[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-116982-en":3,"doc-seo-116982-105":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},116982,8796095360427,"Lucas Martin","https://ap-avatar.wpscdn.com/davatar_994ba38a5ba835b3df7d355c54d3ed8d",8,"Research & Report","Causal scientific explanations from Machine Learning","Machine learning is increasingly used in scientific settings, yet it remains unclear whether such models can deliver genuine scientific explanations rather than only accurate predictions. The paper argues that when machine learning is employed for causal inference, it can support a positive route to causal scientific explanations. It also explains why standard models are unsuitable for causal inference without specialized techniques, and how extracted structural equation models can still fail when confounders, colliders, or physical-law violations distort causal graphs.","Causal scientific explanations from Machine Learning  \nStefan Buijsman∗  \nAccepted for publication in Synthese  \nAbstract  \nMachine learning is used more and more in scientific contexts, from the recent breakthroughs with AlphaFold2 in protein fold prediction to the use of ML in parametrization for large climate/astronomy models. Yet it is unclear whether we can obtain scientific explanations from such models. I argue that when machine learning is used to conduct causal inference we can give a new positive answer to this question. However, these ML models are purpose-built models and there are technical results showing that standard machine learning models cannot be used for the same typeof causal inference. Instead, there is a pathway to causal explanations from predictive ML models through new explainability techniques; specifically, new methods to extract structural equation models from such ML models. The extracted models are likely to suffer from issues though: they will often fail to account for confounders and colliders, as well as deliver simply incorrect causal graphs due to ML models tendency to violate physical laws such as the conservation of energy. In this case, extracted graphs are a starting point for new explanations, but predictive accuracy is no guarantee for good explanations.  \n1 Introduction  \nMachine learning models 1 are quickly gaining ground in scientific practice. A particular success is the use of deep learning model AlphaFold 2 to predict protein folding (Jumper et al. , 2021), but examples abound. There is, for example, usage of deep learning in climate models (Rasp et al. , 2018), astronomy (Agarwal et al. , 2012), and materials science (Schmidt et al. , 2019) . These machine learning models are primarily predictive models: they are used because they can very accurately predict outcomes (e.g. protein folds, or climate parameters) . An important exception is the case of neuroscience(Milkowski, 2013; Stinson, 2018; Piccinini, 2010) because of the direct modeling offered by machine learning models  \n∗ TU Delft, Jaffalaan 5, 2628BX, Delft, The Netherlands. Email: [s.n.r.buijsman@tudelft.nl](s.n.r.buijsman@tudelft.nl)  \n1 It should be noted here that the term ’machine learning’ has both narrow and broad interpretations. In a broad interpretation it is any computer method that solves a problem by fitting a function to data. In that case, simple models such as those based on linear regression count as machine learning. I follow the narrower definition of machine learning common in the literature discussed here, where the term is only applied to methods such as deep neural networks and random forest algorithms, which are distinguished by their use of a large number of parameters and non-linearity.  \nin that area. I will set aside that field here to focus on the other areas of science. There, it is not clear that opaque machine learning models, which predict well but where we do not understand why they arrive at a certain prediction (Das and Rad, 2020) can be used to arrive at scientific explanations (L´opez-Rubio and Ratti, 2021; Sre´ckovi´c et al. , 2021) . The most advanced machine learning tools, deep neural networks, present us with two difficulties: they are very complex (easily containing millions of parameters that are fine-tuned based on large data sets) and they lack internal representations that are legible to us (other models have explicit variables standing for e.g. physical quantities, deep neural networks have activation functions that respond to complex combinations of input features which we cannot interpret) . As a result, it is difficult to see how we can acquire causal explanations when using these complex models. There are worries that “if you do molecular biology with machine learning techniques, and if you want to have the best machine learning performances, then you cannot even in principle elaborate fully-fledged mechanistic explanations.”(L´opez-Rubio and Ratti, 2021, p.3152)  \nR","cbCaiuchDQEXWSN4","https://ap.wps.com/l/cbCaiuchDQEXWSN4","pdf",394633,1,19,"English","en",105,"# Introduction\n## Machine learning in scientific practice\n## Limits of predictive models for explanation\n## From predictive ML to causal explanation\n## Extracting structural equation models","[{\"question\":\"Why is it uncertain whether machine learning models can produce scientific explanations?\",\"answer\":\"Predictive machine learning models are designed for accurate outcomes but often lack legible internal representations, making it difficult to understand why predictions occur. The paper also notes technical obstacles for using standard ML models for causal inference.\"},{\"question\":\"What new positive route does the paper propose for causal scientific explanations?\",\"answer\":\"It argues that when machine learning is used to conduct causal inference, it can yield causal scientific explanations. The approach relies on explainability techniques that extract structural equation models from predictive ML models.\"},{\"question\":\"What problems can arise in the extracted causal models?\",\"answer\":\"Extracted graphs may fail to account for confounders and colliders, and can produce incorrect causal structures due to ML tendencies to violate physical laws, such as energy conservation. Predictive accuracy is therefore not a guarantee of good causal explanations.\"}]","Causal scientific explanations from Machine Learning | PDF",1785672950,48,{"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},"causal-scientific-explanations-from-machine-learning","",{"@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/causal-scientific-explanations-from-machine-learning/116982/",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","2026-08-02",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 is it uncertain whether machine learning models can produce scientific explanations?","Question",{"text":75,"@type":76},"Predictive machine learning models are designed for accurate outcomes but often lack legible internal representations, making it difficult to understand why predictions occur. The paper also notes technical obstacles for using standard ML models for causal inference.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What new positive route does the paper propose for causal scientific explanations?",{"text":80,"@type":76},"It argues that when machine learning is used to conduct causal inference, it can yield causal scientific explanations. The approach relies on explainability techniques that extract structural equation models from predictive ML models.",{"name":82,"@type":73,"acceptedAnswer":83},"What problems can arise in the extracted causal models?",{"text":84,"@type":76},"Extracted graphs may fail to account for confounders and colliders, and can produce incorrect causal structures due to ML tendencies to violate physical laws, such as energy conservation. Predictive accuracy is therefore not a guarantee of good causal explanations.","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":21,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},"General","general"]