DebiasPI: Inference-time Debiasing by Prompt Iteration of a Text-to-Image Generative Model

2025-01-28

AI Summary

AI text-to-image models exhibit systematic biases across demographic attributes—including gender, race, age, occupation, body type, and skin tone. This paper introduces DebiasPI, a training-free, inference-time prompting method that dynamically optimizes textual prompts via a closed-loop feedback mechanism to achieve multi-attribute fairness. Its key contributions are: (1) the first real-time feedback control loop integrating attribute classifiers and latent-state probing; (2) fine-grained, interactive joint intervention across multiple attributes; and (3) empirical discovery of cross-attribute intervention effects—e.g., optimizing racial distribution improves gender diversity but reduces skin-tone diversity—as well as latent bias patterns, such as significantly degraded generation quality for light skin tones. Evaluated on news headline visualization, DebiasPI effectively achieves balanced generation across gender and race without model retraining.

Technology Category

• Computer Vision: Bias, Fairness & Privacy
• Philosophy and Ethics of AI: Bias, Fairness & Equity
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems
• Social Networks and Social Media: Fairness and bias in social network and social media analysis

Abstract

Ethical intervention prompting has emerged as a tool to counter demographic biases of text-to-image generative AI models. Existing solutions either require to retrain the model or struggle to generate images that reflect desired distributions on gender and race. We propose an inference-time process called DebiasPI for Debiasing-by-Prompt-Iteration that provides prompt intervention by enabling the user to control the distributions of individuals' demographic attributes in image generation. DebiasPI keeps track of which attributes have been generated either by probing the internal state of the model or by using external attribute classifiers. Its control loop guides the text-to-image model to select not yet sufficiently represented attributes, With DebiasPI, we were able to create images with equal representations of race and gender that visualize challenging concepts of news headlines. We also experimented with the attributes age, body type, profession, and skin tone, and measured how attributes change when our intervention prompt targets the distribution of an unrelated attribute type. We found, for example, if the text-to-image model is asked to balance racial representation, gender representation improves but the skin tone becomes less diverse. Attempts to cover a wide range of skin colors with various intervention prompts showed that the model struggles to generate the palest skin tones. We conducted various ablation studies, in which we removed DebiasPI's attribute control, that reveal the model's propensity to generate young, male characters. It sometimes visualized career success by generating two-panel images with a pre-success dark-skinned person becoming light-skinned with success, or switching gender from pre-success female to post-success male, thus further motivating ethical intervention prompting with DebiasPI.

Problem

• AI Image Bias
• Representation Inequality
• Fairness in AI

Innovation

• DebiasPI
• Bias Mitigation
• Feature Balance