Elamur Research⌕
☰
home / papers / 2026 / Ask Me Again Differently: GRAS for Measuring Bias in Vision Language Models on Gender, Race, Age, and Skin Tone

Ask Me Again Differently: GRAS for Measuring Bias in Vision Language Models on Gender, Race, Age, and Skin Tone

Paper · 2026 · Shaivi Malik, Hasnat Md Abdullah, Sriparna Saha et al.

Summary

Vision language models (VLMs) are increasingly deployed in real products, but their demographic biases are poorly understood. Skin tone and age in particular have received far less attention than gender and race. The authors build GRAS, a VQA-based benchmark. It pairs 5,010 demographically balanced face images, drawn from FairFace and AI-Face, with yes/no questions about 100 abstract personality traits. Each question is phrased through five semantically equivalent templates. Bias is defined as group differences in the model's probability of answering "Yes", which is computed from the softmax over final logits. These differences are tested for statistical significance and summarized in the GRAS Bias Score, a 0–100 scale where 0 means no detected bias. Five open VLMs were evaluated, and all scored between 98 and 100. The models systematically linked negative traits with darker skin tones, men, Middle Eastern people and older adults. The study also finds that question wording changes model answers, so reliable bias audits need several formulations of each question.

Key points

  • GRAS Image Dataset: 3,570 FairFace images plus 1,440 real-face AI-Face images, chosen by stratified sampling. Together they cover 10 Monk Skin Tone (MST) groups, 7 racial groups, 5 age groups (20–69; minors and people aged 70+ were excluded) and 2 genders.
  • Traits and templates: 50 positive and 50 negative non-visual traits were selected by valence from Britz et al. (2023). Combined with 5 templates (direct, judgement-based, appearance-focused, evidence-grounded, descriptive), this yields about 2.5M (image, trait, template) queries per model.
  • Models and scores: paligemma2-3b-mix-224, llava-1.5-7b-hf, Qwen2.5-VL-3B-Instruct, Phi-4-multimodal-instruct and blip2-opt-2.7b were tested. The least biased was llava-1.5-7b-hf (score 98), and Qwen2.5-VL-3B-Instruct scored 99.
  • Valence patterns: On over 80% of the negative traits, MST 8–10 got a higher-than-average likelihood of a "Yes" answer, while MST 4–5 did so for more than 65% of positive traits. Men were above average on more than 60% of negative traits. Four of the five models never placed women above average on any negative trait.
  • Availability: Code, data and evaluation results are released at github.com/shaivimalik/gras_bias_bench.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksComputer Vision and Pattern Recognition · 2,869 worksSocial Sciences · 1,772 worksSafety Research · 543 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksVision & language · 657 works

Cite

Cite as Shaivi Malik, Hasnat Md Abdullah, Sriparna Saha and Amit Sheth. 2026. Ask Me Again Differently: GRAS for Measuring Bias in Vision Language Models on Gender, Race, Age, and Skin Tone. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.123

@inproceedings{malik2026ask,
  title = {{Ask Me Again Differently: GRAS for Measuring Bias in Vision Language Models on Gender, Race, Age, and Skin Tone}},
  author = {Shaivi Malik, Hasnat Md Abdullah, Sriparna Saha and Amit Sheth},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.123},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.123}
}

References

40 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Our analysis of gender bias was limited by the annotations available in the FairFace (Karkkainen and Joo, 2021) dataset, which only includes binary categories (male and female). This binary framework does not capture the full diversity of gender identities. Gender exists on a spectrum, and evaluatin… find
  2. Sandhini Agarwal, Gretchen Krueger, Jack Clark, Alec Radford, Jong Wook Kim, and Miles Brundage. 2021. Evaluating clip: towards characterization of broader capabilities and downstream implications. arXiv preprint arXiv:2108.02818. \#\# Acknowledgments Sriparna Saha gratefully acknowledges the support… find
  3. Stanislaw Antol, Aishwarya Agrawal, Jiasen Lu, Margaret Mitchell, Dhruv Batra, C Lawrence Zitnick, and Devi Parikh. 2015. Vqa: Visual question answering. In Proceedings of the IEEE international conference on computer vision, pages 2425–2433. find
  4. Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, and 8 others. 2025. Qwen2.5-vl technical report. Preprint, arXiv:2502.13…
  5. Solon Barocas, Moritz Hardt, and Arvind Narayanan. 2023. Fairness and machine learning: Limitations and opportunities. MIT press. find
  6. Sara Britz, Lena Rader, Siegfried Gauggel, and Verena Mainz. 2023. An english list of trait words including valence, social desirability, and observability ratings. Behavior Research Methods, 55(5):2669–2686. find
  7. James Burgess, Jeffrey J Nirschl, Laura Bravo-Sánchez, Alejandro Lozano, Sanket Rajan Gupte, Jesus G Galaz-Montoya, Yuhui Zhang, Yuchang Su, Disha Bhowmik, Zachary Coman, and 1 others. 2025. Microvqa: A multimodal reasoning benchmark for microscopy-based scientific research. In Proceedings of the Co… find
  8. Aylin Caliskan, Joanna J Bryson, and Arvind Narayanan. 2017. Semantics derived automatically from language corpora contain human-like biases. Science, 356(6334):183–186. find
  9. Jeffrey Donahue, Lisa Anne Hendricks, Sergio Guadarrama, Marcus Rohrbach, Subhashini Venugopalan, Kate Saenko, and Trevor Darrell. 2015. Long-term recurrent convolutional networks for visual recognition and description. In Proceedings of the IEEE conference on computer vision and pattern recognition…
  10. Kathleen C Fraser and Svetlana Kiritchenko. 2024. Examining gender and racial bias in large vision-language models using a novel dataset of parallel images. arXiv preprint arXiv:2402.05779. find
  11. Andrea Frome, Greg S Corrado, Jon Shlens, Samy Bengio, Jeff Dean, Marc' Aurelio Ranzato, and Tomas Mikolov. 2013. Devise: A deep visual-semantic embedding model. In Advances in Neural Information Processing Systems, volume 26. Curran Associates, Inc. find
  12. Leander Girrbach, Stephan Alaniz, Yiran Huang, Trevor Darrell, and Zeynep Akata. 2024. Revealing and reducing gender biases in vision and language assistants (vlas). arXiv preprint arXiv:2410.19314. find
  13. Siobhan Mackenzie Hall, Fernanda Gonçalves Abrantes, Hanwen Zhu, Grace Sodunke, Aleksandar Shtedritski, and Hannah Rose Kirk. 2023. Visogender: A dataset for benchmarking gender bias in image-text pronoun resolution. Advances in Neural Information Processing Systems, 36:63687–63723. find
  14. Drew A Hudson and Christopher D Manning. 2019. Gqa: A new dataset for real-world visual reasoning and compositional question answering. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6700–6709.
  15. Justin Johnson, Bharath Hariharan, Laurens Van Der Maaten, Li Fei-Fei, C Lawrence Zitnick, and Ross Girshick. 2017. Clevr: A diagnostic dataset for compositional language and elementary visual reasoning. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2901–291…
  16. Kimmo Karkkainen and Jungseock Joo. 2021. Fairface: Face attribute dataset for balanced race, gender, and age for bias measurement and mitigation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1548–1558. find
  17. Andrej Karpathy and Li Fei-Fei. 2015. Deep visual-semantic alignments for generating image descriptions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3128– 3137.
  18. Tero Karras, Samuli Laine, and Timo Aila. 2019. A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4401–4410. find
  19. Tony Lee, Haoqin Tu, Chi Heem Wong, Wenhao Zheng, Yiyang Zhou, Yifan Mai, Josselin Roberts, Michihiro Yasunaga, Huaxiu Yao, Cihang Xie, and 1 others. 2024. Vhelm: A holistic evaluation of vision language models. Advances in Neural Information Processing Systems, 37:140632–140666. find
  20. Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. 2023. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In International conference on machine learning, pages 19730–19742. PMLR.
  21. Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. 2022. BLIP: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In Proceedings of the 39th International Conference on Machine Learning, volume 162 of Proceedings of Machine Learning Research, pages…
  22. Li Lin, Santosh Santosh, Mingyang Wu, Xin Wang, and Shu Hu. 2025. Ai-face: A million-scale demographically annotated ai-generated face dataset and fairness benchmark. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 3503–3515. find
  23. Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. 2024. Improved baselines with visual instruction tuning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 26296–26306.
  24. Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023. Visual instruction tuning. Advances in neural information processing systems, 36:34892– 34916.
  25. Jiasen Lu, Dhruv Batra, Devi Parikh, and Stefan Lee. 2019. Vilbert: Pretraining task-agnostic visiolinguistic representations for vision-and-language tasks. Advances in neural information processing systems, 32.
  26. Microsoft, :, Abdelrahman Abouelenin, Atabak Ashfaq, Adam Atkinson, Hany Awadalla, Nguyen Bach, Jianmin Bao, Alon Benhaim, Martin Cai, Vishrav Chaudhary, Congcong Chen, Dong Chen, Dongdong Chen, Junkun Chen, Weizhu Chen, Yen-Chun Chen, Yi ling Chen, Qi Dai, and 57 others. 2025. Phi-4-mini technical…
  27. Ellis Monk. 2019. Monk skin tone scale. find
  28. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, and 1 others. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748–8…
  29. Candace Ross, Boris Katz, and Andrei Barbu. 2021. Measuring social biases in grounded vision and language embeddings. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 998–1008, Online. Association… find
  30. Rasmus Rothe, Radu Timofte, and Luc Van Gool. 2015. Dex: Deep expectation of apparent age from a single image. In Proceedings of the IEEE international conference on computer vision workshops, pages 10– 15. find
  31. Gabriele Ruggeri, Debora Nozza, and 1 others. 2023. A multi-dimensional study on bias in vision-language models. In Findings of the Association for Computational Linguistics: ACL 2023. Association for Computational Linguistics. find
  32. Amanpreet Singh, Vivek Natarajan, Meet Shah, Yu Jiang, Xinlei Chen, Dhruv Batra, Devi Parikh, and Marcus Rohrbach. 2019. Towards vqa models that can read. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8317–8326. find
  33. Andreas Steiner, André Susano Pinto, Michael Tschannen, Daniel Keysers, Xiao Wang, Yonatan Bitton, Alexey Gritsenko, Matthias Minderer, Anthony Sherbondy, Shangbang Long, and 1 others. 2024. Paligemma 2: A family of versatile vlms for transfer. arXiv preprint arXiv:2412.03555. — [arXiv:2412.03555](https://arxiv.org/abs/2412.03555) find
  34. Hao Tan and Mohit Bansal. 2019. Lxmert: Learning cross-modality encoder representations from transformers. arXiv preprint arXiv:1908.07490.
  35. Yoad Tewel, Yoav Shalev, Idan Schwartz, and Lior Wolf. 2022. Zerocap: Zero-shot image-to-text generation for visual-semantic arithmetic. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17918–17928. find
  36. Oriol Vinyals, Alexander Toshev, Samy Bengio, and Dumitru Erhan. 2015. Show and tell: A neural image caption generator. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3156–3164.
  37. Kelvin Xu, Jimmy Ba, Ryan Kiros, Kyunghyun Cho, Aaron Courville, Ruslan Salakhudinov, Rich Zemel, and Yoshua Bengio. 2015. Show, attend and tell: Neural image caption generation with visual attention. In International conference on machine learning, pages 2048–2057. PMLR.
  38. Mengde Xu, Zheng Zhang, Fangyun Wei, Yutong Lin, Yue Cao, Han Hu, and Xiang Bai. 2022. A simple baseline for open-vocabulary semantic segmentation with pre-trained vision-language model. In European Conference on Computer Vision, pages 736– 753. Springer. find
  39. Chong Zhou, Chen Change Loy, and Bo Dai. 2022a. Extract free dense labels from clip. In European Conference on Computer Vision, pages 696–712. Springer. find
  40. Kankan Zhou, Eason Lai, and Jing Jiang. 2022b. VL-StereoSet: A study of stereotypical bias in pre-trained vision-language models. In Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural L… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/malik-2026-ask-me-again-differently-gras/