AI solves 100-year-old mystery to supercharge scientific discovery

Breakthrough AI model could help advance everything from ultra-powerful batteries, to archaeology

Researchers from Columbia Engineering say a century-old puzzle could hold the key to next-generation batteries (iStock/ Getty Images)

Scientists in the US have solved a century-old puzzle that could fast-track the development of powerful, longer-lasting batteries.

The longstanding problem involved determining the exact atomic structures of nanocrystals – tiny, disordered materials critical for advancing everything from electronics to archaeology.

Previous methods involve shining an X-ray beam through large, pure crystals to produce clear patterns. But this approach does not work on nanocrystals, which come in the form of a powder and scatter X-rays into indecipherable patterns.

Using a custom-built artificial intelligence algorithm, a team from Columbia Engineering in New York were able to observe the pattern produced by nanocrystals in order to infer the material’s atomic structure.

“The AI solved this problem by learning everything it could from a database of many thousands of known, but unrelated, structures,” said Simon Billinge, professor of materials science and of applied physics and applied mathematics at Columbia Engineering.

“Just as ChatGPT learns the patterns of language, the AI model learned the patterns of atomic arrangements that nature allows.”

The tool they developed, called PXRDnet, is trained on tens of thousands of known materials that allows it to figure out the structure of crystals as small as 10 angstronms – thousands of times thinner than a human hair.

It marks a major advance in materials science, allowing researchers to dramatically expand the ability to identify and characterise nanomaterials that were inaccessible with traditional methods.

It also demonstrates the massive progress made with artificial intelligence in recent years, with the researchers noting that such a discovery would have previously been thought impossible.

“When I was in middle school, the field was struggling to build algorithms that could tell cats from dogs,” said Gabe Guo, who led the project at Columbia.

“Now, studies like ours underscore the massive power of AI to augment the power of human scientists and accelerate innovation to new levels.”

The research was published in the journal Nature Materials on Monday, in a study titled ‘Ab initio structure solutions from nanocrystalline powder diffraction data via diffusion models’.

“What particularly excites me is that with relatively little background knowledge in physics or geometry, AI was able to learn to solve a puzzle that has baffled human researchers for a century,” said Hod Lipson, chair of the Department of Mechanical Engineering at Columbia Engineering.

“This is a sign of things to come for many other fields facing long-standing challenges.”