With the widespread adoption of Large Language Models (LLMs), the prevalence of iterative interactions among these models is anticipated to increase. Notably, recent advancements in multi-round self-improving methods allow LLMs to generate new examples for training subsequent models. At the same time, multi-agent LLM systems, involving automated interactions among agents, are also increasing in prominence. Thus, in both short and long terms, LLMs may actively engage in an evolutionary process. We draw parallels between the behavior of LLMs and the evolution of human culture, as the latter has been extensively studied by cognitive scientists for decades. Our approach involves leveraging Iterated Learning (IL), a Bayesian framework that elucidates how subtle biases are magnified during human cultural evolution, to explain some behaviors of LLMs. This paper outlines key characteristics of agents' behavior in the Bayesian-IL framework, including predictions that are supported by experimental verification with various LLMs. This theoretical framework could help to more effectively predict and guide the evolution of LLMs in desired directions.
Bias Amplification in RAG: Poisoning Knowledge Retrieval to Steer LLMs
Faculty of Data Science, City University of Macau（数据科学学院，澳门城市大学）；Key Laboratory of Computing Power Network and Information Security, Ministry of Education, Shandong Computer Science Center, Qilu University of Technology (Shandong Academy of Sciences)（计算能力网络与信息安全重点实验室，教育部，山东计算机科学中心，齐鲁工业大学（山东科学院））；Shandong Provincial Key Laboratory of Computing Power Internet and Service Computing, Shandong Fundamental Research Center for Computer Science（山东省计算能力互联网与服务计算重点实验室，山东省计算机科学基础研究中心）
BRIDGE the Gap: Mitigating Bias Amplification in Automated Scoring of English Language Learners via Inter-group Data Augmentation
School of Computing, University of Georgia, Athens, GA, USA（佐治亚大学计算机学院，美国亚特兰大，GA）；Educational Testing Service, Princeton, NJ, USA（教育测试服务，美国新泽西州普林斯顿，NJ）；AI4STEM Education Center, University of Georgia, Athens, GA, USA（AI4STEM教育中心，佐治亚大学，美国亚特兰大，GA）；The Hong Kong Polytechnic University, Hong Kong, China（香港理工大学，中国香港）
School of Physics, Peking University（北京大学物理系）；Peking University（北京大学）；Center for High Energy Physics, Peking University（北京大学高能物理中心）；University of California, Berkeley（加州大学伯克利分校）；School of Physics and Astronomy, Beijing Normal University（北京师范大学物理天文学院）；Beijing Normal University（北京师范大学）；Key Laboratory of Multi-scale Spin Physics, Ministry of Education, Beijing Normal University（北京师范大学多尺度自旋物理重点实验室）；Beijing Computational Science Research Center（北京计算科学研究中心）
Mitigating Gender Bias Amplification in Distribution by Posterior Regularization
Tsinghua University（清华大学）；University of California, Los Angeles（加州大学洛杉矶分校）
先进的机器学习技术提高了自然语言处理的性能。然而，最近的研究，例如，Zhao et al.（2017）表明，这些技术无意中捕获了隐藏在语料库中的社会偏见，并进一步放大了它。然而，他们的分析仅针对模型的最佳预测。在本文中，我们从分布的角度研究了性别偏见放大问题，并证明了偏见是放大的标签上的预测概率分布的观点。我们进一步提出了一种基于后验正则化的偏差缓解方法。在性能损失很小的情况下，我们的方法几乎可以消除分布中的偏置放大。我们的研究有助于理解偏差放大效应。
The Fairness Collapse Phenomenon: Bias Amplification in Language Models Trained on Synthetic Data
Laboratoire Hubert Curien（于贝尔·屈里安实验室）；CNRS（法国国家科学研究中心）；Université Claude Bernard Lyon 1（里昂第一大学）；Université Lumière Lyon 2（里昂第二大学）；École Centrale de Lyon（里昂中央理工学院）
