[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-145548-105":59,"doc-detail-145548-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","learning-to-discuss-strategically-a-case-study-on-one-night-ultimate-werewolf-neurips-2024","Learning to Discuss Strategically - A Case Study on One Night Ultimate Werewolf - NeurIPS 2024","","Communication is a core mechanism for exchanging information and beliefs, yet many agent approaches using large language models still miss explicit control over discussion tactics required in communication games. This paper studies One Night Ultimate Werewolf as a setting where role changes increase uncertainty and strategic complexity. The work analyzes Perfect Bayesian Equilibria in scenarios with and without discussion, showing that discussion reshapes players’ utilities by altering beliefs. It proposes an RL-instructed language agent framework and validates effectiveness across multiple game settings.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/learning-to-discuss-strategically-a-case-study-on-one-night-ultimate-werewolf-neurips-2024/145548/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/learning-to-discuss-strategically-a-case-study-on-one-night-ultimate-werewolf-neurips-2024/145548.png","ImageObject",300,407,{"name":92,"@type":93},"Đào","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-04","2026-08-26",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",12,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"Why is strategic discussion important in One Night Ultimate Werewolf (ONUW)?","Question",{"text":113,"@type":114},"Role changes and limited actions during night and one daytime discussion increase uncertainty. Players must infer final roles from statements while managing misleading or concealed information.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What theoretical results does the paper provide?",{"text":118,"@type":114},"It analyzes the existence of Perfect Bayesian Equilibria in ONUW under two scenarios: with discussion and without discussion, including how beliefs determine utilities when discussion is present.",{"name":120,"@type":111,"acceptedAnswer":121},"How does the proposed RL-instructed framework improve performance?",{"text":122,"@type":114},"It trains a discussion policy using reinforcement learning to select tactics based on current observations. Experiments show the framework approximates PBEs more closely and generalizes better than direct prompting.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},145548,1787724674,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":130,"read_time":145},1374402968488,"https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0","Learning to Discuss Strategically: A Case Study on One Night Ultimate Werewolf  \nXuanfa Jin∗, 1 ,3 , Ziyan Wang∗ ,2 , Yali Du†,2 , Meng Fang4 , Haifeng Zhang†, 1 ,3 ,5 , Jun Wang†,6  \n1Institute of Automation, Chinese Academy of Sciences, 2 Cooperative AI Lab, Department of Informatics, King’s College London, 3 School of Artificial Intelligence, University of Chinese Academy of Sciences, 4University of Liverpool, 5Nanjing Artificial Intelligence Research of IA, 6AI Centre, Department of Computer Science, UCL  \nAbstract  \nCommunication is a fundamental aspect of human society, facilitating the exchange of information and beliefs among people. Despite the advancements in large language models (LLMs), recent agents built with these often neglect the control over discussion tactics, which are essential in communication scenarios and games.  \nAs a variant of the famous communication game Werewolf, One Night Ultimate Werewolf (ONUW) requires players to develop strategic discussion policies due to the potential role changes that increase the uncertainty and complexity of the game. In this work, we first present the existence of the Perfect Bayesian Equilibria (PBEs) in two scenarios of the ONUW game: one with discussion and one without. The results showcase that the discussion greatly changes players’ utilities by affecting their beliefs, emphasizing the significance of discussion tactics. Based on the insights obtained from the analyses, we propose an RL-instructed language agent framework, where a discussion policy trained by reinforcement learning (RL) is employed to determine appropriate discussion tactics to adopt. Our experimental results on several ONUW game settings demonstrate the effectiveness and generalizability of our proposed framework. The project page of our paper:  \n[one-night-ultimate-werewolf.github.io](one-night-ultimate-werewolf.github.io).  \n1 Introduction  \nMany games such as StarCraft [1, 2], Diplomacy [3, 4] can approximate various fundamental issues in real life, and studying these games contributes to a better understanding of the functioning of our society [5] . Designing artificial intelligence (AI) agents that can play these games well has attracted a lot of attention [6, 2, 3] . Fortunately, recent large language models (LLMs) have demonstrated significant potential in constructing intelligent agents in numerous tasks [7–10] due to their impressive reasoning and emergent generalization abilities [11–13] . Moreover, LLM-based agents have achieved approaching or even surpassing human performance in games like Chess[14], Minecraft [15–17], Avalon [18–20] and Werewolf [21–23], etc.  \nIn the game Werewolf [24], the hidden roles and uncertain discussions greatly influence the gameplay, making it challenging for players. Compared to it, the One Night Ultimate Werewolf (ONUW) game adds roles that can change the roles of players, and all players only have one nighttime to take action  \nand one daytime to discuss and vote. Therefore, the key challenge of the ONUW game is to deduce ∗Equal contribution. Author ordering is determined by coin flip.  \n†Correspondence to: 〈[yali.du@kcl.ac.uk](yali.du@kcl.ac.uk)〉, 〈[haifeng.zhang@ia.ac.cn](haifeng.zhang@ia.ac.cn)〉, 〈[jun.wang@cs.ucl.ac.uk](jun.wang@cs.ucl.ac.uk)〉 .  \n38th Conference on Neural Information Processing Systems (NeurIPS 2024) .  \nand distinguish the final roles of all players from their statements. Meanwhile, due to the sequential execution of actions at night, the information obtained by the former player might be not reliable, leading to the rise of reasoning difficulty and uncertainty. For example, Seer and Robber are two roles in the ONUW game with special abilities, and the Seer saw a player was a Werewolf at night. However, if the Robber switched that player’s role later on, the information obtained by Seer becomes invalid. So players need to discuss strategically, leveraging others’ statements and their own prior knowledge to guide other players to re","cbCaiogMkePdpMAz","https://ap.wps.com/l/cbCaiogMkePdpMAz","pdf",778823,38,"English","# Introduction\n## Communication games and LLM-based agents\n## One Night Ultimate Werewolf as a Bayesian game\n## Perfect Bayesian Equilibria and discussion scenarios","[{\"question\":\"Why is strategic discussion important in One Night Ultimate Werewolf (ONUW)?\",\"answer\":\"Role changes and limited actions during night and one daytime discussion increase uncertainty. Players must infer final roles from statements while managing misleading or concealed information.\"},{\"question\":\"What theoretical results does the paper provide?\",\"answer\":\"It analyzes the existence of Perfect Bayesian Equilibria in ONUW under two scenarios: with discussion and without discussion, including how beliefs determine utilities when discussion is present.\"},{\"question\":\"How does the proposed RL-instructed framework improve performance?\",\"answer\":\"It trains a discussion policy using reinforcement learning to select tactics based on current observations. Experiments show the framework approximates PBEs more closely and generalizes better than direct prompting.\"}]","Learning to Discuss Strategically - A Case Study on One Night Ultimate Werewolf - NeurIPS 2024 | PDF",96]