{"id":641393,"date":"2026-06-16T09:55:35","date_gmt":"2026-06-16T09:55:35","guid":{"rendered":"https:\/\/www.newsbeep.com\/uk\/641393\/"},"modified":"2026-06-16T09:55:35","modified_gmt":"2026-06-16T09:55:35","slug":"ai-learned-the-rules-of-the-universe-and-that-became-a-problem","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/uk\/641393\/","title":{"rendered":"AI Learned the Rules of the Universe and That Became a Problem"},"content":{"rendered":"

\"Quijote<\/a>Two images from the Quijote simulations used in this study. The panels show the same region of the Universe, but in different cosmological models. The top image corresponds to the standard \u039bCDM model, while the bottom image shows a universe with massive neutrinos and modified gravity. The differences are subtle, but they reveal how changes in the underlying physics can affect the formation and distribution of cosmic structures. Credit: Francisco Villaescusa-Navarro<\/p>\n

AI could accelerate the hunt for new physics, but sometimes it knows too much to see what\u2019s right in front of it.<\/p>\n

Artificial intelligence could make it much cheaper and faster to search for new laws of physics, according to a new study published in the Journal of Cosmology and Astroparticle Physics (JCAP). But the research also points to an unexpected downside. In some situations, AI can become so dependent on its previous training that it struggles to recognize genuinely new phenomena.<\/p>\n

AI has become an important tool in cosmology, helping researchers analyze enormous amounts of data about the universe. Yet investigating ideas that go beyond the current standard cosmological model, known as \u039bCDM, remains an extremely expensive computational challenge.<\/p>\n

While \u039bCDM successfully explains many observed features of the universe, including its expansion and the large-scale distribution of galaxies, scientists do not believe it tells the whole story. Recent observations suggest that phenomena such as massive neutrinos, modified gravity, and evolving dark energy could reveal physics that lies beyond the current model.<\/p>\n

Exploring these possibilities requires researchers to generate vast numbers of detailed simulations of virtual universes, each based on different physical assumptions. Producing these simulations often demands enormous computing power and time.<\/p>\n

Transfer Learning Offers a Faster Route<\/p>\n

The researchers investigated whether a machine learning approach called transfer learning could reduce that burden.<\/p>\n

Transfer learning allows an AI system to apply knowledge gained from one task to help it learn another task more efficiently. Rather than starting from scratch, the AI builds on what it has already learned.<\/p>\n

For this study, the team first trained a neural network using simulations based on \u039bCDM. This initial training process, known as pretraining, gave the AI a foundation before it was exposed to more complex cosmological models that include possible new physics.<\/p>\n

\u201cIt\u2019s basically a shortcut,\u201d explains Adrian Bayer a cosmologist at the Flatiron Institute and Princeton University, co-author of the study. \u201cUsually people train the AI directly on the most computationally expensive simulations. What we do instead is first use simpler and less expensive \u039bCDM simulations to give the AI an idea of what\u2019s happening, and only afterward move to the more complex models.\u201d<\/p>\n

Bayer compares the process to learning from textbooks. \u201cYou first read a basic book to get an idea of the knowledge,\u201d says Bayer, \u201cand then move to the really complicated book.\u201d<\/p>\n

According to Veena Krishnaraj, an undergraduate student at Princeton University and the paper\u2019s first author, this approach prevents the AI from having to \u201cdigest everything at once.\u201d<\/p>\n

The strategy proved highly effective. In some cases, transfer learning reduced the number of costly simulations required by more than a factor of ten.<\/p>\n

When Prior Knowledge Becomes a Problem<\/p>\n

The study also revealed a less obvious challenge known as negative transfer.<\/p>\n

Using Bayer\u2019s textbook analogy, imagine a medical student learning from introductory materials and later encountering a rare disease that resembles a common illness. Existing knowledge is usually helpful, but it can sometimes lead to the wrong conclusion.<\/p>\n

A similar problem can arise in AI systems. Certain signals produced by new physics can look very similar to patterns the AI already learned from the standard cosmological model. When that happens, the AI may interpret the new information through the lens of its earlier training, making it more difficult to recognize something truly different.<\/p>\n

The researchers saw this effect while studying simulations that included massive neutrinos. Some of the observable consequences of neutrino mass closely resemble changes associated with an existing \u039bCDM parameter called \u03c38, which measures how strongly matter clusters throughout the universe.<\/p>\n

Because the two effects can appear so similar, the pretrained neural network initially had trouble telling them apart.<\/p>\n

\u201cThe negative transfer is not random. It is driven by underlying physical degeneracies in the model,\u201d says Krishnaraj. In other words, different physical parameters can create nearly identical observable signatures, making it difficult for the AI to correctly separate them. \u201cSo this is something we need to be aware of and try to mitigate,\u201d she concludes.<\/p>\n

Promise and Risks for Future Cosmology<\/p>\n

The findings illustrate both the benefits and potential pitfalls of applying foundation model strategies to physics. These approaches are conceptually similar to the techniques used in modern generative AI systems and large language models.<\/p>\n

As the authors note in the paper, pretraining can speed up inference, \u201cbut may also hinder learning new physics.\u201d<\/p>\n

So far, the method has only been tested using simulations. However, the researchers believe it provides an important foundation for future applications involving real astronomical observations.<\/p>\n

That could become increasingly valuable as next-generation cosmological surveys begin producing unprecedented volumes of high-precision data about the universe. If used carefully, transfer learning could help scientists analyze that information far more efficiently while continuing the search for physics beyond the Standard Model.<\/p>\n

The paper, \u201cTransfer Learning Beyond the Standard Model,\u201d by Veena Krishnaraj, Adrian E. Bayer, Christian Kragh Jespersen, and Peter Melchior, is now available in JSTAT.<\/p>\n

Never miss a breakthrough: Join the SciTechDaily newsletter.<\/a>
Follow us on Google<\/a> and Google News<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"Two images from the Quijote simulations used in this study. The panels show the same region of the…\n","protected":false},"author":2,"featured_media":641394,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[554,733,4308,11745,9025,57020,86,56,54,55],"class_list":["post-641393","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-astrophysics","tag-cosmology","tag-sissa","tag-technology","tag-uk","tag-united-kingdom","tag-unitedkingdom"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/641393","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=641393"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/641393\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/641394"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=641393"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=641393"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=641393"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}