Summary: A new study utilizes Koopman operator learning to prove that certain complex, chaotic systems have fundamental predictability limits that cannot be overcome by infinite training data. By designing adversarial systems to map where machine learning models collapse, the team explained the mathematical root causes of LLM hallucinations, while introducing a highly efficient algorithm with built-in error bounds that successfully mapped hidden Arctic sea ice patterns using a standard laptop.
Key Facts
- The Infinite Data Myth Overthrown: The research demonstrates that the common tech-industry philosophy of “more data equals guaranteed learning” is mathematically incorrect. Certain highly complex or chaotic problems feature layered patterns that are hidden or impossible to neatly separate, meaning the absolute best an algorithm can score is a coin-flip ($50/50$), rendering the problem mathematically unsolvable regardless of dataset size.
- Why Chatbots Hallucinate: The mathematical instabilities that break long-term physical prediction explain why large language models (LLMs) like ChatGPT or Claude confidently invent false information over time. In highly sensitive systems, minute variations in the starting prompt trigger compounding errors that send the model down wildly separate pathways, preserving short-term coherence while entirely losing contact with reality.
- The Two Pillars of Machine Learning Failure: Dr. Matthew Colbrook’s team identified two specific structural reasons why AI modeling naturally breaks down when interacting with complex environments:
- Data Insufficiency Verification Failure: The machine learning algorithm possesses no internal mathematical mechanism to determine when it has ingested enough training samples to output a stable, provably certain prediction.
- Hidden Pattern Obfuscation: Critical tracking coordinates within the dynamic architecture remain mathematically hidden or deeply tangled, making them impossible for standard neural nets to differentiate.
- The Chaos Frequency Problem: When an AI analyzes a chaotic system (where tiny changes in starting parameters yield massive divergence), the Koopman operator produces a continuous spread of overlapping frequencies rather than clean, isolated tracking variables. This explains why short-term forecasts remain accurate, while long-term system projections systematically collapse.
- The Provably Reliable Algorithm: To solve this structural vulnerability, the researchers engineered a novel, mathematically rigorous algorithm featuring built-in, immutable error bounds. This toolkit gives researchers a real-time certainty meter, verifying exactly when an AI’s output can be trusted without requiring multi-million-dollar supercomputers.
- The Laptop vs. Supercomputer Benchmark: When stress-tested against 40 years of Arctic climate records, the team’s custom algorithm identified long-lost structural decay patterns in the ice sheets. It consistently outperformed the world’s leading commercial AI systems while running entirely on a basic, consumer-grade standard laptop at a fraction of the computational cost.
Source: University of Cambridge
When can we trust the results we get from AI, and when is learning impossible? Researchers have shown that there are some problems that even the most powerful AI can reliably solve, no matter how much data it’s given.
The researchers, from the University of Cambridge and the University of California Santa Barbara, designed ‘adversarial’ mathematical systems designed to fool any AI algorithm. Like ethical hackers stress-testing the security of a network, these adversarial systems were designed to map out exactly where and why AI prediction breaks down.
Many real-world systems – like those in oceans, the human brain, or robotics – are too complex to describe neatly with equations, so researchers often learn how they behave by using machine learning. But these AI methods don’t always work well, returning unreliable results or poor predictions.
Sometimes, however, providing reliable solutions is fundamentally impossible, even with infinite data. The adversarial systems developed by the researchers may help developers and users of AI systems know whether they’re working on a solvable or unsolvable problem, build methods that work, and avoid wasting time, effort or AI tokens when a problem is beyond the bounds of possibility.
Their results, reported in the journal Nature Communications, could also help explain why popular AI chatbots like ChatGPT or Claude can be accurate in the short term, but can drift or hallucinate over time.
“We’re probing the boundaries of what you can and can’t do with AI,” said lead author Dr Matthew Colbrook, from Cambridge’s Department of Applied Mathematics and Theoretical Physics. “It’s so important to understand what problems can’t be solved with these methods, because otherwise you end up wasting a lot of time and money.”
Colbrook and his co-authors used an approach called Koopman operator learning, which turns complicated nonlinear behaviour into a linear form that’s easier to analyse.
“What we were doing with these ‘adversaries’ was trying to figure out the types of systems that are hard or impossible to predict, and the types of systems that could be adapted to return reliable results,” said Colbrook.
The researchers identified two main reasons why machine learning breaks down when analysing complex systems: either the algorithm can’t tell when it’s seen enough data to return a reliable result, or patterns in the system can be hidden or hard to distinguish.
“In a lot of AI research, a common assumption is that if we just collect more data, learning will eventually work,” said Colbrook. “But we found this is often wrong. Learning is often layered, and requires multiple steps in the right order to work.”
When a system is chaotic — meaning tiny differences in starting conditions lead to wildly different trajectories, like a choose your own adventure story — the Koopman operator often ends up with a continuous spread of frequencies rather than clean, distinct modes. Short-term prediction was accurate, but long-term prediction became fundamentally unreliable, because the sensitivity to initial conditions compounds over time.
The same mathematical instability that defeats prediction algorithms may also explain why AI chatbots confidently fabricate facts: small changes in a question can send the chatbot down an entirely different path, one that looks plausible word-by-word but loses its grip on reality over longer outputs.
The researchers developed a way to classify these problems based on how many steps are needed to solve them. Where the data is not sufficiently layered or in the right order, the best an algorithm can do – even with infinite data – is 50/50, essentially classifying the problem as unsolvable.
The team also produced a new, provably reliable and highly efficient algorithm with built-in error bounds: essentially giving AI researchers a way to know when they’re able to trust the answer, at a fraction of the cost of most supercomputers.
The researchers tested their approach on over 40 years of Arctic sea ice data. Using their algorithm they found hidden patterns in how the ice is declining, and were able to outperform current leading AI models at a fraction of the cost, on a standard laptop.
“We’re at the stage now where there have been a lot of flashy examples and success stories in AI, but it’s vital that we also ask how certain the models are, and how we know whether they’re certain,” said Colbrook. “Otherwise, we’re building on very shaky foundations.”
Key Questions Answered:
A: Imagine trying to predict the path of smoke rising from a campfire. The smoke twists, loops, and breaks apart in a highly complicated, non-linear way that is nearly impossible to track with basic equations. A Koopman operator is a mathematical technique that takes this messy, non-linear behavior and maps it into an alternative, abstract landscape where the movement acts like a straight, linear line. By converting complex systems into this linear form, Dr. Matthew Colbrook’s team could stress-test the math, creating “adversarial networks” to pinpoint exactly where the equations break down and become impossible for an AI to solve.
A: Chatbots process text much like a chaotic weather system processes atmosphere, word by word, the path depends entirely on the starting point. When a system is chaotic, tiny shifts in the initial input send the path down wildly different trajectories. In a chatbot, changing a single letter or word in a long prompt can cause the AI to veer off its factual path. Word-by-word, the response sounds completely plausible, but over a long output, the compounding sensitivity to that minor initial change causes the model to drift away from reality, resulting in a hallucination.
A: The tech sector needs to stop blindly throwing massive computing power and uncurated datasets at complex problems. Instead, developers can use the Cambridge-UCSB team’s new algorithm, which features built-in error bounds. Think of it like a built-in dashboard gauge that tells you exactly how certain an AI model is about its output. By using this method, scientists can immediately distinguish whether a complex problem is solvable or fundamentally impossible, saving millions of dollars, cutting out wasted supercomputing time, and highlighting hidden patterns using basic, affordable laptops.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this AI research news
Author: Sarah Collins
Source: University of Cambridge
Contact: Sarah Collins – University of Cambridge
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Adversarial dynamical systems characterize when data-driven learning succeeds or fails” by Matthew J. Colbrook, Igor Mezić & Alexei Stepanenko. Nature Communications
DOI:10.1038/s41467-026-74220-8
Abstract
Adversarial dynamical systems characterize when data-driven learning succeeds or fails
Many systems resist analytical modeling, making data-driven inference of dynamics important. Yet data-driven methods can fail to converge or generalize, leaving open a central question: When can system behavior be learned reliably from data, and when is such learning impossible?
We answer this question using adversarial dynamical systems to identify the boundary between accessible and inaccessible regimes. In Koopman operator learning, a leading framework for representing nonlinear dynamics through linear spectral objects, we design optimal data-driven spectral algorithms with convergence and certification guarantees under conditions arising broadly in physical systems.
This yields a convergence theory for Koopman-operator approximations and resolves a longstanding open problem in Koopman spectral analysis. Conversely, by constructing adversarial systems, we prove matching impossibility results: without these conditions, no single-sequence limiting procedure can guarantee learning, regardless of data quality. These results sharply characterize when data-driven spectral learning can succeed and when it must fail. We validate the framework on oscillators, chaotic fluid flows and Arctic sea ice concentration forecasting.
In the latter, we uncover hidden modes of Arctic sea ice decline, deliver long-range forecasts with geographic error bounds, and outperform state-of-the-art dynamical and deep learning models at substantially lower computational cost, enabling real-time deployment on standard CPUs.
