Scientists at the University of New Hampshire just added 25 previously unknown magnetic compounds to humanity's toolkit—and they did it by teaching an AI system to read decades of scientific papers.
The breakthrough matters because the magnets powering everything from electric vehicles to wind turbines rely almost entirely on rare earth elements. These materials are expensive, mostly imported, and increasingly hard to source. Finding alternatives has been a slow grind: researchers know many magnetic compounds exist in the scientific literature, but extracting that knowledge and testing it in the lab takes years.
Automating the Discovery Process
The UNH team trained an AI to scan and interpret experimental details from scientific papers, then fed that data into computer models that predict whether a material is magnetic and how much heat it can withstand before losing that property. The result is the Northeast Materials Database—a searchable collection of 67,573 magnetic materials, including those 25 new candidates that retain their magnetism at high temperatures.
We're a new kind of news feed.
Regular news is designed to drain you. We're a non-profit built to restore you. Every story we publish is scored for impact, progress, and hope.
Start Your News Detox"By accelerating the discovery of sustainable magnetic materials, we can reduce dependence on rare earth elements, lower the cost of electric vehicles and renewable energy systems, and strengthen the U.S. manufacturing base," said Suman Itani, the doctoral student in physics who led the work.
What makes this approach powerful is speed. Instead of researchers spending years hunting through papers and running individual experiments, they can now query a database and identify the most promising materials to test in the lab. It's the difference between searching a library card catalog and using Google.
The work, published in Nature Communications, shows how large language models—the same technology behind modern AI assistants—can be repurposed for scientific discovery. Jiadong Zang, physics professor and co-author, notes the team is "optimistic that our experimental database and growing AI technologies will make this goal achievable."
The next phase is obvious: take these 25 candidates into the laboratory and see which ones actually perform as predicted. But the real win here is the infrastructure. Future researchers will build on this database, adding new discoveries and refining predictions. What started as a way to solve one bottleneck—finding alternatives to rare earth magnets—has created a template for accelerating materials science itself.
