Artificial intelligence promises clear answers from complex data.
Yet new research suggests there are limits that no amount of data can overcome.
Some problems, scientists now show, cannot be solved by AI at all.
A study explores where machine learning succeeds and where it fails.
Researchers from the University of Cambridge and the University of California Santa Barbara designed mathematical systems meant to expose these limits.
Their findings reveal a deeper truth.
In some cases, learning is not just difficult, it is impossible.
Modern science often relies on AI to study systems that are too complex for traditional equations.
These include ocean currents, brain activity, and robotic motion.
Instead of writing down exact rules, scientists collect data and train algorithms to learn patterns.
This approach has led to major breakthroughs.
Still, it does not always work.
Models can give unstable results or predictions that drift over time.
Dr Matthew Colbrook, the study’s lead author from Cambridge’s Department of Applied Mathematics and Theoretical Physics, explained the goal.
“We’re probing the boundaries of what you can and can’t do with AI,” he said.
“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.” To test these limits, the team built what they call adversarial systems.
These are carefully designed problems that look ordinary but hide features that confuse any algorithm.
The researchers found two main reasons why AI fails in complex systems.
In some cases, the algorithm cannot tell when it has seen enough data.
It keeps learning without ever reaching a reliable conclusion.
In others, patterns exist but remain hidden or too subtle to detect.
More striking is the discovery that some problems cannot be solved at all.
Even with unlimited data and perfect algorithms, the best possible outcome is no better than chance.
Colbrook described the assumption many researchers make.
“In a lot of AI research, a common assumption is that if we just collect more data, learning will eventually work,” he said.
“But we found this is often wrong.
Learning is often layered, and requires multiple steps in the right order to work.” When these steps are missing or misordered, the system becomes unsolvable.
One of the clearest examples comes from chaotic systems.
These are systems where tiny differences at the start lead to large differences later.
Weather patterns offer a familiar case.
A small change in temperature or wind can lead to very different outcomes over time.
The study shows that in such systems, AI can make accurate short-term predictions.
Over longer periods, however, errors grow quickly.
Small uncertainties multiply until predictions lose meaning.
The researchers used a method called Koopman operator learning to analyze this behavior.
This approach transforms complex systems into a simpler form that can be studied more easily.
Still, even with this tool, chaos creates a barrier.
Instead of clear patterns, the system shows a continuous spread of behaviors.
This makes long-term prediction unreliable.
The findings also shed light on why AI systems sometimes produce false or misleading results.
Large language models can....
