Examining Gender and Racial Bias in Large Vision-Language Models Using a Novel Dataset of Parallel Images

TL;DR

This paper investigates gender and racial biases in large vision-language models by introducing a novel dataset of similar images differing in gender and race, revealing significant bias-related response differences.

Contribution

The paper presents PAIRS, a new dataset of parallel images designed to test bias in LVLMs, and demonstrates bias detection through model response analysis.

Findings

LVLMs show significant bias based on gender and race in responses.

The PAIRS dataset enables controlled bias testing in vision-language models.

Bias varies depending on the perceived characteristics of individuals in images.

Abstract

Following on recent advances in large language models (LLMs) and subsequent chat models, a new wave of large vision-language models (LVLMs) has emerged. Such models can incorporate images as input in addition to text, and perform tasks such as visual question answering, image captioning, story generation, etc. Here, we examine potential gender and racial biases in such systems, based on the perceived characteristics of the people in the input images. To accomplish this, we present a new dataset PAIRS (PArallel Images for eveRyday Scenarios). The PAIRS dataset contains sets of AI-generated images of people, such that the images are highly similar in terms of background and visual content, but differ along the dimensions of gender (man, woman) and race (Black, white). By querying the LVLMs with such images, we observe significant differences in the responses according to the perceived gender or race of the person depicted.