By listening to the tech giants showering the public with conversational agents modeled on ChatGPT, we almost forget how immense the potential of machine learning is in the field of research.
British researchers have proven it once again by designing a super-powerful rare earth-free magnet thanks to contributions from a proprietary AI model. An advance that might seem anecdotal, but which is full of very concrete implications.
Rare earths, the ball and chain of electric transport
We are thinking in particular of the field of electric automobiles. Today, nearly 80% of these vehicles are powered by motors that contain powerful magnets, and the industry therefore finds itself in a quite paradoxical situation. Indeed, the first argument of electric vehicles is to reduce the use of fossil fuels, which have a disastrous impact on the environment in a context where global warming already represents a major existential threat.
The problem is that these magnets, which are essential to the democratization of electric cars, are generally produced from rare earthlike the famous neodymium. These are materials whose extraction is absolutely catastrophic for local ecosystems, in particular because these processes tend to contaminate the soil in an almost irreversible manner on our time scale. In addition, this supply chain consumes astronomical amounts of energy, is the source of serious geopolitical friction, and poses major ethical and humanitarian problems.
And the situation is unlikely to improve, knowing that analysts predict an explosion in the number of electric vehicles over the coming years. Many laboratories and institutions have therefore embarked on the pursuit of an alternative, with the objective of reducing this industry’s dependence on rare earths.
AI to the rescue
But materials science is an often thankless discipline because of its formidable complexity. To understand and exploit new materials, we must first explore their chemical, mechanical, thermal and electrical properties — extremely tedious background work. Add to this the economic considerations which further complicate the equation, and it is easy to understand why revolutions are so rare in this area; most often, we have to be content with small incremental advances. The startup Niron Magnetics, cited by New Atlas, is a good example. It succeeded in producing the first high-performance magnet without rare earths more than ten years ago… but the process is still not ready for mass production.
In recent years, materials scientists have therefore started to turn to machine learning to accelerate the discovery process, with already very encouraging results. At the end of 2022, researchers, for example, presented MG3Net, an AI model which made it possible to discover 31 million unknown theoretical materials – including some with potentially very interesting properties.
This is where Materials Nexus comes in, a startup that is betting heavily on this technology. Its engineers developed an AI model designed specifically to guide researchers toward a rare earth-free supermagnet. Using the combinatorial power of machine learning, this system screened tens of millions of possible chemical compositions while taking into account other important economic and environmental factors.
At the end of the process, this allowed the researchers to identify the theoretical recipe for such a magnet, which they named MagNex. After synthesizing and testing this material in partnership with the University of Sheffield, they were pleased to note that it is not only very efficient, but also that it would theoretically cost 20% less to mass produce. To top it all, the carbon footprint of this MagNex would be 70% lower than that of the traditional sector.
A whole new horizon for research
It remains to be seen whether this magnet will succeed in seducing the industry. But the most interesting thing is that it is a new example which perfectly illustrates the potential of machine learning in research — whether in materials sciences or elsewhere.
The more time passes, the more researchers have access to fantastic new tools that push the limits of disciplines such as particle physics or structural biology. To date, the best example is undoubtedly AlphaFoldthe DeepMind model that revolutionized protein science by producing what can be compared to a gigantic catalog of molecular origami.
At the time, Arthur D. Levinson, founder of a company associated with the project, saw it as an early example of a great paradigm shift in science. “ This shows how computational methods are set to transform research, and hold promise for accelerating the discovery process. “, he explained.
These new results once again confirm this prognosis. Beyond the work of Materials Nexus, it will therefore be appropriate to keep an eye on the democratization of artificial intelligence in the scientific sphere, because this technology could well be one of the engines that will allow our civilization to move into a new era.
🟣 To not miss any news on the WorldOfSoftware, subscribe on Google News. And if you love us, .