# Images Speak Louder than Words: Understanding and Mitigating Bias in Vision-Language Model from a Causal Mediation Perspective

arXiv: https://arxiv.org/abs/2407.02814 · HTML: https://mlsanity.com/paper/2407.02814/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Placed in Social bias and fairness in VLMs, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 23 of its 35 references that are in the map.
- First posted on arXiv 3 Jul 2024.

## Placed in

- Multimodal LLMs / vision-language models / [Social bias and fairness in VLMs](https://mlsanity.com/mllm/social-bias-and-fairness-in-vlms/)

## Builds on (most cited first, 23 of 23)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://mlsanity.com/paper/1801.04381/) (2018)
- [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://mlsanity.com/paper/2102.05918/) (2021)
- [VQA: Visual Question Answering](https://mlsanity.com/paper/1505.00468/) (2015)
- [ViLBERT: Pretraining Task-Agnostic Visiolinguistic Representations for Vision-and-Language Tasks](https://mlsanity.com/paper/1908.02265/) (2019)
- [Flickr30k Entities: Collecting Region-to-Phrase Correspondences for Richer Image-to-Sentence Models](https://mlsanity.com/paper/1505.04870/) (2015)
- [Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings](https://mlsanity.com/paper/1607.06520/) (2016)
- [ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision](https://mlsanity.com/paper/2102.03334/) (2021)
- [Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks](https://mlsanity.com/paper/1604.02878/) (2016)
- [Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints](https://mlsanity.com/paper/1707.09457/) (2017)
- [Measuring Bias in Contextualized Word Representations](https://mlsanity.com/paper/1906.07337/) (2019)
- [Analyzing the Structure of Attention in a Transformer Language Model](https://mlsanity.com/paper/1906.04284/) (2019)
- [Gender Bias in Contextualized Word Embeddings](https://mlsanity.com/paper/1904.03310/) (2019)
- [Alpha-CLIP: A CLIP Model Focusing on Wherever You Want](https://mlsanity.com/paper/2312.03818/) (2023)
- [DeAR: Debiasing Vision-Language Models with Additive Residuals](https://mlsanity.com/paper/2303.10431/) (2023)
- [Directional Bias Amplification](https://mlsanity.com/paper/2102.12594/) (2021)
- [Worst of Both Worlds: Biases Compound in Pre-trained Vision-and-Language Models](https://mlsanity.com/paper/2104.08666/) (2021)
- [VQA-E: Explaining, Elaborating, and Enhancing Your Answers for Visual Questions](https://mlsanity.com/paper/1803.07464/) (2018)
- [Counterfactually Measuring and Eliminating Social Bias in Vision-Language Pre-training Models](https://mlsanity.com/paper/2207.01056/) (2022)
- [FairerCLIP: Debiasing CLIP's Zero-Shot Predictions using Functions in RKHSs](https://mlsanity.com/paper/2403.15593/) (2024)
- [Men Also Do Laundry: Multi-Attribute Bias Amplification](https://mlsanity.com/paper/2210.11924/) (2022)

## Built on by (most cited first, 13 of 13)

- [MemMA: Coordinating the Memory Cycle through Multi-Agent Reasoning and In-Situ Self-Evolution](https://mlsanity.com/paper/2603.18718/) (2026)
- [Bias Is a Subspace, Not a Coordinate: A Geometric Rethinking of Post-hoc Debiasing in Vision-Language Models](https://mlsanity.com/paper/2511.18123/) (2025)
- [Interpretable Debiasing of Vision-Language Models for Social Fairness](https://mlsanity.com/paper/2602.24014/) (2026)
- [Bias in Gender Bias Benchmarks: How Spurious Features Distort Evaluation](https://mlsanity.com/paper/2509.07596/) (2025)
- [Beyond Spurious Signals: Debiasing Multimodal Large Language Models via Counterfactual Inference and Adaptive Expert Routing](https://mlsanity.com/paper/2509.15361/) (2025)
- [Uncovering Cultural Representation Disparities in Vision-Language Models](https://mlsanity.com/paper/2505.14729/) (2025)
- [DSO: Direct Steering Optimization for Bias Mitigation](https://mlsanity.com/paper/2512.15926/) (2025)
- [Freeze and Reveal: Exposing Modality Bias in Vision-Language Models](https://mlsanity.com/paper/2508.07432/) (2025)
- [Vision-Language Models Suppress Female Representations Under Ambiguous Input](https://mlsanity.com/paper/2605.31556/) (2026)
- [SDGBiasBench: Benchmarking and Mitigating Vision--Language Models' Biases in Sustainable Development Goals](https://mlsanity.com/paper/2605.21919/) (2026)
- [KG-FairDiff: Knowledge Graph-Guided Prompt Refinement for Demographically Fair Text-to-Image Generation](https://mlsanity.com/paper/2606.01282/) (2026)
- [Guardrail-Agnostic Societal Bias Evaluation in Large Vision-Language Models](https://mlsanity.com/paper/2608.29590/) (2026)
- [ViD: Vision-Dominant Gender Bias Mitigation for Large Vision-Language Models](https://mlsanity.com/paper/2609.16647/) (2026)

## Abstract (arXiv)

Vision-language models (VLMs) pre-trained on extensive datasets can inadvertently learn biases by correlating gender information with specific objects or scenarios. Current methods, which focus on modifying inputs and monitoring changes in the model's output probability scores, often struggle to comprehensively understand bias from the perspective of model components. We propose a framework that incorporates causal mediation analysis to measure and map the pathways of bias generation and propagation within VLMs. This approach allows us to identify the direct effects of interventions on model bias and the indirect effects of interventions on bias mediated through different model components. Our results show that image features are the primary contributors to bias, with significantly higher impacts than text features, specifically accounting for 32.57% and 12.63% of the bias in the MSCOCO and PASCAL-SENTENCE datasets, respectively. Notably, the image encoder's contribution surpasses that of the text encoder and the deep fusion encoder. Further experimentation confirms that contributions from both language and vision modalities are aligned and non-conflicting. Consequently, focusing on blurring gender representations within the image encoder, which contributes most to the model bias, reduces bias efficiently by 22.03% and 9.04% in the MSCOCO and PASCAL-SENTENCE datasets, respectively, with minimal performance loss or increased computational demands.