通知：联邦节假日调休：哥伦布日 / 原住民日（Columbus Day / Indigenous Peoples' Day），今日（13日）无更新。通知：联邦节假日调休：哥伦布日 / 原住民日（Columbus Day / Indigenous Peoples' Day），今日（13日）无更新。
arXiv 2505.19212cs.CLcs.AIcs.CY
当伦理与收益相悖时：道德两难情境下的大语言模型智能体
When Ethics and Payoffs Diverge: LLM Agents in Morally Charged Social Dilemmas
ETH Zürich（苏黎世联邦理工学院）；University of Zurich（苏黎世大学）；Carnegie Mellon University（卡内基梅隆大学）；University of Michigan（密歇根大学）；Max Planck Institute for Intelligent Systems, Tübingen（图宾根人工智能研究所）；University of Toronto（多伦多大学）；Vector Institute（向量研究所）
Steffen Backmann, David Guzman Piedrahita, Terry Jingchen Zhang, Emanuel Tewolde, Rada Mihalcea, Bernhard Schölkopf, Zhijing Jin
Recent advances in LLMs have enabled their use in complex agentic roles, involving decision-making with humans or other agents, making ethical alignment a critical concern. While prior work has examined LLMs' moral judgment and strategic behavior separately, there is limited understanding of how they act when moral imperatives directly conflict with profit incentives. We introduce \msimfull (\msim) to evaluate how LLMs behave in the prisoner's dilemma and public goods game embedded in morally charged contexts, varying moral framing, opponent behavior, and survival pressure across nine models. Beyond measuring behavior, we estimate the causal effect of each factor via average treatment effects (ATEs) and analyze agents' own reasoning traces to characterize the motives behind their choices. We find that no model remains consistently moral, with cooperation rates ranging from 7.9\% to 76.3\%. Game structure and moral framing are the strongest causal drivers of moral behavior, while reasoning-trace analysis reveals distinct motive profiles across models, ranging from predominantly payoff-maximizing to moral- and reputation-oriented. Together, these results expose the situational brittleness of current LLMs' moral behavior and the risk of deploying them where profit incentives conflict with ethical guidelines.
The Staircase of Ethics: Probing LLM Value Priorities through Multi-Step Induction to Complex Moral Dilemmas
University of Chinese Academy of Sciences（中国科学院大学）；Institute of Computing Technology, Chinese Academy of Sciences（中国科学院计算技术研究所）；Media Synthesis and Forensics Lab（媒体合成与取证实验室）；Zhongguancun Laboratory（中关村实验室）
Deconstructing The Ethics of Large Language Models from Long-standing Issues to New-emerging Dilemmas: A Survey
Rutgers University（罗格斯大学）；University of Technology Sydney（悉尼科技大学）；Ohio State University（俄亥俄州立大学）；University of Notre Dame（圣母大学）；Washington University in St. Louis（圣路易斯华盛顿大学）；University of Chicago（芝加哥大学）；University of Illinois Chicago（伊利诺伊大学芝加哥分校）；Georgia Institute of Technology（佐治亚理工学院）；University of California, Los Angeles（加利福尼亚大学洛杉矶分校）；Carnegie Mellon University（卡内基梅隆大学）；Stony Brook University（石溪大学）；University of Connecticut（康涅狄格大学）；University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Boise State University（博伊西州立大学）
Rare Diseases, Common Dilemmas: LLMs Prioritize Equal Resource Distribution over Patient Benefit in Decision-Making
Harvard T.H. Chan School of Public Health（哈佛陈曾熙公共卫生学院）；Harvard Medical School（哈佛医学院）；Clalit Health Services（克拉利特医疗服务机构）；Harvard College（哈佛学院）；Harvard-MIT Program in Health Sciences and Technology（哈佛-麻省理工健康科学与技术项目）
Role-Play Paradox in Large Language Models: Reasoning Performance Gains and Ethical Dilemmas
University of Toronto（多伦多大学）；University of Alberta（阿尔伯塔大学）；University of Waterloo（滑铁卢大学）；Mcgill University（麦吉尔大学）；Renmin University of China（中国人民大学）；The Australian National University（澳大利亚国立大学）
University of Dhaka（达卡大学）；Daffodil International University（达福迪国际大学）；Islamic University of Technology（伊斯兰技术大学）；University of Central Florida（佛罗里达中央大学）；King Fahad University of Petroleum and Minerals（国王法赫德石油与矿物大学）；SDAIA - KFUPM Joint research Center for Artificial Intelligence（SDAIA-KFUPM人工智能联合研究中心）
Between a Rock and a Hard Place: The Tension Between Ethical Reasoning and Safety Alignment in LLMs
Department of Computer Science and Technology, Institute for AI, BNRist, Tsinghua University（计算机科学与技术系，人工智能研究院，BNRist，清华大学）；IDG/McGovern Institute for Brain Research, Tsinghua University（IDG/麦克戈文脑科学研究院，清华大学）；Chinese Institute for Brain Research (CIBR)（中国脑科学研究院（CIBR））；Shanghai Jiao Tong University（上海交通大学）；ByteDance（字节跳动）
Given the fast rise of increasingly autonomous artificial agents and robots, a key acceptability criterion will be the possible moral implications of their actions. In particular, intelligent persuasive systems (systems designed to influence humans via communication) constitute a highly sensitive topic because of their intrinsically social nature. Still, ethical studies in this area are rare and tend to focus on the output of the required action. Instead, this work focuses on the persuasive acts themselves (e.g. "is it morally acceptable that a machine lies or appeals to the emotions of a pers…
Are Dilemmas and Conflicts in LLM Alignment Solvable? A View from Priority Graph
The Hong Kong University of Science and Technology（香港科学与技术大学）；The Hong Kong University of Science and Technology (Guangzhou)（香港科学与技术大学（广州））；New York University（纽约大学）；National University of Singapore（新加坡国立大学）
Exploring and steering the moral compass of Large Language Models
Ernst Strüngmann Institute for Neuroscience（恩斯特·施特伦格曼神经科学研究所）；Max Planck Society（马克斯·普朗克学会）；Laboratory for Clinical Neuroscience（临床神经科学实验室）；Centre for Biomedical Technology（生物医学技术中心）；Universidad Politécnica de Madrid（马德里理工大学）
We test the hypothesis that language models trained with reinforcement learning from human feedback (RLHF) have the capability to "morally self-correct" -- to avoid producing harmful outputs -- if instructed to do so. We find strong evidence in support of this hypothesis across three different experiments, each of which reveal different facets of moral self-correction. We find that the capability for moral self-correction emerges at 22B model parameters, and typically improves with increasing model size and RLHF training. We believe that at this level of scale, language models obtain two capab…
Antisocial Analagous Behavior, Alignment and Human Impact of Google AI Systems: Evaluating through the lens of modified Antisocial Behavior Criteria by Human Interaction, Independent LLM Analysis, and AI Self-Reflection
