Imagine trying to write down the rules for something as wild and messy as a hurricane — all the swirling wind, changing pressure, and unpredictable movements. That is basically what scientists have been struggling to do for complex systems like weather, ocean currents, or the human heart. Now, a team at Clarkson University in Potsdam, New York, has built an AI tool that can do exactly that: find the hidden mathematical equations behind the world's most chaotic phenomena, just by looking at data.
The tool is called KANDy, short for Kolmogorov-Arnold Networks for Dynamics. Most AI models today are good at predicting what will happen next, but they work like "black boxes" — they give you an answer without explaining why. KANDy is different. It does not just guess the future; it discovers the actual mathematical rules driving a system's behavior.
Research Associate Kevin Slote and Electrical and Computer Engineering Research Assistant Professor Jeremie Fish led the work under Erik Bollt at Clarkson. They adapted a type of neural network called KANs — think of neural networks as computer systems inspired by the human brain — and turned it into something that can crack open complex systems and show their inner workings.
The team tested KANDy on all kinds of tough problems: systems that jump between states, systems that change smoothly over time, and even chaotic partial differential equations, which are notoriously difficult math problems. KANDy handled them all. In one striking test, the tool successfully uncovered important structure in something called the Hopf fibration, a mathematical object that even seasoned mathematicians find mind-bending.
So why does this matter? For scientists and engineers, understanding the equations behind a system means understanding the system itself. Rather than just watching patterns and making guesses, they can now peer under the hood. That could help with predicting disease spread, designing better engines, or understanding climate patterns — anywhere messy, nonlinear behavior has kept researchers stuck.
The research was published on the arXiv preprint server and the team has made KANDy available on GitHub so other scientists can try it out. The hope is that this tool opens doors for anyone studying systems that have resisted traditional analysis.
For Bollt and his team, this work is just the beginning. As AI tools like KANDy grow more powerful, the line between prediction and understanding continues to blur — and that may be the most exciting equation of all.