Designing Agentic AI for Subsurface Geoscience:
Concepts, Scientific Principles, and Lessons from GeoVerse Lab
by Prof. Ju-won Oh
English
1/12/2026 | 0900 - 1230
KIGAM
Course Description
This half-day course introduces the concepts and design principles of agentic AI for subsurface geoscience. It examines how agentic systems differ from conventional machine-learning models, general-purpose large language models, and conversational AI assistants, and discusses how agents may coordinate information, data, analytical tools, and scientific tasks under human supervision.
The course is lecture- and demonstration-based rather than a coding or system-building workshop.GeoVerse Lab (www.geoverselab.com), an agentic research ecosystem developed by the instructor, will be presented as an illustrative case study. Selected demonstrations will be used to explain the motivations, design choices, scientific workflows, and governance principles behind an agent-based geoscience research environment.
GeoVerse Lab is not presented as a standard or universal reference architecture. Instead, it provides one practical perspective from which participants can examine alternative ways of organizing agentic systems. The emphasis will be on transferable concepts, including task decomposition, human-agent collaboration, scientific validation, provenance, reproducibility, accumulated knowledge, and responsible autonomy.
Participants will be encouraged to critically consider which ideas may be relevant to their own subsurface problems and how different scientific, organizational, and technical conditions may lead to different system designs.
Learning objectives
By the end of the course, participants will be able to:
Explain the differences between conventional AI/ML models, LLM-based assistants, and agentic AI systems.
Identify the types of subsurface and geoscientific tasks that may or may not be suitable for agentic approaches.
Describe the conceptual components and information flows that may be involved in an agent-based scientific workflow.
Explain key design considerations for coordinating scientific knowledge, data, analytical tools, and human decisions.
Recognize major scientific risks, including hallucination, weak provenance, non-reproducible results, hidden assumptions, and inappropriate automation.
Discuss how validation procedures, human oversight, approval gates, and accumulated research knowledge can be incorporated into an agentic research environment.
Distinguish between generalizable design principles and implementation-specific choices demonstrated through the GeoVerse Lab case study.
Formulate relevant questions and design considerations for developing or evaluating their own agentic AI systems.
Target Participant Profile
This course is intended for:
Geoscientists, geophysicists, geologists, and subsurface engineers interested in the emerging role of agentic AI.
Professionals working in oil and gas, mineral exploration, geothermal energy, carbon storage, and related subsurface domains.
Researchers and graduate students exploring AI-assisted scientific workflows.
Data scientists and software professionals collaborating with geoscience teams.
Technical and research managers considering the opportunities, limitations, and governance of agentic AI.
The course is particularly suitable for participants who wish to understand the conceptual foundations and design choices behind agentic systems without requiring a coding-intensive treatment.
Prerequisites
No programming or prior experience in building AI agents is required.
- General familiarity with a geoscience, geophysical, or subsurface research or operational workflow.
- An interest in how AI may support scientific analysis, knowledge management, or decision-making.
- Basic familiarity with AI/ML or large language models will be helpful but is not essential. The course will focus on concepts, demonstrations, and design reasoning rather than mathematical derivations or coding exercises.
- To gain some context before the course, participants are encouraged, but not required, to explore the GeoVerse Lab public website at www.geoverselab.com. The website provides an introductory view of the lab’s evolving multi-agent research environment, including its organizational concept, AI researchers, research areas, and research infrastructure. No prior experience with or access to the GeoVerse system is required.
PROF. JU-WON OH
Professor, Department of Mineral Resources and Energy Engineering
Founder and Principal Investigator, GeoVerse Lab
Jeonbuk National University
Prof. Ju-Won Oh is a Professor in the Department of Mineral Resources and Energy Engineering at Jeonbuk National University, Republic of Korea, and the founder and principal investigator of GeoVerse Lab.
His research spans geophysical exploration, seismic imaging and full-waveform inversion, high-performance geophysical computing, machine learning for geophysical data processing, and agentic AI for geoscience.
He received his B.Sc. in Earth and Environmental Sciences and Ph.D. in Energy Resources Engineering from Seoul National University.
Before joining Jeonbuk National University in 2017, he conducted postdoctoral research at Seoul National University and King Abdullah University of Science and Technology (KAUST).
He currently leads the development of GeoVerse Lab, a multi-agent research environment that explores how human researchers and AI agents can collaborate across scientific literature,
geoscience data, analysis, validation, and knowledge accumulation.
His current work focuses particularly on scientifically trustworthy, traceable, reproducible, and human-governed agentic AI workflows for Earth and subsurface research.
