AI discovers hidden physics rules in plasma by learning from dust particles

Physicists at Emory University have used machine learning to uncover physical laws that were invisible to traditional analysis — not by crunching massive datasets, but by carefully watching dust particles move through plasma in a laboratory chamber.

The breakthrough came from a deliberate constraint: instead of feeding an AI system billions of internet pages like ChatGPT, the team designed a neural network that could learn something genuinely new from a small amount of carefully measured data. "We showed that we can use AI to discover new physics," says Justin Burton, an experimental physicist at Emory. "Our AI method is not a black box: we understand how and why it works."

Plasma — ionized gas where electrons and ions move freely — makes up 99.9% of the visible universe. It appears in solar wind, lightning, and Saturn's rings. Dusty plasma, which contains charged dust particles alongside the ions and electrons, shows up in surprising places: the Moon's surface (which is why astronauts' suits get covered in dust), Earth's ionosphere, and even wildfire smoke, where charged soot particles can disrupt firefighter communications.

In Burton's lab, tiny plastic particles suspended in a vacuum chamber act as a simplified model of these complex systems. Using a laser sheet and high-speed camera, the team created a tomographic imaging method to track the three-dimensional motion of dozens of particles over several minutes. The AI then analyzed these trajectories to infer the forces governing particle interactions.

The results corrected longstanding assumptions. The neural network described non-reciprocal forces — where a leading particle attracts a trailing particle, but the trailing particle always repels the leading one — with over 99% accuracy. It's similar to how two boats create wakes that affect each other differently depending on their relative positions.

The team also found that a classical theory about dust particle charge was incomplete. While larger particles do carry larger charges, the relationship isn't perfectly proportional to size — it depends on plasma density and temperature. Another theory assumed force drop-off with distance followed a fixed pattern regardless of particle size. The AI revealed that particle size actually matters here too.

"What's even more interesting is that we show that some common theoretical assumptions about these forces are not quite accurate," says Ilya Nemenman, a theoretical physicist and co-senior author. "We're able to correct these inaccuracies because we can now see what's occurring in such exquisite detail."

Nemenman studies collective behavior — how individual interactions create whole-system patterns — from cells in tumors to flocking birds. He saw the dusty plasma project as a test case for a broader framework. "For all the talk about how AI is revolutionizing science, there are very few examples where something fundamentally new has been found directly by an AI system," he notes.

The key was designing the right neural network structure. The team spent over a year in weekly meetings, eventually distilling the problem into three independent contributions to particle motion: drag force, environmental forces like gravity, and particle-to-particle forces. "Once we came up with the correct structure of the network to train, it turned out to be fairly simple," Nemenman says.

Their physics-based neural network runs on a desktop computer and offers what they call a universal framework — applicable to colloids like paint and ink, clusters of cells in living organisms, and other systems made of large numbers of interacting particles. Nemenman is planning to teach this approach at an upcoming visiting professorship in Germany, training students from around the world to use AI to infer physics in living systems, not just plasmas.

But the researchers are clear-eyed about AI's role. Expert human physicists remain essential to design the right network structure, interpret results, and validate findings. "It takes critical thinking to develop and use AI tools in ways that make real advances in science, technology, and the humanities," Burton says. The technology opens doors, but people still have to decide which ones matter most.