Interactive Discovery and Exploration of Visual Bias in Generative Text-to-Image Models
For AI ethics researchers and practitioners, ViBEx provides a novel interactive method to systematically uncover and analyze visual biases in T2I models, addressing the challenge of exploring the vast output space.
The paper introduces ViBEx, an interactive tool for discovering visual biases in generative text-to-image models, enabling experts to uncover previously undocumented biases through a prompting tree interface and CLIP-based probing.
Bias in generative Text-to-Image (T2I) models is a known issue, yet systematically analyzing such models' outputs to uncover it remains challenging. We introduce the Visual Bias Explorer (ViBEx) to interactively explore the output space of T2I models to support the discovery of visual bias. ViBEx introduces a novel flexible prompting tree interface in combination with zero-shot bias probing using CLIP for quick and approximate bias exploration. It additionally supports in-depth confirmatory bias analysis through visual inspection of forward, intersectional, and inverse bias queries. ViBEx is model-agnostic and publicly available. In four case study interviews, experts in AI and ethics were able to discover visual biases that have so far not been described in literature.