Based on recent headlines and social media discussions, one might easily think that AI will revolutionize the power grid, eradicate diseases globally, and even handle my holiday shopping. However, it might just be a considerable amount of exaggeration circulating out there.
This week, we unveiled a new package titled Hype Correction. This compilation of stories examines how society is beginning to grapple with the actual capabilities of AI versus mere hype.
One of my top picks from that package is an article by my colleague David Rotman, who conducted a thorough analysis of AI in materials research. AI holds the potential to revolutionize the discovery of new materials—innovation that could be particularly beneficial within the climate tech sector, which requires innovative batteries, semiconductors, magnets, and more.
However, the domain still has to demonstrate its capability to create materials that are genuinely novel and valuable. Can AI truly enhance materials research? What might that entail?
For researchers seeking fresh solutions to power the globe (or to find cures or achieve any number of other significant objectives), a new material could entirely shift the landscape.
The challenge is that developing materials is both complex and time-consuming. Take plastic, for instance—the first fully synthetic plastic emerged in 1907, yet it wasn’t until around the 1950s that companies could produce the extensive variety we recognize today. (And of course, while it is tremendously useful, plastic also leads to a multitude of societal issues.)
In recent years, materials science has stagnated somewhat—David has been covering this field for nearly 40 years, and as he mentions, there have been only a few significant commercial advancements during that span. (Lithium-ion batteries are one of them.)
Could AI truly change everything? The possibility is an enticing one, and firms are racing to explore it.
Lila Sciences, located in Cambridge, Massachusetts, is exploring the use of AI models to reveal new materials. The company not only trains an AI model on the latest scientific literature but also integrates it into an automated laboratory, allowing it to learn from experimental findings. The objective is to expedite the iterative process of inventing and testing new materials and to analyze research from angles that humans might overlook.
Earlier this year, at an MIT Technology Review event, I had the opportunity to hear David interview Rafael Gómez-Bombarelli, a cofounder of Lila. While discussing the company’s initiatives, Gómez-Bombarelli acknowledged that AI materials discovery has yet to experience a significant breakthrough. However, this may change in the future.
Gómez-Bombarelli explained that the models trained by Lila are delivering insights that are “as profound [as] or deeper than those our domain scientists possess.” He added that in the future, AI might “think” differently from human scientists when addressing problems: “There will be a necessity to convert scientific reasoning by AI into the way we conceptualize the world.”
It’s thrilling to witness this form of enthusiasm in materials research, yet a complex and lengthy journey lies ahead before we can confidently assert that AI has significantly altered the discipline. A substantial hurdle is that while it’s one matter to receive suggestions from a model regarding new experimental techniques or potential structures, it’s another entirely to synthesize a material and demonstrate that it is both novel and functional.
You might recall that a couple of years ago, Google’s DeepMind reported using AI to forecast the structures of “millions of new materials” and successfully creating hundreds of them in a lab.
However, as David highlights in his story, following that announcement, some materials scientists pointed out that various purportedly novel materials were essentially just slightly altered versions of known variants. Others couldn’t even exist under standard conditions (as the simulations were conducted at ultra-low temperatures, where atomic movement is minimal).
It’s conceivable that AI could energize materials discovery, heralding a new era filled with superconductors, batteries, and magnets unlike any we have encountered before. But for now, I suggest viewers temper their expectations.
This article is part of The Spark, MIT Technology Review’s weekly climate newsletter. To have it delivered to your inbox every Wednesday, sign up here.
