Visual input increases bias in multimodal large language models
Melange: A Synthetic Visual Bias Benchmark for Multi-modal Large Language Models
Computer Vision and Pattern Recognition
Summary
Biases in large language models have been studied before, but less is known about biases in models that use both text and images together. The authors created Melange, a new set of carefully made test images matched with text to better study bias in these multi-modal models. They tested 23 models and found that when models see images alongside text, they tend to show more biased answers compared to text alone. Even though models correctly recognize visual details, they still often give answers that align with stereotypes.
What this means in practice
- •For ai ethics teams: Evaluate and mitigate social biases in multi-modal AI systems by testing them with Melange’s visual bias benchmark.
- •For ai model developers: Use Melange to compare fairness impact of text-only versus multi-modal inputs during model development.
Authors
Razieh Rezaei, Ehsaneddin Asgary, Masoud Jalili Sabet
Abstract
Multi-modal Large Language Models (MLLMs) have seen broad adoption, but their biases are still far less examined than those of standard LLMs, creating possible social harms when deployed in real-world settings. Prior efforts to extend bias analyses from LLMs into the multi-modal space either cover only a few bias categories or do not ensure that the groups being compared are clearly addressed by visual cues in the images. To overcome these limitations, we introduce Melange, a wide-ranging synthetic visual benchmark covering seven categories for assessing social biases in MLLMs. It is grounded in the widely used language-only BBQ dataset (Bias Benchmark for Question Answering) from BigBench, carefully extending its constructed contexts via controlled generation of visual scenarios so the target attributes appear visually and clearly. To maximize reliability, our generated images pass through two stages of validation: automated screening using visual question answering (VQA) and subsequent manual inspection by human annotators. By systematically evaluating 23 open-source MLLMs across three matched evaluation settings---text-only, image-only, and image-with-context---Melange isolates the fairness effect of the input modality. Switching from text to vision reduces fairness by 0.291 on average, a gap that varies by model family rather than by scale alone. A perception control confirms that this gap is not an artifact of visual comprehension: models identify the depicted attributes at 90--95\% accuracy yet still fail to abstain. When models do commit to a biased answer, polarity-aware stereotype alignment shows they choose the human-stereotypical response roughly three-quarters of the time.