[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-145551-105":59,"doc-detail-145551-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","ai-werewolf-agent-with-reasoning-using-role-patterns-and-heuristics","AI Werewolf Agent with Reasoning Using Role Patterns and Heuristics","","AIWolf project enables contests where automatic agents play Werewolf (“Mafia”), an imperfect-information game with extremely limited revealed information beyond each agent’s own role. As a result, role probability estimates are not sufficient to provide theoretically solid reasoning, and explanations based only on probabilities are hard to justify convincingly. The proposed agent targets inference reasoning by focusing on the seer roles (genuine and fake) and defining role patterns from their utterances as the game progresses.",{"@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/ai-werewolf-agent-with-reasoning-using-role-patterns-and-heuristics/145551/",{"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/ai-werewolf-agent-with-reasoning-using-role-patterns-and-heuristics/145551.png","ImageObject",300,407,{"name":92,"@type":93},"Bill Black","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-28","2026-08-26",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",13,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"Why is reasoning difficult in the Werewolf game compared with perfect-information games?","Question",{"text":113,"@type":114},"Werewolf is an imperfect-information game where players see very limited information besides their own role, so confident theoretical inference from role probabilities alone is hard.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What is the main idea behind the proposed AIWolf agent’s reasoning?",{"text":118,"@type":114},"The agent infers roles by tracking the game’s progress and using role patterns defined from utterances of the genuine and fake seers.",{"name":120,"@type":111,"acceptedAnswer":121},"How does the method use seer-related assumptions in the five-player setting?",{"text":122,"@type":114},"It assumes that when two players claim to be seers, they correspond to a genuine seer and a possessed, then builds a reasoning table covering possible inspection outcomes.","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},145551,1787724697,{"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":24,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":130,"read_time":105},24189269381491,"https://ap-avatar.wpscdn.com/avatar/160000cf11732dd8392?x-image-process=image/resize,m_fixed,w_180,h_180&k=1788146458752108895","AI Werewolf Agent with Reasoning Using Role Patterns and Heuris  \ntics  \nIssei Tsunoda Yoshinobu Kano  \nFaculty of Informatics, Shizuoka University, Hamamatsu, Shizuoka, Japan  \n[itsunoda@kanolab.net](itsunoda@kanolab.net), [kano@inf.shizuoka.ac.jp](kano@inf.shizuoka.ac.jp)  \nAbstract  \nThe AIWolf project has been holding contests for these years to play the Werewolf game (“Mafia”) by automatic agents. A difficulty of the Werewolf game is that the game is an imperfect information game, very small limited amount of information is shown to players, other than the player’sown role information. Therefore, inference of probabilities for each player agent's role could not be confident theoretically, difficult to utter appropriate reasons when simply based on the probabilities. Focusing on a genuine seer and a fake seer, we implemented our player agent system that can make inferences depending on the progress of the game, defining role patterns based on the utterances of the genuine and fake seers.  \n1 Introduction  \nThe AlphaGO [1] system defeated the human champion player in Go. However, AI game player is still far from being successful in the Werewolf game that requires complex communications, in addition to the nature of an imperfect information game, while Go is a perfect information game. Playing the Werewolf game would be the next grand research challenge for the AI players.  \nIn order to promote such a research challenge, the AIWolf project [2] has been holding competitions every year to play the Werewolf game automatically. We describe our Werewolf player agent system which participated the AIWolfDial 2019 shared task (the natural language division of the 2019 competition of AIWolf) [3] . Our AIWolf agents use the Japanese language, while the shared task organizers automatically translate the system I/O to connect with English agents.  \n1.1 The Werewolf Game  \nWe briefly explain the rules of the werewolf game in this section. Before starting a game, each player is assigned a hidden role from the game master (a server system in case of the AIWolf competition) . The most common roles are “villager” and “werewolf”. Each role (and a player of that role) belongs either to a villager team or a werewolf team. The goal of a player is for any of the team members to survive, not necessarily the player him/herself.  \nWhile there are many variations of the Werewolf game exists, we only explain the AIWolfDial 2019 shared task setting in this paper.  \nThere are other roles than the villager and the werewolf: a seer and a possessed. A seer belongs to the villager team, who has a special talent to“divine” a specified player to know whether the player is a human or a werewolf; the divine result is notified the seer only. A possessed belongs to the werewolf team but if he/her is divined by a seer, then its result is human.  \nA game consists of “days”, and a “day” consists of “daytime” and “night”. During the daytime phase, each player talks freely. At the end of the daytime, a player will be executed by votes of all of the remained players. In the night phase, special role players use their abilities: a werewolf can attack and kill a player, and a seer can divine a player. The victory condition of the villager team is to execute all werewolfs (a possessed may be alive), and the victory condition of the werewolf team is to make the number of villager team less than the number of werewolf team. A game in the AIWolfDial 2019 shared task have five players: a seer, a werewolf, a possessed, and two villagers.  \nIn the shared task, Day 0 does not start games but conversations e.g. greetings. A daytime consists of several turns; a turn is a synchronized talks  \nof agent, i.e. the agents cannot refer to other agents’ talks ofthe same turn.  \nAn AIWolf agent communicates with an AIWolf server to perform a game. Other than vote, divine, and attack actions, an agent communicatesin natural language only. An agent may insert an anchor symbol (e.g.“>>Agent[","cbCaimXjSNcvz8Y8","https://ap.wps.com/l/cbCaimXjSNcvz8Y8","pdf",489898,"English","# Abstract\n# Introduction\n## The Werewolf Game\n# Related Works\n# Method\n## Reasoning table of inspections and role combinations","[{\"question\":\"Why is reasoning difficult in the Werewolf game compared with perfect-information games?\",\"answer\":\"Werewolf is an imperfect-information game where players see very limited information besides their own role, so confident theoretical inference from role probabilities alone is hard.\"},{\"question\":\"What is the main idea behind the proposed AIWolf agent’s reasoning?\",\"answer\":\"The agent infers roles by tracking the game’s progress and using role patterns defined from utterances of the genuine and fake seers.\"},{\"question\":\"How does the method use seer-related assumptions in the five-player setting?\",\"answer\":\"It assumes that when two players claim to be seers, they correspond to a genuine seer and a possessed, then builds a reasoning table covering possible inspection outcomes.\"}]","AI Werewolf Agent with Reasoning Using Role Patterns and Heuristics | PDF"]