{"id":1975,"date":"2025-10-23T11:44:17","date_gmt":"2025-10-23T10:44:17","guid":{"rendered":"https:\/\/babel.isa.uma.es\/kipr\/?p=1975"},"modified":"2025-10-23T11:44:17","modified_gmt":"2025-10-23T10:44:17","slug":"a-quantitative-demonstration-based-on-mdps-of-the-increasing-need-of-a-world-model-learnt-or-given-as-the-complexity-of-the-task-and-the-performance-of-the-agent-increase","status":"publish","type":"post","link":"https:\/\/babel.isa.uma.es\/kipr\/?p=1975","title":{"rendered":"A quantitative demonstration based on MDPs of the increasing need of a world model (learnt or given) as the complexity of the task and the performance of the agent increase"},"content":{"rendered":"\n<h4 class=\"wp-block-heading\">Jonathan Richens, David Abel, Alexis Bellot, Tom Everitt,  <strong>General agents contain world models,<\/strong>  arXiv cs:AI, Sep. 2025, <a href=\"https:\/\/arxiv.org\/abs\/2506.01622v4\">arXiv:2506.01622<\/a>.<\/h4>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p>Are world models a necessary ingredient for flexible, goal-directed behaviour, or is model-free learning sufficient? We provide a formal answer to this question, showing that any agent capable of generalizing to multi-step goal-directed tasks must have learned a predictive model of its environment. We show that this model can be extracted from the agent&#8217;s policy, and that increasing the agents performance or the complexity of the goals it can achieve requires learning increasingly accurate world models. This has a number of consequences: from developing safe and general agents, to bounding agent capabilities in complex environments, and providing new algorithms for eliciting world models from agents. \n<\/p>\n<\/blockquote>\n","protected":false},"excerpt":{"rendered":"<p>Jonathan Richens, David Abel, Alexis Bellot, Tom Everitt, General agents contain world models, arXiv cs:AI, Sep. 2025, arXiv:2506.01622. Are world <span class=\"ellipsis\">&hellip;<\/span> <span class=\"more-link-wrap\"><a href=\"https:\/\/babel.isa.uma.es\/kipr\/?p=1975\" class=\"more-link\"><span>Read More &rarr;<\/span><\/a><\/span><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[84],"tags":[426,205,493,568],"class_list":["post-1975","post","type-post","status-publish","format-standard","hentry","category-reinforcement-learning-in-ai","tag-internal-model","tag-model-based-reinforcement-learning","tag-model-free-reinforcement-learning","tag-probabilistic-model"],"_links":{"self":[{"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=\/wp\/v2\/posts\/1975"}],"collection":[{"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1975"}],"version-history":[{"count":1,"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=\/wp\/v2\/posts\/1975\/revisions"}],"predecessor-version":[{"id":1976,"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=\/wp\/v2\/posts\/1975\/revisions\/1976"}],"wp:attachment":[{"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1975"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1975"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/babel.isa.uma.es\/kipr\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1975"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}