Berkeley professor protests rejection of toxic computing

The Department of Defense and the US Geological Survey have launched two separate challenges to explore using artificial intelligence and machine learning to accelerate the USGS’s task of assessing the availability and mining potential of 50 critical minerals . Challenges enlist the help of outside researchers and AI developers to solve a tricky problem.

The end goal of the challenges isn’t to fully automate assessments, said Graham Lederer, a research geologist at the USGS Geology, Energy and Minerals Science Center. Rather, it’s about finding a way to use AI to speed up the process of digitizing long-buried maps and other data and using it to understand the country’s mineral resource landscape.

“Right now, we’re basically processing one product at a time or one type of mineral deposit at a time, because it’s so manual and requires a lot of human effort,” Lederer said. “But if you configure the machine learning algorithms well and generalize the solutions well, [we] could make all 50 products simultaneously, which would transform the way we work and the pace at which we work.

A map of the locations of critical mineral deposits in the United StatesImage: USGS

Although the government’s geologists and its military technology researchers may seem odd, the collaboration could allow the former to speed up geological assessments that are the first step in building a critical national supply chain. Given the national security implications as well as the need to deploy clean energy technology as quickly as possible, this mission takes on added urgency.

“Part of the problem is that we collected so much data 70, 80, 90 years ago, before digital data collection was possible and effective,” said Joshua Elliott, program manager working on innovation. information to DARPA, the research arm of the DOD. “We’re a little behind because we were so early. Much of our data is actually locked away in these incredibly valuable and incredibly detailed maps that have been produced by [USGS experts] two generations ago.

There are essentially two steps – corresponding to the two challenges – that are required to actually use existing maps to locate minerals. The first is, Elliott said, “extremely accurate georeferencing” to digitize hard copies and scans of maps into a usable format. Next, researchers must extract important features such as fault lines and geological formations to determine where mineral deposits might be. By hand, it would be an extremely time-consuming process, but Elliott said AI and machine learning have the potential to “significantly reduce” the time it could take.

A before and after image of an 1894 economic geology map of Jackson Peak, California that has been digitized.

A before and after image of an 1894 economic geology map of Jackson Peak, California that has been digitized.Image: USGS/DARPA; Protocol

At present, Elliott said, agencies estimate that about 10% of the USGS’s core database of 100,000 maps has even been partially digitized, and an even smaller share has had the data mined. in a usable way.

But time is running out. The 2020 Energy Act set the target of completing all critical mineral assessments within four years. This is the time it takes for the USGS to assess the availability of a single mineral.

“That requires an order of magnitude increase in efficiency,” Lederer said. “We have a lot of tools, but we need to scale those tools to really be able to do that and effectively, and that’s where the AI ​​and machine learning efforts come in.”

It’s not an easy task, however, according to Asitang Mishra, senior data scientist for the AI ​​and Analytics Group at NASA’s Jet Propulsion Laboratory, who has taken a close look at the kind of AI tools that might be needed. AI or machine learning is well suited for highly repetitive analysis that would take hours or even days, he said, which bodes well for the new effort.

The challenge is an opportunity to tap into the expertise of researchers who have the skills to use AI to digitize existing data or extract relevant features and “exploit the broader segment of the ecosystem of innovation that exists,” Elliott said. This could include private sector mining companies, academic research groups or even individuals. For each challenge, DARPA is offering $10,000 for first prize, $3,000 for second, and $1,000 for third.

For the georeferencing challenge, participating groups and individuals will receive 1,000 or more maps for the AI ​​to learn. To be successful, the challenge announcement states that teams will need to “precisely geolocate a map of an unknown location and coordinate system by adjusting coordinate points that can be referenced to known locations in one or more maps of base”.

For the Feature Extraction Challenge, participants will receive labeled maps and legends. They will use these to train the AI ​​to identify features that could be used to identify mineral locations.

Even if no clear answer to the USGS predicament emerges from the challenges, Mishra said at least the agencies will have a better idea of ​​the gap in using machine learning for these tasks.

Ilya Jackson, a postdoctoral associate focused on applying AI to supply chains at the MIT Center for Transportation and Logistics who is not affiliated with the challenge, said the effort is “a step in the right direction. ” and “worth a try”, but its success will depend on the quality of the data.

“If you have a project that has a lot of high-quality data, AI will work and be easy to implement, Jackson said.

Ultimately, DARPA’s plan is to create a grand, multi-year challenge that will involve enlisting teams that are already familiar with the task of scanning and map mining to use AI and machine learning to do this work. This first step, Mishra said, is “simply understanding the waters.”

The highest priority minerals are those necessary for the energy transition, and in particular for batteries. Nickel, manganese, lithium and graphite are vital in this respect, and the competition for these on a global scale has become increasingly tough. China, Elliott said, has been actively ramping up its mining capacity and now dominates the supply chains of many of them.

World map of mineral imports.

World map of mineral imports.Image: USGS

“There are certainly concerns that have been expressed within the US government about whether or not our reliance on them creates a strategic disadvantage,” Elliott said. Russia is also a major source of critical minerals, as are other countries without democratically elected leaders, which could create further supply chain vulnerabilities. In the Democratic Republic of Congo, for example, the two largest cobalt mines were once owned by American companies, but have since been sold to Chinese companies, compromising the strategic availability of these minerals. One such company is being investigated for allegedly flouting royalty payments.

The United States building its own supply chain of critical minerals could mitigate some of these risks and help its allies accelerate their own transition to clean energy.

“The information needed to make downstream decisions about where mining should take place, where should production take place? All of this needs to be informed of where the resources are or where they could be,” Lederer said.

AI could help answer these questions sooner rather than later. DARPA will announce the challenge winners next month.

Comments are closed.