Social Bias
Social bias in artificial intelligence, particularly large language models (LLMs) and related architectures like text-to-image generators, is a significant area of research focusing on identifying, measuring, and mitigating the perpetuation of societal prejudices within AI systems. Current efforts concentrate on developing robust bias detection methods, including novel datasets and evaluation metrics, and exploring various mitigation strategies such as adversarial training, counterfactual data augmentation, and prompt engineering techniques to reduce bias amplification. Understanding and addressing these biases is crucial for ensuring fairness, equity, and trustworthiness in AI applications across diverse sectors, impacting both the development of responsible AI and the broader societal implications of AI deployment.
Papers
Identifying Implicit Social Biases in Vision-Language Models
Kimia Hamidieh, Haoran Zhang, Walter Gerych, Thomas Hartvigsen, Marzyeh Ghassemi
Benchmarking Bias in Large Language Models during Role-Playing
Xinyue Li, Zhenpeng Chen, Jie M. Zhang, Yiling Lou, Tianlin Li, Weisong Sun, Yang Liu, Xuanzhe Liu
Do the Right Thing, Just Debias! Multi-Category Bias Mitigation Using LLMs
Amartya Roy, Danush Khanna, Devanshu Mahapatra, Vasanthakumar, Avirup Das, Kripabandhu Ghosh
AI Can Be Cognitively Biased: An Exploratory Study on Threshold Priming in LLM-Based Batch Relevance Assessment
Nuo Chen, Jiqun Liu, Xiaoyu Dong, Qijiong Liu, Tetsuya Sakai, Xiao-Ming Wu