DARPA seeks a few good AI coders to help America find its own rare minerals

There's some pocket change in it for you, if you can crack this nut

DARPA – the US government's boffinry nerve center – is offering up to $10,000 to programmers who can whip up some AI to help find rare earth minerals on our home world and ease US supply constraints of critical materials needed by the energy and defense industry.

To win the money, developers need to compete in a contest that's split into two separate tasks: the Map Georeferencing Challenge, in which maps of unknown regions and coordinate systems need to be pinpointed on a base map by a machine-learning model; and the Map Feature Extraction Challenge, in which an model needs to be trained to identify all the different kinds of polygons, points, lines, text, and other legends on maps.

These may seem removed from the process of finding critical minerals, though it's hoped the technology developed through these challenges can be eventually used be geologists to automate and speed up their search for vital materials.

Participants in the competition will be given a limited set of images to train their models. The winners can expect to receive $10,000 for first place, $3,000 for second place, and $1,000 for third, in each of the challenges. Given that competent AI developers can command fat salaries at the moment, this prize money is more of a small bonus than anything else, in our eyes.

"The USGS's critical mineral resource assessments are at the heart of our domestic supply and production of critical minerals," Anne Fischer, deputy director of DARPA's Defense Sciences Office, said in a statement this week, referring to the US Geological Survey.

"We want to have a measurable, immediate impact on the USGS's ability to reach some of its objectives, especially in ways that are critical to national security."

Using computational methods could help speed up the process of finding critical minerals, which traditionally requires manual inspections of images and reports by geologists. DARPA and USGS believe that automating some parts of the work means experts will have more time to tackle other issues related to easing supply constraints.

The other kind of supply chain security

Rare earth minerals are vital components in producing everyday objects. Lithium and cobalt, for example, are used to build computers, batteries, and solar panels. Lesser known elements with exotic names like lanthanum, samarium, or praseodymium are needed for night-vision goggles, nuclear reactor control rods, and aircraft engines. 

China tends to be the dominant source for these types of minerals, and the US increasingly relies on imports of the stuff to build the technology needed for its own national security. A report [PDF] reviewing American supply chains, compiled by the Biden administration last year, urged the federal government to strengthen national mineral supplies and rebuild the US industrial base.

Legislation such as the Energy Act of 2020 and the Bipartisan Infrastructure Law compelled the USGS to assess the country's sources of vital minerals and study mine wastes. The USGS published a list of 50 minerals considered critical earlier this year, and is also providing data for DARPA's AI for Critical Mineral Assessment Competition.

"The first step in any supply chain is the flow of naturally occurring materials from reservoirs in the Earth to human societies," Graham Lederer, a physical scientist with the USGS, told The Register.

"Information that helps characterize and quantify Earth's reservoirs, such as the amount and probability of occurrence of critical mineral resources in the United States, forms basic information needed before decisions can be made about how materials should be extracted, process, and used to produce goods that society deems important."

The competition will help both agencies figure out the capabilities of today's AI systems so they can best explore how they might be deployed in the future. 

"The outcome of a successful solution would be to greatly accelerate one bottleneck in the overall assessment process, allowing USGS to devote more effort to other components of the mineral resource assessment workflow, such as developing mappable criteria for mineral systems and deposit types, integrating varied datasets, and disseminating scientific information to the public," Lederer told us.

"By accelerating the integration of geological and mineral resource information embedded in thousands of individual maps this project will enhance the ability of the USGS to provide information in a timely manner. That said, improving supply chain security for critical mineral commodities involves many different stages including production, processing, and recycling as well as continued federal, state, and private sector engagement."

Registration for the Map Georeferencing Challenge opened on Monday and closes on August 26, while registration for the Map Feature Extraction Challenge will launch on August 29 and ends September 9. ®

PS: DARPA, through its SocialCyber program, is meanwhile probing the use of AI to detection things like social-engineering attacks on developers so as to get into their code and disrupt the software supply chain.