Wentao Yu’s neural network spotted a pattern in dusty plasma data that no human had seen before: particles influencing each other in ways that broke long-standing assumptions about force symmetry.
How the AI uncovered non-reciprocal forces in plasma
Researchers at Emory University trained a custom neural network on experimental data from a dusty plasma—a mixture of ionized gas and charged dust grains—to model interactions between particles. The AI identified non-reciprocal forces, where particle A affects particle B differently than B affects A, with over 99% accuracy. This level of precision revealed flaws in existing theoretical models that assumed forces were mutual.
Why this method changes how physics is discovered
The team emphasized their approach is not a black box; they can trace how the AI reaches its conclusions, making it a transparent tool for scientific inquiry. Ilya Nemenman, co-senior author, noted the framework could apply to other complex systems like cell cultures or industrial fluids where particle interactions are too intricate for traditional math. Last time a similar shift occurred, computational methods in the 1980s allowed physicists to simulate chaotic systems, but this is the first time AI has directly revealed new physical laws from experimental data.
What this means for future research across fields
Vyacheslav Lukin of the National Science Foundation said the work exemplifies how AI-driven insights in plasma physics could advance understanding of living systems, where collective behavior dominates. The researchers plan to test the method in biological contexts, such as modeling how cells communicate through mechanical forces. While the current study focused on plasma, the underlying technique is designed to be adaptable to any system with many interacting parts.
What is a dusty plasma?
A dusty plasma is an ionized gas that contains tiny solid particles, such as dust grains, which become electrically charged and interact with the plasma and each other.
Can this AI method be used outside of physics?
Yes, the researchers say the framework could apply to other many-body systems, including biological systems like groups of living cells or industrial materials such as paints and inks.
