The Korea Institute of Geoscience and Mineral Resources (KIGAM) has begun developing artificial intelligence agents to analyze complex geological data. The goal is to build six AI agents specialized in tasks such as sinkhole risk prediction, mineral exploration and geological cross-section generation, along with a system to operate them in an integrated manner — improving analytical accuracy while cutting processing time.
Geological data is considered particularly difficult to apply digital transformation (DX) and AI transformation (AX) to. Because it involves probing deep underground, large-scale data accumulation is harder than with general industrial data, and numerous variables are at play. Korea's geology is especially challenging to analyze, as repeated plate collisions, compressions and deformations over eons have left strata ranging from more than 2 billion years old to young formations created only tens of millions of years ago.
To overcome these limitations, KIGAM is developing an "integrated AI agent operating framework." Kwon Ji-hoe, head of the Geo-AI Convergence Research Laboratory, said, "Geological data is a field that is both scientific and hermeneutic or humanistic in nature. Since we cannot directly look underground, researchers must infer conditions by combining surface information, sensor signals and past records — and we are progressively building the digital infrastructure so that AI can assist this process."
For five years starting in 2020, KIGAM has laid the groundwork by gathering geological maps, drilling data, groundwater records and seismic waveform data in one place. The Geo Big Data Open Platform now holds 3,012 datasets, 118 thematic maps on geological resources and 133,674 reports and papers.
Building on the Geo AI Platform established in 2024, KIGAM began developing the six AI agents this year. The Geo AI Platform is a system on which researchers collect and process data and develop, train and validate AI models. The specialized infrastructure was built in recognition that data in the geological resources field is too heterogeneous to be directly applied to general-purpose AI. The six AI agents will be developed around areas with strong interconnectivity and utility.
Over the medium to long term, KIGAM will also complete an AI orchestrator called "GAKAAI" that links and operates the AI agents in an integrated fashion. When a user requests a geological analysis of a specific area, the system will automatically collect the required data and call upon the relevant specialized AI agents to present comprehensive results.
Kwon assessed the current state of the AI, however, as "a junior researcher who has built up a knowledge base but is still refining hands-on field interpretation know-how." She added, "Given the high data complexity, the optimization process that enables AI to fully support the insights of field experts is critical. Our goal is to advance the integrated system around 2030 and unveil a fully fledged AI partner."
