[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-119389-en":3,"doc-seo-119389-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},119389,1099514067438,"River Wang","https://ap-avatar.wpscdn.com/avatar/100002539ee87300030?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780474512215547542",8,"Research & Report","The Limits of Machine Learning Models of Misinformation","Judgments of misinformation are grounded in the informational preferences of the communities that make them, yet informational norms evolve and create distribution shifts that undermine the adequacy of machine-learning models. The work identifies five kinds of distribution shifts and evaluates three improvement strategies: larger static training sets, social engineering, and dynamic sampling. It argues the first is inadequate, the second unethical, and dynamic sampling superior. Overall prospects are limited because both epistemic and non-epistemic values are hard to operationalize dynamically in machine code, leaving such systems closer to recommender systems than truth detectors.","AI & SOCIETY  \n[https://doi.org/10.1007/s00146-025-02324-8](https://doi.org/10.1007/s00146-025-02324-8)  \nThe Limits of Machine Learning Models of Misinformation  \nAdrian K. Yee1  \nReceived: 26 February 2025 / Accepted: 14 March 2025 © The Author(s) 2025  \nAbstract  \nJudgments of misinformation are made relative to the informational preferences of the communities making them. However, informational standards change over time, inducing distribution shifts that threaten the adequacy of machine learning models of misinformation. After articulating five kinds of distribution shifts, three solutions for enhancing success are discussed: larger static training sets, social engineering, and dynamic sampling. I argue that given the idiosyncratic ontology of misinformation, the first option is inadequate, the second is unethical, and thus the third is superior. However, I conclude that the prospects for machine learning models of misinformation are far weaker than most have presupposed, given that both epistemic and non-epistemic values are difficult to operationalize dynamically in machine code, rendering them surprisingly at most a species of recommender systems rather than literal truth detectors.  \nKeywords Misinformation · machine learning · philosophy of social science · philosophy of science · social epistemology  \n1 Introduction  \nMisinformation has recently been a topic of discussion in the intersection of philosophy of science and artificial intelligence via ‘machine learning models of misinformation’(MMMs): the usage of computers to automate the process of identifying misinformation in text and images (Yee 2023a, b; Harris 2024) . MMMs are of central interest to citizens and governments, with some advocating their usage for regulating societies’ information ecosystems, such as in Malta (Cassar 2023) and Saudi Arabia (Altiabi 2022) . However, the overall adequacy of MMMs remains highly contested, despite scoring highly on standard machine learning metrics of accuracy, precision, and recall (Cf. Khan et al. 2021) . Given how widespread misinformation is alleged to be, and given the high stakes that governments, citizens, and private industry have in ensuring that information flowing through the internet and other sources is as high quality as possible, it is important to reflect on the increasing usage of MMMsand analyze their methodological foundations.  \nThis paper focuses on a neglected methodological problem in MMM research. Judgments of misinformation are  \n* Adrian K. Yee [adrianyee@ln.edu.hk](adrianyee@ln.edu.hk)  \n1 Lingnan University, Hong Kong Catastrophic Risk Centre, Tuen Mun, China  \nalways made relative to a background epistemic community whose informational norms are in sufficient equilibrium: an act of communication is judged as misinformation whenever violations of epistemic conventions surrounding usage of that kind of information are perceived to have been made, typically concerning the truth or misleadingness of a piece of information. However, empirical and historical studies have shown that sufficient equilibrium is often merely transient concerning prevailing informational norms, rendering these systems the result of stochastic processes whose nonstationarity is particularly difficult to model. This suggests that when one tries to automate manual judgments of misinformation in the form of MMMs, any ostensible successes of test data relative to training data become underwhelming when there are strong empirical reasons to be skeptical of the success of real-world applications due to what been called the MMM problem of ‘temporal generalizability’(Stepanova and Ross 2013) . While the problem of distribution shifts in machine learning more generally has been studied in great detail (Gama et al. 2014), the context of MMMs remains comparatively under studied with unclear causes and solutions for such shifts.  \nAs I will argue in this paper, social and computer scientists ought to reconsider what the appropriate ","cbCaibwYSwGbvbDG","https://ap.wps.com/l/cbCaibwYSwGbvbDG","pdf",716136,1,14,"English","en",105,"# Introduction\n## Motivations and contested adequacy of MMMs\n## The MMM problem of temporal generalizability\n# Understanding judgments of misinformation\n## Four categories of informational disputes\n## Epistemic and non-epistemic value disagreements\n# Distribution shifts in misinformation judgments\n# Improving empirical adequacy of MMMs\n## Larger static training sets\n## Social engineering and dynamic sampling","[{\"question\":\"Why can machine learning models of misinformation fail in real-world use?\",\"answer\":\"Because judgments depend on community informational norms that change over time, producing distribution shifts that make model adequacy degrade outside the conditions of training and tests.\"},{\"question\":\"What are the three proposed solutions to improve success for MMMs?\",\"answer\":\"The paper discusses larger static training sets, social engineering, and dynamic sampling, then evaluates them ethically and methodologically.\"},{\"question\":\"According to the paper, are MMMs reliable as literal truth detectors?\",\"answer\":\"No. The argument concludes that due to the idiosyncratic ontology of misinformation and difficulties operationalizing epistemic and non-epistemic values dynamically, MMMs are at best more like recommender systems than literal truth detectors.\"}]","The Limits of Machine Learning Models of Misinformation | PDF",1785724058,35,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"the-limits-of-machine-learning-models-of-misinformation","",{"@graph":36,"@context":85},[37,54,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/the-limits-of-machine-learning-models-of-misinformation/119389/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":62,"encodingFormat":61,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-03",true,{"@type":65,"interactionType":66,"userInteractionCount":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why can machine learning models of misinformation fail in real-world use?","Question",{"text":75,"@type":76},"Because judgments depend on community informational norms that change over time, producing distribution shifts that make model adequacy degrade outside the conditions of training and tests.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What are the three proposed solutions to improve success for MMMs?",{"text":80,"@type":76},"The paper discusses larger static training sets, social engineering, and dynamic sampling, then evaluates them ethically and methodologically.",{"name":82,"@type":73,"acceptedAnswer":83},"According to the paper, are MMMs reliable as literal truth detectors?",{"text":84,"@type":76},"No. The argument concludes that due to the idiosyncratic ontology of misinformation and difficulties operationalizing epistemic and non-epistemic values dynamically, MMMs are at best more like recommender systems than literal truth detectors.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]