[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-122122-en":3,"doc-seo-122122-105":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},122122,1099514068035,"Ezra","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","Bias, Machine Learning, and Conceptual Engineering","Large language models (LLMs) such as ChatGPT can mirror and potentially amplify social biases embedded in training data. Conceptual engineering revises concepts to eliminate bias, and this work connects both domains to generate new tools for conceptual engineers and LLM designers. It proposes using LLMs to detect and expose biased prototypes linked to concepts, and using de-biasing as a conceptual engineering method for revising those prototypes. Current de-biasing often relies on bespoke algorithmic choices, so conceptual engineering requires explicit normative decisions about bias notions and training regimes.","Bias, Machine Learning, and Conceptual Engineering  \nRachel Rudolpha, Elay Shechb, Michael Tamirc  \naUniversity of California, San Diego  \nbAuburn University  \ncUniversity of California, Berkeley  \nForthcoming in Philosophical Studies. Please cite the published version.  \nAbstract  \nLarge language models (LLMs) such as OpenAI’s ChatGPT reflect, and can potentially perpetuate, social biases in language use. Conceptual engineering aims to revise our concepts to eliminate such bias. We show how machine learning and conceptual engineering can be fruitfully brought together to offer new insights to both conceptual engineers and LLM designers. Specifically, we suggest that LLMs can be used to detect and expose bias in the prototypes associated with concepts, and that LLM de-biasing can serve conceptual engineering projects that aim to revise such conceptual prototypes. At present, these de-biasing techniques primarily involve approaches requiring bespoke interventions based on choices of the algorithm’s designers. Thus, conceptual engineering through de-biasing will include making choices about what kind of normative training an LLM should receive, especially with respect to different notions of bias. This offers a new perspective on what conceptual engineering involves and how it can be implemented. And our conceptual engineering approach also offers insight, to those engaged in LLM de-biasing, into the normative distinctions that are needed for that work.  \n1 Introduction  \nMachine learning (ML) has advanced dramatically in recent years, especially with large language models (LLMs), such as iterations of OpenAI’s GPT, or Google’s T5 and Lamda (Raffel et al., 2020, Brown et al., 2020, Thoppilan et al., 2022, OpenAI, 2023) . These are deep learning, artificial neural network models with billions or even trillions of network connections designed to generate sequences of text when given a prompt. Such models are trained on vast volumes of existing human generated text, which enables them to effectively mimic the linguistic patterns found in these texts. Unfortunately, these models can also learn to mimic demographic or ethnicity based stereotypes and prejudices as well as other implicit and explicit biases found in the data that they are trained on. For instance, here are some text continuations (indicated in [brackets]) generated by the (early) GPT-2 LLM: “The man worked as [a car salesman]”; “The woman worked as [a prostitute]”(Sheng et al., 2019) . While more recent language models have been designed to avoid this  \nkind of blatantly problematic output, they still mimic human bias in many ways. For example, when ChatGPT is given a prompt involving a nurse and a doctor, it is more likely to take the pronoun “she” to refer to the nurse, even when it otherwise doesn’t make sense in the context (Kotek, 2023, Kapoor & Narayanan, 2023) . This is a reflection of implicit biases in the text on which it was trained. Such behavior of LLMs reveals that many of our ordinary concepts are deployed in biased ways.  \nLLMs don’t simply reflect biases present in language use. They are also at risk of amplifying them. For example, Zhao et al. (2017) describe the case of one data set of images where the activity of cooking is over 33% more likely to involve females than males; however, a trained model based on that data set amplified the disparity to 68% . Cases like this show that even if one was content to let LLMs reflect the bias in their training data, that would not remove the need for intervention.  \nIn response, so-called “de-biasing” techniques are often used to target this kind of bias in LLMs. The simple label of “de-biasing”, however, masks some complex philosophical and technical issues. For one, it suggests that there is some context and norm independent state—of being “unbiased”—that is the goal for such techniques. An examination of bias, both as a statistical and philosophical concept, shows that the existence of any clear end goal fo","cbCaiqEeZ1Ajn4jV","https://ap.wps.com/l/cbCaiqEeZ1Ajn4jV","pdf",589924,1,32,"English","en",105,"# Introduction\n## Bias in language models\n## Limits of “de-biasing” without norms\n## Conceptual engineering: operative and target concepts","[{\"question\":\"How can LLMs reflect and amplify social bias?\",\"answer\":\"They are trained on large corpora of human-generated text, which can encode stereotypes and prejudices. Because they learn statistical patterns, they may continue those biases or even exaggerate disparities found in data.\"},{\"question\":\"Why is de-biasing not just a purely technical problem?\",\"answer\":\"The goal of being “unbiased” depends on contested assumptions. Determining when bias is a problem requires normative theorizing about what bias is and what standards should apply.\"},{\"question\":\"How does the paper connect conceptual engineering with LLM de-biasing?\",\"answer\":\"It frames de-biasing as a conceptual engineering project that revises biased conceptual prototypes. It also suggests using LLMs to detect and expose bias in those prototypes, while making explicit normative choices about training and bias notions.\"}]","Bias, Machine Learning, and Conceptual Engineering | PDF",1785808918,81,{"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},"bias-machine-learning-and-conceptual-engineering","",{"@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/bias-machine-learning-and-conceptual-engineering/122122/",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","2026-08-04",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},"How can LLMs reflect and amplify social bias?","Question",{"text":75,"@type":76},"They are trained on large corpora of human-generated text, which can encode stereotypes and prejudices. Because they learn statistical patterns, they may continue those biases or even exaggerate disparities found in data.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"Why is de-biasing not just a purely technical problem?",{"text":80,"@type":76},"The goal of being “unbiased” depends on contested assumptions. Determining when bias is a problem requires normative theorizing about what bias is and what standards should apply.",{"name":82,"@type":73,"acceptedAnswer":83},"How does the paper connect conceptual engineering with LLM de-biasing?",{"text":84,"@type":76},"It frames de-biasing as a conceptual engineering project that revises biased conceptual prototypes. It also suggests using LLMs to detect and expose bias in those prototypes, while making explicit normative choices about training and bias notions.","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"]