<\/p>\n
The models were even more likely to stereotype people by demographic group than the human participants in the original study. On the study\u2019s segregation scale, where 2 means every group has been completely confined to its own job niche, human participants scored 0.84. The models scored roughly 65% higher, with OpenAI\u2019s reasoning model o3 scoring 1.83, close to the maximum possible.<\/p>\n
That\u2019s because LLMs \u201creally are eager to create generalizations from limited data,\u201d says Ryan Liu, a PhD student at Princeton University and a coauthor of the study, which was published in a paper at ICML in Seoul in July. \u201cThat\u2019s literally a lot of what they\u2019re optimized for.\u201d Every decision-maker, human or machine, faces a trade-off between sticking with what worked before and trying something new that might work better\u2014a phenomenon psychologists call the \u201cexploration-exploitation dilemma.\u201d It\u2019s like choosing between a new restaurant and your reliable favorite.\u00a0<\/p>\n
Because LLMs are trained on math, coding, and science problems\u2014tasks that reward generalizing from just a few examples\u2014they can settle on a hunch too early. And the same instinct that helps LLMs crack logic puzzles also makes them quick to stereotype. In the experiment, newer models with higher reasoning capabilities, such as OpenAI\u2019s o3 and DeepSeek\u2019s R1, showed even stronger biases. When LLMs rush to generalize in social settings, \u201cthat\u2019s when things tend to go wrong,\u201d says Liu. OpenAI and Anthropic did not respond to requests for comment.<\/p>\n
The finding is especially relevant now that chatbots are gaining improved memory and personalization features, says Angelina Wang, a computer scientist at Cornell University who did not work on the study. When a chatbot draws on its previous conversation history, it can \u201cover-index on the same kinds of behaviors it\u2019s experienced before\u201d and form biases, she says. Simply having chatbots remember less isn\u2019t a fix, though, because users want chatbots to remember what they say. \u201cWe still are trying to figure out just the right amount that isn\u2019t too much or too little,\u201d says Wang.<\/p>\n
Telling the model to be fair didn\u2019t change its behavior much. \u201cEither it can\u2019t put these values into action or that process is being submerged under the tendency to try to optimize for the goal of getting the most correct hires,\u201d says Liu. But promising the models an additional bonus for diverse hiring made them far less biased. The trick, then, is to design goals that \u201cincorporate desirable social values in order to make the large language model act in socially desirable ways,\u201d says Liu.<\/p>\n<\/p><\/div>\n
#humans #form #biases #hiring<\/p>\n","protected":false},"excerpt":{"rendered":"
The models were even more likely to stereotype people by demographic group than the human…<\/p>\n","protected":false},"author":1,"featured_media":17858,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7],"tags":[12028,12030,506,12031,6213,12029],"class_list":["post-17857","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-stories","tag-ai-is-more-likely-than-humans-to-form-biases-when-hiring-ai-doesnt-just-learn-stereotypes-from-its-training-it-can-cook-up-new-ones","tag-biases","tag-form","tag-hiring","tag-humans","tag-too"],"featured_image_urls":{"full":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b.jpg",1200,600,false],"thumbnail":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b-150x150.jpg",150,150,true],"medium":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b-300x150.jpg",300,150,true],"medium_large":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b-768x384.jpg",640,320,true],"large":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b-1024x512.jpg",640,320,true],"1536x1536":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b.jpg",1200,600,false],"2048x2048":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b.jpg",1200,600,false],"covernews-featured":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b-1024x512.jpg",1024,512,true],"covernews-medium":["https:\/\/news678.top\/wp-content\/uploads\/2026\/07\/ai-slant1b-540x340.jpg",540,340,true]},"author_info":{"display_name":"admin","author_link":"https:\/\/news678.top\/?author=1"},"category_info":"Stories<\/a>","tag_info":"Stories","comment_count":"0","_links":{"self":[{"href":"https:\/\/news678.top\/index.php?rest_route=\/wp\/v2\/posts\/17857","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/news678.top\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/news678.top\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/news678.top\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/news678.top\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=17857"}],"version-history":[{"count":0,"href":"https:\/\/news678.top\/index.php?rest_route=\/wp\/v2\/posts\/17857\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/news678.top\/index.php?rest_route=\/wp\/v2\/media\/17858"}],"wp:attachment":[{"href":"https:\/\/news678.top\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=17857"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/news678.top\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=17857"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/news678.top\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=17857"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}