Can we Debias Social Stereotypes in AI-Generated Images? Examining Text-to-Image Outputs and User Perceptions

2025/05/27 by Saharsh Barve, Barve, Saharsh, Axiu Mao +7 · 2 citations
Social Sciences · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Computational and Text Analysis Methods #Cultural diversity #Debiasing #Diversity (politics) #Ethics and Social Impacts of AI #FOS: Computer and information sciences #Human-Computer Interaction (cs.HC) #Operationalization #Perception #Relevance (law) #Rubric #Set (abstract data type) #Stereotype (UML)

paper · pdf · doi:10.48550/arxiv.2505.20692

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/05/27 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

Recent advances in generative AI have enabled visual content creation through text-to-image (T2I) generation. However, despite their creative potential, T2I models often replicate and amplify societal stereotypes -- particularly those related to gender, race, and culture -- raising important ethical concerns. This paper proposes a theory-driven bias detection rubric and a Social Stereotype Index (SSI) to systematically evaluate social biases in T2I outputs. We audited three major T2I model outputs -- DALL-E-3, Midjourney-6.1, and Stability AI Core -- using 100 queries across three categories -- geocultural, occupational, and adjectival. Our analysis reveals that initial outputs are prone to include stereotypical visual cues, including gendered professions, cultural markers, and western beauty norms. To address this, we adopted our rubric to conduct targeted prompt refinement using LLMs, which significantly reduced bias -- SSI dropped by 61% for geocultural, 69% for occupational, and 51% for adjectival queries. We complemented our quantitative analysis through a user study examining perceptions, awareness, and preferences around AI-generated biased imagery. Our findings reveal a key tension -- although prompt refinement can mitigate stereotypes, it can limit contextual alignment. Interestingly, users often perceived stereotypical images to be more aligned with their expectations. We discuss the need to balance ethical debiasing with contextual relevance and call for T2I systems that support global diversity and inclusivity while not compromising the reflection of real-world social complexity.

Citations

• BetterBench: Assessing AI Benchmarks, Uncovering Issues, and Establishing Best Practices
• Interpretations, Representations, and Stereotypes of Caste within Text-to-Image Generators
• Breaking the Global North Stereotype: A Global South-centric Benchmark Dataset for Auditing and Mitigating Biases in Facial Recognition Systems
• DiffInject: Revisiting Debias via Synthetic Data Generation using Diffusion-based Style Injection
• Survey of Bias In Text-to-Image Generation: Definition, Evaluation, and Mitigation
• The Male CEO and the Female Assistant: Evaluation and Mitigation of Gender Biases in Text-To-Image Generation of Dual Subjects
• SCoFT: Self-Contrastive Fine-Tuning for Equitable Image Generation
• Finetuning Text-to-Image Diffusion Models for Fairness
• Analysing Gender Bias in Text-to-Image Models using Object Detection
• Multimodal Composite Association Score: Measuring Gender Bias in Generative Multimodal Models
• Social Biases through the Text-to-Image Generation Lens
• Stable Bias: Analyzing Societal Representations in Diffusion Models
• Sensing Wellbeing in the Workplace, Why and For Whom? Envisioning Impacts with Organizational Stakeholders
• Charting the Sociotechnical Gap in Explainable AI: A Framework to Address the Gap in XAI
• A Validity Perspective on Evaluating the Justified Use of Data-driven Decision-making Algorithms
• Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding
• Hierarchical Text-Conditional Image Generation with CLIP Latents
• Assessing the Fairness of AI Systems: AI Practitioners' Processes, Challenges, and Needs for Support
• Overcoming Failures of Imagination in AI Infused System Development and Deployment
• Closing the AI Accountability Gap: Defining an End-to-End Framework for Internal Algorithmic Auditing
• Closing the AI accountability gap
• A Survey on Bias and Fairness in Machine Learning
• Implicit Diversity in Image Summarization
• AI4People—An Ethical Framework for a Good AI Society: Opportunities, Risks, Principles, and Recommendations
• Model Cards for Model Reporting
• Datasheets for Datasets
• Stereotypes in Search Engine Results: Understanding The Role of Local and Global Factors
• Identifying Stereotypes in the Online Perception of Physical Attractiveness

Cited by

• Aligned but Stereotypical? How System Prompts Shape Demographic Bias in LLM-Based Text-to-Image Models
• When Cars Have Stereotypes: Auditing Demographic Bias in Objects from Text-to-Image Models

Related

• Hidden Bias in the Machine: Stereotypes in Text-to-Image Models
• T2IBias: Uncovering Societal Bias Encoded in the Latent Space of Text-to-Image Generative Models
• Social Biases through the Text-to-Image Generation Lens
• Survey of Bias In Text-to-Image Generation: Definition, Evaluation, and Mitigation
• Bias in Generative AI