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. Mapping The Boundaries Of AI 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. Convergent general-purpose methods for Koopman learning. (CREDIT: Nature Communications) 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 …