AI数字孪生难以预测人类行为，产生'哈哈镜'扭曲

September 4, 2026 report AI digital twins struggle to predict human behavior, creating 'funhouse mirror' distortions by Paul Arnold, Phys.org Paul Arnold Author Meet our staff & contributors Learn about our editorial standards edited by Gaby Clark, reviewed by Robert Egan Gaby Clark Scientific Editor Meet our editorial team Behind our editorial process Robert Egan Senior Editor Meet our editorial team Behind our editorial process Editors' notes This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility: fact-checked peer-reviewed publication trusted source proofread The GIST Add as preferred source Credit: Image generated by the editorial team using AI for illustrative purposes. While many fear artificial intelligence will replace humans, using AI to take over some human roles has benefits. Companies can use the technology to conduct surveys and polls, while behavioral scientists can run experiments on digital twins to gather faster insights without risking harm or distress to real participants. But there's a huge snag. Today's digital twins are just not up to it. New research published in the journal Science Advances found that digital twins seriously misrepresent human behavior. What is a digital twin? A digital twin is a large language model built from a person's past data, such as demographics, personality traits and past choices. Digital twins are designed to predict how that person would think or act. Advocates even suggest they could one day negotiate or make choices on a person's behalf. But this latest paper is giving the field a major reality check. That reality check came in the form of a massive study involving 19 separate experiments led by Tianyi Peng of Columbia University in New York. Researchers compared the responses of 1,784 real people with those of their AI twins to see how accurately the models could predict human choices. The topics they were quizzed on included hiring decisions, political attitudes, privacy choices and news consumption. Mega-study overview. Credit: Science Advances (2026). DOI: 10.1126/sciadv.aeh8260 Each digital twin was created by feeding more than 500 prior answers from each individual into a base AI model. So how well did they do? Not at all well. "The digital twins of today may best be described as funhouse mirrors that systematically distort human behavior," the study authors wrote. The answers were less varied than those of their human counterparts. Overall, when it came to accurately predicting human choices, the twins performed only slightly better than the same AI without any of the person's past data. And there were five specific reasons for this, according to the paper. Five ways AI gets us wrong First, the twins did not do a good enough job of capturing what made each person different. Their answers were less varied than those of real people, making them too similar to one another. They also relied too heavily on stereotypes, often falling back on broad demographic traits rather than the characteristics of the individual. Another issue is ideological bias. The twins tended to be more optimistic than real people about human behavior and technology. They also represented some groups better than others, performing more accurately for people with higher incomes and levels of education. Finally, the twins were too rational. Compared with real people, they tended to display greater knowledge and were more likely to choose the rational answer, even when the person they were supposed to represent did not. And the bottom line, according to the team? "Digital twins are not yet 'ready for primetime.'" Written for you by our author Paul Arnold, edited by Gaby Clark, and fact-checked and reviewed by Robert Egan—this article is the result of careful human work. We rely on readers like you to keep independent science journalism alive. If this reporting matters to you, please consider a donation (especially monthly). You'll get an ad-free account as a thank-you. Publication details Tianyi Peng et al, Digital twins are funhouse mirrors: Five systematic distortions, Science Advances (2026). DOI: 10.1126/sciadv.aeh8260 Journal information: Science Advances Key concepts Large language modelsAI alignment Who's behind this story? Paul Arnold BSc Biology from University of London. BBC documentary producer with world travel experience. Freelances from southern Spain. Full profile → Gaby Clark MA in English, copy editor since 2021 with experience in higher education and health content. Dedicated to trustworthy science news. Full profile → Robert Egan Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile → © 2026 Science X Network Citation: AI digital twins struggle to predict human behavior, creating 'funhouse mirror' distortions (2026, September 4) retrieved 8 September 2026 from https://techxplore.com/news/2026-09-ai-digital-twins-struggle-human.html This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only. Your digital twin might save your life Dec 9, 2025 A virtual copy of your brain? Scientists say it's closer than you think May 1, 2026 Fake models for fast fashion? What AI clones mean for our jobs—and our identities Apr 24, 2025 A 'digital twin' of your brain could predict mental health issues, and slow cognitive decline Oct 14, 2025 Revealing the hidden logic behind AI's judgments of people Apr 13, 2026 Digital brain twins hinge on what living brains reveal, not computing power alone Sep 1, 2026 Memristor chip breaks the capacity limit of brain-inspired associative memory 5 hours ago Invention may bring smart bionic eye into sight 4 hours ago Whole-body expansion and contraction make a robot seem more alive 6 hours ago A tiny crystal tilt may bring greener magnetic memory closer to everyday devices 6 hours ago Tracking energy flow reveals voltage pattern that slashes waste heat by over 60% 8 hours ago Editorial proposes limits on generative AI's role in scientific research 12 hours ago