Scientists Use AI to Crack Nature’s Hidden Laws

Scientists have long relied on simplified equations to understand the world, from Newton’s laws of motion to modern models of weather and electronics. But many real-world systems are so complex that finding those clean mathematical rules has become nearly impossible.

Now, a new artificial intelligence (AI) system developed at Duke University may change that. The AI framework, described in a study published December 17 in NPJ Complexity, can analyze raw data from complex systems and automatically generate simple equations that describe how those systems evolve over time. 

In doing so, it revives a core goal of science, which is to turn overwhelming complexity into understandable laws.

Teaching AI to Think Like a Scientist

The new system is inspired by the work of early ‘dynamicists,’ scientists who studied systems that change over time. Just as Isaac Newton derived equations linking force and motion, the AI takes time-series data, or measurements collected as a system evolves, and searches for mathematical relationships that govern its behavior.

What makes the approach remarkable is its ability to reduce extreme complexity. Many natural and engineered systems involve hundreds or even thousands of interacting variables. While traditional machine learning models can fit such data, they often act like black boxes, offering predictions without insight.

This AI does things differently. It searches for hidden structures that allow a complex, nonlinear system to be represented by a much simpler, nearly linear model. According to Buyoan Chen, director of Duke’s General Robotics Lab and senior author of the study:

“We increasingly have the raw data needed to understand complex systems, but not the tools to turn that information into the kinds of simplified rules scientists rely on. Bridging that gap is essential.”

From Chaos to Clarity

This approach builds on an idea first proposed by mathematician Bernard Koopman in the 1930s, which suggested that nonlinear systems could be represented using linear mathematics, at least in theory. In practice, doing so required massive numbers of equations that quickly exceeded human intuition.

The Duke team’s AI tackles this challenge by combining deep learning with physics-inspired constraints. It identifies patterns in how systems evolve over time, then distills those patterns into a smaller set of variables that still capture the system’s essential behavior.

The result is a compact mathematical description that behaves like a linear system while retaining the accuracy needed for long-term predictions.

To test the framework, the researchers applied it to a wide range of systems, including pendulums, electrical circuits, climate models, and neural activity patterns. Despite their differences, each system revealed a small number of hidden variables that governed its dynamics.

In many cases, the AI’s models were more than ten times smaller than those produced by previous machine learning methods, while remaining accurate over extended time horizons.

Why This Matters for Science and Engineering

The ability to extract simple equations from complex data could reshape how scientists work across disciplines. Weather forecasting, electrical engineering, mechanical systems, and biological research all depend on understanding how systems change over time.

Rather than replacing traditional theory, the AI acts as a powerful assistant, helping researchers uncover structures they might otherwise miss. The equations it produces are interpretable, meaning scientists can examine, test, and refine them rather than simply trusting opaque predictions.

This interpretability is critical. As AI becomes more deeply embedded in scientific research, tools that enhance understanding, rather than obscure it, are increasingly valuable.

By showing that AI can uncover the hidden simplicity inside complex systems, the Duke team’s work suggests a future where AI helps accelerate scientific discovery, not by guessing answers, but by revealing the rules that make the world intelligible.

More Must-Reads:

• Researchers Uncover Hidden Bias in Cancer Diagnosis AI
• Can AI Ever Be Proven Conscious? A Philosopher Says Probably Not
• AI Learns Math Better by Arguing With Itself