AI finds rare-earth-free magnet candidates (materials at scale)

47:35permanent magnets i.e. permanent magnets that don't require these rare earth me minerals that like we have to mine for that are really hard to find that maybe we have to make a deal with China with and so on and so forth like that's crucial for reducing the manufacturing cost and the supply chain vulnerabilities of electric vehicles and electric vehicles if we can make them cheaper easier to make they can be the future right not just electric vehicles I mean batteries go in everything right medical devices renewable energy systems And what this team at the University of New Hampshire has done is use an AI system to extract data from thousands of scientific reports and train a model

48:16that will predict the magnetic behavior of materials of novel materials, right? And you you have like now a searchable database of something like 67,000 magnetic compounds. And out of those they found 25 new candidates that will remain magnetic at high temperatures. >> Very interesting. Yeah, because the the hotter you make a magnet, the more jiggling there is, which means that the local alignment between your magnetic elements. Like in iron, for example, the iron has like a veence electron that creates a gross magnetic field around the atom. Why? When iron becomes magnetic, like when you take a paperclip and you put it around a permanent magnet, that paperclip becomes magnetic,

48:58right? Because all of the iron atoms sort of >> align to be make a big magnetic field. Well, if you heat up that paperclip, you're done. But if you put that paper clip in the freezer and you come back hours later, it'll still be magnetic. The point is, we want to find magnetic materials that remain magnetic at high temperatures because that will give us industry applications, right? And that's what this is doing. And it it vastly accelerates material science. >> Um, you know, because millions of time consuming lab tests can now be churned through with this machine learning algorithm. and you can find like a searchable digital resource that you can then be like okay I want to find a

49:38particular magnetic material that does this and this you got something >> this is this is great so this was out of nature communications would it be >> so it the I mean obviously the thing that comes to my mind is >> it makes me think of like a magnetic version of alpha fold >> um yeah kind of is it is >> it's like a material science version of of of it like directionally speaking details um I I think it's it's it's still a little far from I give you a novel material and you predict this is still there's all of these um there's all of these thousands of scientific papers that have already come out with a bunch of stuff. How do I centralize that and create an understanding?

50:18>> So so rather so it's more so taking the existing data and making it uh making someone be able to iterate on ideas based on existing data. Mhm. >> Alphafold was really sort of saying here's all possible protein structures. >> Yeah. From all of the sequences that we know, even the ones that we haven't found, >> which is a different thing cuz that's generating net new data. >> That's like versus just interpreting past data. Okay. So, there's a there's a difference there. >> Yeah. But it's it's getting there, right? We're getting close. >> That's very cool. Very very very very good. >> Yeah. >> Story number three is a climate science story with a bit of anthropology. >> Yeah. Uh there's new science about what's actually what actually went wrong

51:00on Easter Island. >> Yes. >> What went wrong? >> So Easter Island is very famous, right? These this is Rapanui. It has those um giant statues of um these like heads with torsos that are facing out into the ocean. And for the longest time, the