Primary Convener:
Muthukumaran Ramasubramanian, University of Alabama in Huntsville

Convener:
Grant Boquet, Organization Not Listed
Armin Mehrabian, NASA Goddard Space Flight Center, ADNET Systems Inc.
Jieping Ye, Zhejiang Laboratory
Bishwaranjan Bhattacharjee, IBM Research USA
Geeth De Mel, IBM Research UK

Chair:
Grant Boquet, Organization Not Listed
Geeth De Mel, IBM Research UK

Large language models (LLMs) and agentic workflows are revolutionising science, driving advances in literature discovery, data analysis, coding, and cross-domain synthesis. This session explores both the opportunities and the challenges of applying these tools, with emphasis on geosciences, where LLMs are rapidly becoming essential research instruments.

We invite discussions on safety, integrity, and governance. Key themes include specialised guardrails against hallucinations and bias, rigorous evaluation with uncertainty quantification, reproducibility, data governance, and human-in-the-loop approaches. Presentations addressing ethical, fairness, and responsible AI practices are also encouraged. Equally important are contributions showcasing real-world applications such as taxonomy, data curation, geologic mapping, environmental protection, resource exploration, hazard forecasting, and disaster response. Domain-specific systems like GeoGPT and other tailored models are particularly welcome.

By combining case studies with lessons learned, this session fosters collaboration between AI researchers and domain scientists, offering best practices for integrating LLMs into broader scientific decision-making tasks.

Index Terms
1916 Data and information discovery
1942 Machine learning
1954 Natural language processing

Suggested Itineraries:
Machine Learning and AI
Open Science and Open Data

Cross-Listed:
V - Volcanology‚ Geochemistry and Petrology
EP - Earth and Planetary Surface Processes
GC - Global Environmental Change

Neighborhoods:
1. Science Nexus