An AI learns to see crystals in 3D — and spots battery materials words alone missed

Ask a computer to invent a better battery material and it usually only gets to read about crystals — not actually look at them. Most crystals, the neatly ordered solids that make up more than 90 percent of solid materials on Earth, are stored as 3D maps of atoms, a language today's AI models barely speak. A team at MIT has now taught one to see. Their model pairs those 3D atomic maps with rich written descriptions of each crystal — what it is made of, how it is put together and what it might be good for — and learns the two together. In a paper in Nature Communications, the model that learned this way beat one trained on text alone, and when it was asked to find materials suited to solid-state lithium-ion batteries (a safer, longer-lasting kind of battery), its top picks included materials other researchers had already independently flagged as promising. 'Structure is really at the heart of the materials science field,' said senior author Ju Li. 'If you understand the structure, you can infer quite a lot about the properties and performance of materials.'

SDG 9 · Industry, innovation and infrastructure — Mechanism: tech innovation