DURHAM, N.C., December 17, 2023 – Researchers at Duke University have developed a new artificial intelligence framework capable of deciphering the underlying rules governing some of the most complex systems in nature and technology, offering a powerful new lens for scientific finding.
Unlocking Nature’s Code with AI
Table of Contents
A new AI framework simplifies complex systems, revealing hidden patterns and offering insights into everything from weather to electrical circuits.
- The AI can reduce systems with thousands of interacting variables to simpler rules.
- Inspired by the work of historical “dynamicists” like Isaac Newton, it translates data into understandable equations.
- The framework identifies stable states within complex systems,helping predict behavior and detect instability.
- Researchers envision a future where AI assists in automated scientific discovery, acting as “machine scientists.”
Just as isaac Newton, often considered the first “dynamicist,” developed equations linking force and motion, this AI analyzes evolving data and generates equations that accurately describe that behavior. this breakthrough offers a new way to understand change over time, a fundamental challenge across numerous scientific disciplines.
A New Tool for Understanding Change Over Time
The research, published online in the journal npj Complexity, introduces a powerful method for scientists to study systems that evolve – including weather patterns, electrical circuits, mechanical devices, and biological signals. “Scientific discovery has always depended on finding simplified representations of complicated systems,” explained Professor Yilun Chen of Duke’s Department of Mechanical Engineering and Materials Science.
How the AI Reduces Complexity
To test the framework,researchers applied it to diverse systems,from a swinging pendulum to nonlinear electrical circuits,climate models,and neural circuits. Despite their differences, the AI consistently identified a small number of hidden variables governing their behavior. In many cases, the resulting models were more than 10 times smaller than those produced by earlier machine-learning methods, while still delivering reliable long-term predictions.
“What stands out is not just the accuracy, but the interpretability,” said Chen, who also holds appointments in electrical and computer engineering and computer science. “When a linear model is compact, the scientific discovery process can be naturally connected to existing theories and methods that human scientists have developed over millennia.It’s like connecting AI scientists with human scientists.”
Finding Stability and Warning Signs
The framework goes beyond prediction, identifying stable states – known as attractors – where a system naturally settles. Recognizing these states is crucial for determining if a system is functioning normally, drifting, or approaching instability. “For a dynamicist, finding these structures is like finding the landmarks of a new landscape,” said Sam Moore, the lead author and PhD candidate in Chen’s general Robotics Lab. “Once you know where the stable points are, the rest of the system starts to make sense.”
The researchers emphasize that this method is notably useful when traditional equations are unavailable, incomplete, or too complex to derive. “This is not about replacing physics,” Moore continued. “It’s about extending our ability to reason using data when the physics is unknown, hidden, or too cumbersome to write down.”
Toward Machine Scientists
the team is exploring how the framework could guide experimental design by actively selecting data to reveal a system’s structure more efficiently. They also plan to apply the method to richer data forms, including video, audio, and signals from complex biological systems. This research supports a long-term goal in Chen’s General Robotics lab: to develop “machine scientists” that assist with automated scientific discovery, linking modern AI with the mathematical language of dynamical systems and possibly uncovering the fundamental rules shaping the physical world and living systems.
This work was supported by the National Science Foundation Graduate Research Fellowship, the Army Research Laboratory STRONG program (W911NF2320182, W911NF2220113), the Army Research Office (W911NF2410405), the DARPA FoundSci program (HR00112490372), and the DARPA TIAMAT program (HR00112490419).
