AI Is Transforming How the United States Hunts for Critical Minerals
The United States is ramping up its efforts to secure stable supplies of critical minerals—the elements essential for clean energy technologies, defense systems, and advanced electronics. These minerals include well-known resources like lithium, cobalt, and nickel, along with dozens of other strategically important elements. But America faces a major challenge: a large portion of these resources is imported from regions that are either geopolitically unstable or potential adversaries. To solve this problem, a wide-ranging initiative called Critical Mineral Assessments with AI Support (CriticalMAAS) has been launched to speed up the discovery and evaluation of mineral deposits across the country.
This initiative is backed by the U.S. Geological Survey (USGS), the Defense Advanced Research Projects Agency (DARPA), and the Advanced Research Projects Agency–Energy (ARPA-E). A key part of the effort involves researchers from the USC Viterbi Information Sciences Institute (ISI) and the University of Minnesota, who are building AI-powered tools capable of turning vast amounts of old geological data into modern, searchable, and analyzable resources. Their work is reshaping how mineral assessments are conducted—shifting the process from slow and labor-intensive to fast and scalable.
Automating Geological Map Digitization
A major obstacle in mineral assessment is the enormous archive of over 100,000 historical geological maps produced by the USGS. These maps contain rich information about rock formations, faults, mineral deposits, and other geological features. However, most of them exist only as scanned raster images, which cannot be easily interpreted by computers.
Traditionally, experts would manually digitize these maps, translating symbols, colors, and boundaries into structured data. This can take hours per map. Existing automated tools struggle with geological maps because they vary widely in color schemes, symbols, and labeling styles, and they often mix complex visual elements.
To tackle this, the research teams developed DIGMAPPER, a modular, AI-driven system that automates the entire digitization process. DIGMAPPER identifies polygon features, line structures, point symbols, and interpretive text directly from the maps. It also handles tricky visual elements such as overlapping symbols, small text labels, inconsistent color usage, and unusual shapes in geological units. When tested on more than 100 annotated maps from a DARPA–USGS dataset, it completed digitization in under 25 minutes per map, a massive improvement over the manual approach.
By converting scanned images into structured data, DIGMAPPER unlocks the vast geological knowledge stored in historical maps. This allows researchers and analysts to combine old and new data, conduct large-scale evaluations, and accelerate the identification of areas where critical minerals may be found.
Building a Global Mineral Knowledge Graph
Another major bottleneck in mineral discovery is that information about mining sites worldwide is scattered across thousands of documents—academic papers, government reports, industry filings, and geology databases. These sources contain details about mineral deposits, grades, tonnages, geological settings, and production histories, but the formats vary enormously: long PDFs, complex tables, dense terminology, inconsistent naming schemes, and more.
To unify this, ISI researchers created MinMod, an AI-powered system that compiles and standardizes mineral-site information into a large, machine-readable knowledge graph. MinMod has already processed tens of thousands of documents covering over 679,000 mining sites and 190 different mineral commodities, making it one of the largest publicly available mineral datasets in the world.
The AI extracts relevant details, resolves terminology differences, and organizes the information so that researchers and policymakers can query it effectively. One of MinMod’s most powerful abilities is generating grade-and-tonnage models, which highlight which geological deposit types historically produce the most metal. These models help prioritize exploration efforts, guide national mineral strategies, and point analysts to locations where undiscovered deposits are likely to exist.
By creating this unified framework, MinMod enables geologists, policymakers, and AI systems to work from the same harmonized dataset—something that was extremely difficult before.
Accelerating the National Mineral Assessment Process
The overarching goal of CriticalMAAS is to dramatically speed up the mineral assessment workflow. Under traditional methods, conducting a single assessment could take two years. But with increasing demand and expanding lists of critical minerals, Congress wants the USGS to complete 50 assessments within the next few years.
AI systems like DIGMAPPER and MinMod form the foundation for machine-learning models that can assist analysts in predicting where critical minerals may be located. These tools don’t replace geologists, but they empower them by processing massive datasets automatically, handling repetitive tasks, and revealing patterns that would be impossible to detect manually.
Why Critical Minerals Matter
As the world moves toward green energy, the demand for minerals needed for batteries, electric vehicles, renewable-energy storage, microelectronics, and defense systems is rising sharply. For example:
- Lithium and nickel are essential for electric-vehicle batteries.
- Cobalt is vital for aerospace applications and energy storage.
- Rare earth elements are necessary for wind turbines, electric motors, and precision-guided military systems.
Many of these minerals are currently sourced from just a few countries, which creates supply risks. For instance, a large portion of rare earth elements comes from regions that present geopolitical challenges. Strengthening domestic mineral discovery helps reduce these dependencies and supports economic and national-security goals.
The Broader Scientific Impact
Beyond mineral exploration, the technologies developed in CriticalMAAS can benefit other fields:
Geospatial Science
Automated map digitization can support environmental monitoring, land-use planning, hazard analysis, and more.
Knowledge Graph Research
MinMod demonstrates how AI can organize unstructured scientific literature into interconnected, searchable networks.
Machine Learning for Earth Sciences
The project pushes the boundaries of using AI to interpret complex geological patterns that traditionally required decades of expertise.
What Comes Next
The DIGMAPPER project will be presented at the ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, held November 4–7, 2025, in Minneapolis, Minnesota. MinMod will appear in two papers at the International Semantic Web Conference (ISWC) from November 10–14, 2025, in Nara, Japan.
These tools are becoming part of a larger pipeline aimed at predicting where undiscovered critical mineral deposits may lie. By combining historical maps, global mining datasets, and modern AI techniques, researchers hope to create models capable of guiding future mineral exploration with far greater accuracy and speed.
The long-term vision is clear: use AI to transform every stage of the mineral-assessment process—from digitizing old maps to identifying new deposit targets—ultimately helping the U.S. secure a stable supply of the minerals that power the technologies of tomorrow.
Research Paper:
DIGMAPPER: A Modular System for Automated Geologic Map Digitization
https://doi.org/10.48550/arxiv.2506.16006
