DARPA wants to use AI to find new rare minerals

Secure access to rare Earth minerals is a critical national security issue, as the entire United States' economy is highly dependent on minerals—and the majority of them discovered so far are in China. The Defense Advanced Research Projects Agency, or DARPA, has partnered with a company called HyperSpectral that applies artificial intelligence to spectroscopic data, which could be key to using satellites or drones to find minerals that would be difficult to detect otherwise.

HyperSpectral CEO Matt Thereur explained how it works in an exclusive interview with Defense One. Spectroscopy is the study of how matter interacts with light or other forms of radiation across different wavelengths. The solar radiation a specific mineral or substance emits, due to its unique molecular makeup, is a unique signifier. 

The company until now has focused on food safety. Want to find out if large shipments of raw food are carrying deadly pathogens? Want to know about the new breakout of medication-resistant strep? Spectroscopy can help find the bacteria the eye can’t see. 

“The processing used today takes a couple of days for someone to be able to tell the difference between [drug-resistant and drug-sensitive staphylococcus bacteria], because they actually have to plate and grow the bacteria and then apply antibiotics to it to see which ones kill it, if any. Versus now we're talking about a swab, say from a wound, and we're typically turning results around in a couple of minutes versus several days.”

Where does AI come in? Thereur explained, “Pure samples don't exist in nature. Nature is a very noisy place. So what we're doing with artificial intelligence when we build these models is looking for all the relationships that can sometimes be obscured by the noise [such as] if you've got one section of the spectrum being confounded by some other substance within it.”

There are also multiple types of spectroscopic analysis that aren’t easily combined together in a single data picture, which is also where AI helps. The auditory data that comes from human speech is very different from text data related to what combinations of letters and words are most likely to show up together. But combining them together is what makes AI-driven transcription and translation possible. In theory, spectroscopic data from a wide variety of sources could be just as useful. 

“Whether it's absorbance, or reflectance or [Fourier Transform Infrared Spectroscopy] or Raman and or surface-enhanced Raman, it's all about understanding the spectrographic response of those materials and being able to differentiate between different materials,” Thereur said.

What can it reveal? Thereur said the DEA, used a similar technique and, “was able to tell the difference of cocaine that came from one cartel’s area of Colombia versus another.” 

The cooperative agreement with DARPA is in its very early stages, Thereur said, and the Defense Department applications for better understanding where different materials might be are vast. Spectroscopy can be done with a few specific satellites, which makes it potentially useful for intelligence collection—such as finding the existence of specific materials used in enemy or adversary equipment or vehicles. 

The Pentagon is keen on not only getting better access to rare Earth material, but also moving the building of key weapons and supplies much closer to the front lines, rather than rely on supply lines that would be very hard to defend in the Pacific. 

“There is a tremendous amount of applicability and use cases for analysis of spectral data. Yes, there's a tremendous amount,”  Thereur said.