A Comprehensive Survey on Bias and Fairness in Generative AI: Legal, Ethical, and Technical Responses

Xiaojian Lin, Michael Losavio

Abstract: Recent advancements in generative AI, particularly in computer vision and natural language processing, have brought significant innovations and highlighted critical bias and fairness issues. This paper comprehensively reviews bias in generative AI, examining its causes, impacts, and potential solutions from legal, ethical, and technical perspectives. I begin by discussing the current state of bias in generative AI, focusing on racial, gender, and cultural biases in both computer vision and natural language processing. Through case studies, I demonstrate the real-world impacts of these biases. The paper then explores the root causes of bias, including data imbalance and algorithmic design. It discusses the profound social and technical impacts, such as implications for social justice, trust in AI systems, and model per- formance. I review existing domestic and international policies, industry standards, and prac- tices to mitigate AI bias, highlighting their strengths and limitations. The paper concludes with proposed solutions for improving data diversity, developing fairness-aware algorithms, en- hancing regulatory frameworks, and promoting ethical AI education and public awareness. Our study underscores the need for continuous efforts and interdisciplinary collaboration to address bias and ensure fairness in generative AI systems.