Materials-predicting AI from DeepMind could revolutionize electronics, batteries, and solar cells

The materials cookbook has suddenly grown tens of times longer. Modern technologies, from electronics to airplanes, draw on just 20,000 inorganic materials, largely discovered through trial and error; scientists have predicted but not made tens of thousands more. This week, however, researchers report that with a new artificial intelligence (AI), they have predicted the ingredients and properties of another 2.2 million materials. In a companion study, a separate team has shown that such predicted materials can be made efficiently, again with the help of AI. Together, researchers say, the reports foreshadow a new age of materials science, when AI programs and robots will power the search for the makings of novel batteries, superconductors, and catalysts. “It’s very impressive,” says Andrew Rosen, a computational materials scientist at Princeton University. The predictions, published in Nature , are another coup for the AI innovators at DeepMind, an offshoot of Google. Last month, they described an AI algorithm that runs on laptops and can predict the weather as accurately as large, supercomputer-driven models . Prior to that DeepMind developed AlphaFold, an AI that’s able to predict the 3D shape of hundreds of millions of different proteins from their amino acid sequence alone. The new work, Rosen says, “is the AlphaFold equivalent for materials science.” Like previous DeepMind achievements, this one trained an AI wit...
Source: Science of Aging Knowledge Environment - Category: Geriatrics Source Type: research