Papers/2609.01691
🧪 Test?View on arXiv

FairLens: Benchmarking Fairness in Vision-Language Models for High-Stakes Decision-Making

Not provided in the content

fairnessmultimodalbenchmarkinghigh-stakes decision-making
2609.01691
Builder Relevance
80%
2h ago

Abstract

FAIRLENS introduces a benchmark for evaluating fairness and validity in vision-language models used in high-stakes decision-making.

Reality Card

Core Claim

FAIRLENS demonstrates that the primary failure in VLMs is unwarranted inference from visual inputs rather than unequal treatment across demographic groups.

Method / Result

The weakest model inferred qualifications or roles from images on 99% of questions it could not answer.

Limitations

The study highlights that disparity metrics alone may not capture the severity of model failures in high-stakes contexts.

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.
← Back to all papers