper Number
ICIS2025-2532
Paper Type
Short
Abstract
The rapid deployment of AI increases the need for transparency and accountability. However, open-washing, presenting models as open while withholding critical components, undermines scrutiny, reproducibility, and governance. Existing assessments prioritize organization-level openness, rely on small subjective audits, and cannot standardize evidence from unstructured model documentation, which prevents objective model-level transparency measurement at repository scale. We propose a method that integrates rule-based extraction of structured metadata with LLM assisted coding of unstructured documentation. Applied to thousands of Hugging Face repositories across 16 indicators and three dimensions, the method maps the current landscape of transparency. Results show pronounced heterogeneity and stratification: most models disclose only minimal information, while a small subset provide comprehensive documentation. The method enables scalable measurement and yields new empirical evidence to inform open-source AI governance.
Recommended Citation
Li, Yue; Kang, Lele; and Jiang, Qiqi, "Mapping Transparency of Open-Source AI Models: A Large-Scale Analysis on Hugging Face" (2025). ICIS 2025 Proceedings. 12.
https://aisel.aisnet.org/icis2025/public_is/public_is/12
Mapping Transparency of Open-Source AI Models: A Large-Scale Analysis on Hugging Face
The rapid deployment of AI increases the need for transparency and accountability. However, open-washing, presenting models as open while withholding critical components, undermines scrutiny, reproducibility, and governance. Existing assessments prioritize organization-level openness, rely on small subjective audits, and cannot standardize evidence from unstructured model documentation, which prevents objective model-level transparency measurement at repository scale. We propose a method that integrates rule-based extraction of structured metadata with LLM assisted coding of unstructured documentation. Applied to thousands of Hugging Face repositories across 16 indicators and three dimensions, the method maps the current landscape of transparency. Results show pronounced heterogeneity and stratification: most models disclose only minimal information, while a small subset provide comprehensive documentation. The method enables scalable measurement and yields new empirical evidence to inform open-source AI governance.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
20-PublicIS