Judging from headlines and social media posts in recent years, one might reasonably assume that AI is going to fix the power grid, cure the world’s diseases, and finish my holiday shopping for me. But maybe there’s just a whole lot of hype floating around out there.

One of my favorite stories in that package comes from my colleague David Rotman, who took a hard look at AI for materials research. AI could transform the process of discovering new materials—innovation that could be especially useful in the world of climate tech, which needs new batteries, semiconductors, magnets, and more. 

But the field still needs to prove it can make materials that are actually novel and useful. Can AI really supercharge materials research? What could that look like?

For researchers hoping to find new ways to power the world (or cure disease or achieve any number of other big, important goals), a new material could change everything.

The problem is, inventing materials is difficult and slow. Just look at plastic—the first totally synthetic plastic was invented in 1907, but it took until roughly the 1950s for companies to produce the wide range we’re familiar with today. (And of course, though it is incredibly useful, plastic also causes no shortage of complications for society.)

In recent decades, materials science has fallen a bit flat—David has been covering this field for nearly 40 years, and as he puts it, there have been just a few major commercial breakthroughs in that time. (Lithium-ion batteries are one.)

Could AI change everything? The prospect is a tantalizing one, and companies are racing to test it out.

Lila Sciences, based in Cambridge, Massachusetts, is working on using AI models to uncover new materials. The company can not only train an AI model on all the latest scientific literature, but also plug it into an automated lab, so it can learn from experimental data. The goal is to speed up the iterative process of inventing and testing new materials and look at research in ways that humans might miss.

At an MIT Technology Review event earlier this year, I got to listen to David interview Rafael Gómez-Bombarelli, one of Lila’s cofounders. As he described what the company is working on, Gómez-Bombarelli acknowledged that AI materials discovery hasn’t yet seen a big breakthrough moment. Yet.

Gómez-Bombarelli described how models Lila has trained are providing insights that are “as deep [as] or deeper than our domain scientists would have.” In the future, AI could “think” in ways that depart from how human scientists approach a problem, he added: “There will be a need to translate scientific reasoning by AI to the way we think about the world.”

It’s exciting to see this sort of optimism in materials research, but there’s still a long and winding road before we can satisfyingly say that AI has transformed the field. One major difficulty is that it’s one thing to take suggestions from a model about new experimental methods or new potential structures. It’s quite another to actually make a material and show that it’s novel and useful.

You might remember that a couple of years ago, Google’s DeepMind announced it had used AI to predict the structures of “millions of new materials” and had made hundreds of them in the lab.

But as David notes in his story, after that announcement, some materials scientists pointed out that some of the supposedly novel materials were basically slightly different versions of known ones. Others couldn’t even physically exist in normal conditions (the simulations were done at ultra-low temperatures, where atoms don’t move around much).

It’s possible that AI could give materials discovery a much-needed jolt and usher in a new age that brings superconductors and batteries and magnets we’ve never seen before. But for now, I’m calling hype. 

This article is from The Spark, MIT Technology Review’s weekly climate newsletter. To receive it in your inbox every Wednesday, sign up here.