AI Model Reveals Hidden Physics in Dusty Plasma With 99% Accuracy
Physicists at Emory University have used a custom neural network to uncover previously unknown interaction patterns in dusty plasma, a state of matter found everywhere from Saturn's rings to Earth's wildfires. The AI model identified asymmetrical forces between particles with over 99% accuracy and overturned long-standing assumptions about how these forces behave.
The work, published in PNAS, demonstrates that AI can do more than analyze data or make predictions. It can help discover entirely new physical laws.
How the Experiment Worked
The team combined two key elements: a specially designed neural network and precise 3D tracking of particles in a plasma-filled chamber. Researchers suspended tiny plastic particles in a vacuum chamber and used a laser sheet moving through the space while a high-speed camera recorded images. These snapshots were reconstructed to track the motion of dozens of particles over time.
Dusty plasma consists of ionized gas filled with charged particles and dust grains. It appears in many environments: the solar wind streaming from the Sun, lightning strikes, the Moon's surface, and smoke during wildfires.
What the AI Discovered
The neural network separated particle motion into three influences: drag from velocity, environmental forces like gravity, and forces between particles. It then revealed complex interactions that had been difficult to measure before.
One key finding: particles exhibit non-reciprocal forces. A leading particle attracts the trailing particle, but the trailing particle always repels the leading one - similar to how two boats create waves that push and pull differently depending on their positions.
The model also challenged two long-held assumptions. The first held that a particle's electric charge increases in direct proportion to its size. The AI showed the relationship is more complex, depending on plasma density and temperature. The second assumption suggested forces between particles decrease exponentially with distance regardless of particle size. The model revealed that particle size does affect how quickly forces weaken.
Why the AI Design Mattered
Building this model required careful planning. Unlike systems trained on massive datasets, this project had limited experimental data available.
The team spent more than a year refining the neural network design through weekly meetings. "We needed to structure the network to follow the necessary rules while still allowing it to explore and infer unknown physics," said Justin Burton, an Emory professor of experimental physics.
"Once we came up with the correct structure of the network to train, it turned out to be fairly simple," said Ilya Nemenman, an Emory professor of theoretical physics.
The final model runs on a standard desktop computer, making it accessible for other research teams.
Broader Applications
The researchers believe this framework could apply to many-body systems across different fields. Potential applications range from industrial materials like paint and ink to groups of living cells.
Nemenman studies how complex systems emerge from simple interactions. He notes that understanding collective motion has medical relevance - for instance, how cells interact in tumors and whether some break away to become metastatic.
Compared to living systems, dusty plasma offered a simpler environment for testing whether AI could uncover new physical principles. "For all the talk about how AI is changing science, there are very few examples where something fundamentally new has been found directly by an AI system," Nemenman said.
The Human Element Remains Critical
The researchers emphasize that human expertise remains essential. Scientists must design the models carefully, interpret results, and apply critical thinking to ensure real advances in science.
"It takes critical thinking to develop and use AI tools in ways that make real advances in science, technology and the humanities," Burton said.
The National Science Foundation and Simons Foundation funded the research.
Learn more: Explore how AI is transforming scientific discovery with AI for Science & Research resources.
Your membership also unlocks:
