Perspectives: LLM agents reshaping the foundation of geotechnical problem-solving

Abstract

• Discuss potential of Agentic AI application in Geotechnics. • Review the 1st Geotechathon event. • Emphasize the importance of community effort for the future of Geotechnics. This paper explores the transformative potential of Large Language Model (LLM)-based agentic artificial intelligence (AI) in addressing longstanding challenges in geotechnical engineering. It begins by highlighting the significant growth and increasing interest in applying machine learning (ML) and AI techniques across various geotechnical domains, such as soil classification, slope stability analysis, and foundation design. Emphasizing the Gartner Hype Cycle, the authors reflect on the transition from initial enthusiasm toward realistic appraisal and adoption, highlighting current barriers like limited foundational understanding, skepticism about AI reliability, and a lack of standardized practices. The authors then introduce LLM agents as promising solutions for automating the extraction, interpretation, and quantification of qualitative and semi-quantitative geotechnical data. Drawing insights from the 1st GeoTechathon event, an international collaboration involving engineers, data scientists, and AI practitioners, the paper demonstrates practical applications in geotechnical site planning, landslide investigations, liquefaction analysis, and shield tunnel safety evaluation. Each project leveraged basic techniques, including Retrieval-Augmented Generation (RAG), multimodal data integration, and prompt engineering, achieving improvements in efficiency, accuracy, and decision-making processes. The paper concludes by discussing broader implications for interdisciplinary collaboration, ethical considerations, and future directions, emphasizing the necessity for standardized practices, rigorous validation, and enhanced AI literacy to sustainably integrate LLM technologies within the geotechnical engineering community.

Topics & Concepts