A greener way of sourcing essential minerals

By Sarah Murray

With mounting worries over supplies, the race is on to discover new sources or alternatives to the minerals essential for the green transition. But is there a more sustainable way to find and extract these materials — or even replace them altogether?

The most important of these — the rare earths used in batteries, wind turbines and other equipment — are not as rare as the name suggests, at least in geological terms, although they are extremely difficult to separate and refine. However, environmental concerns over mining operations, legal challenges and China’s dominance of global production are all intensifying efforts to find alternative supplies.

Mining operations, meanwhile, damage the environment and the lives of those nearby, prompting legal action from NGOs and pushback from local communities.

In addition, because rare earths and other critical minerals were once readily available, industry has not invested sufficiently in new mining and refining technologies.

“These things were easy to find,” says Mfikeyi Makayi, chief executive of the African arm of mining start-up KoBold Metals Africa. “It was sitting there, so why should you invent some tool to go deeper?”

Backed by Bill Gates and Jeff Bezos and operating in the Democratic Republic of Congo, KoBold is among those companies using AI to identify untapped deposits that can increase supply while minimising the environmental impact of production.

Makayi explains that AI can identify places where there are higher percentages of materials such as lithium, cobalt and nickel per tonne of rock, which means drilling operations can be minimised. “You’re not just digging up holes to find poor-quality material,” she says. “So we’re using AI to turn the art of finding great rocks into a repeatable science.”

AI can also help identify valuable materials in previously untapped sources such as mining industry waste, says Julie Bryce, professor of earth sciences at the University of New Hampshire. “Rare earths are present in lots of materials,” she explains. “But now that they’re so valuable, it’s becoming more economically feasible to go after them.”

Meanwhile, companies are using AI to find out how new combinations of materials might produce “rare earth free” versions of the magnets and other components needed in clean energy infrastructure.

For example, some materials have magnetic properties but, unlike rare earths, cannot retain those properties. “A regular magnetic element like iron by itself doesn’t keep the same magnetisation, so you have to coax it into that state,” says Kathy Christofidou, chair in digital and sustainable metallurgy at the University of Sheffield.

From the periodic table’s vast range of possibilities, AI can identify elements that could do this coaxing far faster than humans. “It would take multiple days to create just one sample to test, and the testing takes a while,” says Christofidou, who is part of a team working with trade association UK AI and materials science company MatNex to develop rare-earth-free materials.

Of course, AI relies on information. This has drawn attention to the world’s materials databases, where work is being done to connect and make accessible sources of global geological data, ranging from digital archives to paper files, government archives and old maps on parchment.

“There’s a collective challenge as an industry where we’re not bringing these bodies together,” says Makayi, who adds that KoBold is working with the governments of Zambia and Congo to help organise their national geological databases and make them more accessible.

“Data to AI is like fuel to a machine,” says Jiadong Zang, professor of materials science at the University of New Hampshire, where he helped build a database of more than 67,000 magnetic materials, including 25 previously unrecognised compounds whose magnetic properties remain constant even at high temperatures.

With limited data on magnetic materials, Zang and his colleagues used large language models to read tens of thousands of papers and extract structural, chemical and domain information on their magnetic properties.

Armed with this “fuel”, AI can combine large numbers of materials at speed, enabling humans to conduct fewer but better physical tests. “You can do experiments in the virtual world before going to the lab,” says Jonathan Bean, co-founder and chief executive of MatNex.

With sufficient data, AI can even come up with ideas humans might have dismissed, says Rafael Gómez-Bombarelli, an MIT professor and co-founder of Lila Sciences, a materials design company that is building AI systems that can learn for themselves.

He cites the company’s experiments in replicating ruthenium and iridium, used in generating green hydrogen. Having been through a process of learning, he says, the AI system came up with a viable idea that the company’s in-house experts would not have pursued.

“We don’t want to lose human creativity,” he says. “But the places where we’ve seen moments of superhuman intelligence are when AI makes those decisions.”

While AI itself relies on power-guzzling data centres, Gómez-Bombarelli believes that the potential for scientists to accelerate the transition to clean energy more than balances out their relatively modest AI use.

“Given the amount of energy that is going into science it will be a net positive,” he says.