American researchers from Emory University in Atlanta taught AI make new physical discoveries, not just look for mistakes or predict research results.
The scientists loaded experimental data into the AI system about a mysterious state of matter called dust plasma — a hot, electrically charged gas filled with small dust particles. After that, the researchers observed, how the AI surprisingly accurately described the strange forces that scientists had not fully understood before.
This demonstrates that artificial intelligence can be used to revealing previously unknown laws governing the interaction of particles in a chaotic system. AI also corrected the established assumptions about plasma physics.
“We have shown that we can use AI to discover new physical phenomena. Our AI method is not a black box: we understand how and why it works. The structure it proposes is also universal. It can potentially be applied to other multi-particle systems to reveal new paths to discovery”, — explains study co-author Professor Justin Burton.
The researchers combined real-world experiments with a carefully developed AI model. They started by studying dust plasma. This state of matter is found everywhere in the universe, from the rings of Saturn and the surface of the Moon to forest fires on Earth.
However, the forces that act between particles in the dust plasma remained poorly understood. This is because the particles affect each other with different strengths. Understanding such interactions using traditional physics has proven to be extremely difficult. Therefore to solve this problem, scientists have created a sophisticated three-dimensional visualization system that allows them to observe the movement of plastic dust particles inside a plasma-filled chamber.
The researchers used a laser knife and a high-speed camera to capture thousands of movements of small particles in three dimensions over time. The detailed trajectories of the particles were used to train an AI model. Most neural network models, that require large datasets to learn, a network of researchers at Emory University has been learning from small but rich datasets and was designed with built-in physical rules, such as taking into account gravity, drag, and interaction forces between particles.
“When you’re researching something new, you don’t have a lot of data to train AI. This meant that we had to develop a neural network that could be trained on a small amount of data and gain new knowledge”, — says senior author of the study, professor Ilya Nemenman.
The neural network decomposed the motion of particles into three components: velocity effects, which include drag, environmental forces, such as gravity, and forces of interaction between particles. The AI provided descriptions of the non-mutual forces with 99% accuracy. In particular, the AI found that the particles in front of them attract the ones behind them, and the ones behind them repel the particles in front of them. Such an asymmetric interaction has been predicted but never modeled before In addition, AI has corrected the erroneous assumptions that have defined plasma theory for years.
For example, one such assumption was that the electric charge of a particle increases in proportion to its size. It turned out that this is not the case. This relationship correlates with the density and temperature of the surrounding plasma. Another false assumption was that the strength of interaction between particles always decreases with distance, regardless of their size. AI has shown that this decrease also depends on the size of the particles, which was previously ignored.
Interestingly, this AI model worked effectively on a regular computer. It created a universal structure, which can now be applied to any multi-particle system, from paints to cell migration in living organisms. This research also demonstrates that AI can go far beyond simple computing.
“Even more interesting is that we show that some common theoretical assumptions about these forces are not entirely accurate. We can correct these inaccuracies because we can now see what is happening in such fine detail. Despite all the talk about how AI is changing science, there are very few examples where something fundamentally new has been discovered directly with the help of AI”, — Professor Nemenman emphasized.
The results of the study are published in the journal PNAS
Source: Interesting Engineering
Контент сайту призначений для осіб віком від 21 року. Переглядаючи матеріали, ви підтверджуєте свою відповідність віковим обмеженням.
Cуб'єкт у сфері онлайн-медіа; ідентифікатор медіа - R40-06029.