VEAT Quantifies Implicit Associations in Text-to-Video Generator Sora and Reveals Challenges in Bias Mitigation

TL;DR

This paper introduces VEAT, a new method to quantify societal biases in text-to-video generators like Sora, revealing that these models often reflect and can amplify harmful stereotypes despite debiasing efforts.

Contribution

The paper develops VEAT and SC-VEAT to measure biases in video generation, demonstrating their effectiveness and exposing challenges in bias mitigation for T2V models.

Findings

Sora exhibits strong associations of race and gender with valence in generated videos.

Debiasing prompts can sometimes increase biases instead of reducing them.

Biases in T2V models correlate with real-world demographic distributions.

Abstract

Text-to-Video (T2V) generators such as Sora raise concerns about whether generated content reflects societal bias. We extend embedding-association tests from words and images to video by introducing the Video Embedding Association Test (VEAT) and Single-Category VEAT (SC-VEAT). We validate these methods by reproducing the direction and magnitude of associations from widely used baselines, including Implicit Association Test (IAT) scenarios and OASIS image categories. We then quantify race (African American vs. European American) and gender (women vs. men) associations with valence (pleasant vs. unpleasant) across 17 occupations and 7 awards. Sora videos associate European Americans and women more with pleasantness (both d>0.8). Effect sizes correlate with real-world demographic distributions: percent men and White in occupations (r=0.93, r=0.83) and percent male and non-Black among award recipients (r=0.88, r=0.99). Applying explicit debiasing prompts generally reduces effect-size magnitudes, but can backfire: two Black-associated occupations (janitor, postal service) become more Black-associated after debiasing. Together, these results reveal that easily accessible T2V generators can actually amplify representational harms if not rigorously evaluated and responsibly deployed.