Our Terms & Conditions | Our Privacy Policy
Unlike typical AI research, where a model predicts outcomes or cleans up data, researchers at Emory University in Atlanta did something unusual. They trained a neural network to discover new physics.
The team achieved this unique feat by feeding their AI system experimental data from a mysterious state of matter called dusty plasma, a hot, electrically charged gas filled with tiny dust particles. The scientists then watched as the AI revealed surprisingly accurate descriptions of strange forces that were never fully understood before.
The development shows that AI can be used to uncover previously unknown laws that govern how particles interact in a chaotic system. Plus, it corrects long-held assumptions in plasma physics and opens the door to studying complex, many-particle systems ranging from living cells to industrial materials in entirely new ways.
“We showed that we can use AI to discover new physics. Our AI method is not a black box: we understand how and why it works. The framework it provides is also universal. It could potentially be applied to other many-body systems to open new routes to discovery,” Justin Burton, one of the study authors and a professor at Emory, said.
How did the AI learn to create laws?
The researchers combined real-world experiments with a carefully designed AI model. They began by studying dusty plasma. This state of matter is found across the universe, from Saturn’s rings and the moon’s surface to wildfire smoke on Earth.
However, despite its cosmic presence, the exact forces acting between the particles in dusty plasma have remained poorly understood. That’s because the system behaves in a non-reciprocal way, which means that the force one particle applies on another isn’t necessarily matched in return.
Understanding such interactions using traditional physics has proven incredibly difficult. So to tackle this problem, the scientists built a sophisticated 3D imaging system to observe how plastic dust particles moved inside a chamber filled with plasma. They used a laser sheet and high-speed camera to capture thousands of tiny particle movements in three dimensions over time.
These detailed trajectories were then used to train a custom neural network. Unlike most AI models that need huge datasets, the Emory team’s network was trained on a small but rich dataset and was engineered with built-in physical rules, like accounting for gravity, drag, and particle-to-particle forces.
“When you’re probing something new, you don’t have a lot of data to train AI. That meant we would have to design a neural network that could be trained with a small amount of data and still learn something new,” said Ilya Nemenman, senior study author and a professor at the university.
The neural network broke down the particle motion into three components: velocity effects (like drag), environmental forces (such as gravity), and inter-particle forces. This allowed the AI to learn complex behaviors while obeying basic physics principles.
As a result, it discovered precise descriptions of the non-reciprocal forces with over 99% accuracy. One surprising insight was that when one particle leads, it pulls the trailing one toward it, but the trailing one pushes the leader away. This kind of asymmetric interaction had been suspected but never clearly modeled before.
Neural network also rectified past assumptions
The AI corrected some faulty assumptions that shaped plasma theory for years. “What’s even more interesting is that we show that some common theoretical assumptions about these forces are not quite accurate. We’re able to correct these inaccuracies because we can now see what’s occurring in such exquisite detail,” Nemenman added.
For instance, one such assumption was that a particle’s electric charge increases exactly with its size—turns out, it doesn’t. Instead, the relationship depends on the surrounding plasma’s density and temperature.
Another mistaken idea was that the force between particles always decreases exponentially with distance, regardless of their size. The AI revealed that this drop-off also depends on how big the particles are, an insight previously overlooked.
The best part is, this AI model ran on something as modest as a desktop computer. It produced a universal framework that can now be applied to all sorts of many-particle systems, from paint mixtures to migrating cells in living organisms. This research also demonstrates that AI can go far beyond crunching numbers. It can actually help scientists discover the hidden rules that govern nature.
“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,” Nemenman said. Hopefully, this work will encourage scientists to explore many other ways in which AI can benefit science and society.
The study is published in the journal PNAS.
Images are for reference only.Images and contents gathered automatic from google or 3rd party sources.All rights on the images and contents are with their legal original owners.
Aggregated From –
Comments are closed.