VIGNETTE: Socially Grounded Bias Evaluation for Vision-Language Models

TL;DR

VIGNETTE introduces a large-scale benchmark to evaluate bias in vision-language models across various social stereotypes and decision-making tasks, revealing nuanced biases in model interpretations.

Contribution

The paper presents VIGNETTE, a comprehensive VQA benchmark with over 30 million images for analyzing complex social biases in VLMs, expanding beyond prior narrow studies.

Findings

VLMs exhibit subtle and complex stereotypical biases.

Models connect visual cues to social traits and hierarchies.

Bias patterns vary across different social dimensions.

Abstract

While bias in large language models (LLMs) is well-studied, similar concerns in vision-language models (VLMs) have received comparatively less attention. Existing VLM bias studies often focus on portrait-style images and gender-occupation associations, overlooking broader and more complex social stereotypes and their implied harm. This work introduces VIGNETTE, a large-scale VQA benchmark with 30M+ images for evaluating bias in VLMs through a question-answering framework spanning four directions: factuality, perception, stereotyping, and decision making. Beyond narrowly-centered studies, we assess how VLMs interpret identities in contextualized settings, revealing how models make trait and capability assumptions and exhibit patterns of discrimination. Drawing from social psychology, we examine how VLMs connect visual identity cues to trait and role-based inferences, encoding social hierarchies, through biased selections. Our findings uncover subtle, multifaceted, and surprising stereotypical patterns, offering insights into how VLMs construct social meaning from inputs.