🧪 Test?View on arXiv
Disentangling Algorithmic Bias from Archival Artifacts: A Controlled Audit of Vision-Language Model Valuation in Metropolitan Museum Archives
Author1, Author2, Author3, Author4, Author5
algorithmic biasmultimodalcultural heritageAI fairness
2609.17572
Builder Relevance
1h ago70%
Abstract
This study audits vision-language models for societal bias by distinguishing algorithmic valuation disparities from archival metadata confounders.
Reality Card
Core Claim
The study demonstrates that artist gender has no statistically significant conditional effect on the valuation of artworks by vision-language models, indicating the need for careful confound control in AI fairness assessments.
Method / Result
High score convergence with no significant gender effect (p > 0.20) across evaluated models.
Limitations
Exclusion of 41.2% unattributed holdings reflects institutional survival bias, which may affect the generalizability of results.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.