Lune

ACL Findings2023

A Multi-dimensional study on Bias in Vision-Language models

Gabriele Ruggeri, Debora Nozza

Abstract

In recent years, joint Vision-Language (VL) models have increased in popularity and capability. Very few studies have attempted to investigate bias in VL models, even though it is a well-known issue in both individual modalities. This paper presents the first multi-dimensional analysis of bias in English VL models, focusing on gender, ethnicity, and age as dimensions. When subjects are input as images, pre-trained VL models complete a neutral template with a hurtful word 5% of the time, with higher percentages for female and young subjects. Bias presence in downstream models has been tested on Visual Question Answering. We developed a novel bias metric called the Vision-Language Association Test based on questions designed to elicit biased associations between stereotypical concepts and targets. Our findings demonstrate that pre-trained VL models contain biases that are perpetuated in downstream tasks.

Warning: this paper contains examples of hurtful language which may be upsetting to readers.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

12 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Align before Fuse: Vision and Language Representation Learning with Momentum Distillation cites ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision
  • Align before Fuse: Vision and Language Representation Learning with Momentum Distillation cites Learning Transferable Visual Models From Natural Language Supervision
  • BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation cites Align before Fuse: Vision and Language Representation Learning with Momentum Distillation
  • BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation cites ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision
  • BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation cites Learning Transferable Visual Models From Natural Language Supervision
  • OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework cites Align before Fuse: Vision and Language Representation Learning with Momentum Distillation
  • OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework cites Learning Transferable Visual Models From Natural Language Supervision
  • ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision cites Learning Transferable Visual Models From Natural Language Supervision
  • New Job, New Gender? Measuring the Social Bias in Image Generation Models cites Learning Transferable Visual Models From Natural Language Supervision
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers12

  1. Vision Language Models are BiasedAn Vo, Khai-Nguyen Nguyen, Mohammad Reza Taesiri, Thi Tuong Vy Dang et al.ICLR 2026 · 68 citations
  2. New Job, New Gender? Measuring the Social Bias in Image Generation ModelsWenxuan Wang, Haonan Bai, Jen-tse Huang, Yuxuan Wan et al.ACM MM 2024 · 19 citations
  3. VisBias: Measuring Explicit and Implicit Social Biases in Vision Language ModelsJen-Tse Huang, Jiantong Qin, Jianping Zhang, Youliang Yuan et al.EMNLP 2025 · 13 citations
  4. VideoLLaMB: Long Streaming Video Understanding with Recurrent Memory BridgesYuxuan Wang, Yiqi Song, Cihang Xie, Yang Liu et al.ICCV 2025 · 7 citations
  5. Metrics for What, Metrics for Whom: Assessing Actionability of Bias Evaluation Metrics in NLPPieter Delobelle, Giuseppe Attanasio, Debora Nozza, Su Lin Blodgett et al.EMNLP 2024 · 4 citations

Builds on14

  1. Learning Transferable Visual Models From Natural Language SupervisionAlec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh et al.ICML 2021 · 47,906 citations
  2. BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and GenerationJunnan Li, Dongxu Li, Caiming Xiong, Steven C. H. HoiICML 2022 · 6,549 citations
  3. Align before Fuse: Vision and Language Representation Learning with Momentum DistillationJunnan Li, Ramprasaath R. Selvaraju, Akhilesh Gotmare, Shafiq R. Joty et al.NeurIPS 2021 · 2,985 citations
  4. ViLT: Vision-and-Language Transformer Without Convolution or Region SupervisionWonjae Kim, Bokyung Son, Ildoo KimICML 2021 · 2,258 citations
  5. OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning FrameworkPeng Wang, An Yang, Rui Men, Junyang Lin et al.ICML 2022 · 1,058 citations

Related papers

  1. VIGNETTE: Socially Grounded Bias Evaluation for Vision-Language ModelsChahat Raj, Bowen Wei, Aylin Caliskan, Antonios Anastasopoulos et al.ACL 2026 · 3 citations
  2. Counterfactually Measuring and Eliminating Social Bias in Vision-Language Pre-training ModelsYi Zhang, Junyang Wang, Jitao SangACM MM 2022 · 11 citations
  3. BiasDora: Exploring Hidden Biased Associations in Vision-Language ModelsChahat Raj, Anjishnu Mukherjee, Aylin Caliskan, Antonios Anastasopoulos et al.EMNLP Findings 2024 · 7 citations
  4. ModSCAN: Measuring Stereotypical Bias in Large Vision-Language Models from Vision and Language ModalitiesYukun Jiang, Zheng Li, Xinyue Shen, Yugeng Liu et al.EMNLP 2024 · 1 citation
  5. See It from My Perspective: How Language Affects Cultural Bias in Image UnderstandingAmith Ananthram, Elias Stengel-Eskin, Mohit Bansal, Kathleen McKeownICLR 2025 · 3 citations
Dusk over the sea between two cliffs drawn in fine vertical lines