Exploring Opportunities, Epistemological Challenges, and Risks of Machine Learning in Volcano Science

Maurizio Petrelli¹, Mónica Ágreda-López¹

• Affiliations: ¹Department of Physics and Geology, University of Perugia 

• Presentation type: Poster

• Presentation time: Monday 16:30 - 18:30, Room Poster Hall

• Poster Board Number: 161

• Programme No: 3.1.53

• Theme 3 > Session 1

---

Abstract

Recent advancements in Machine Learning (ML), combined with significant improvements in computational capabilities, have brought substantial changes to scientific research. These developments have impacted numerous scientific fields, including physics, medicine, chemistry, mathematics, neuroscience, biochemistry, materials science, and engineering. In this work, we review the opportunities, epistemological challenges, and potential risks associated with the application of Machine Learning in Earth Sciences, with a particular focus on igneous petrology and volcanology. We emphasize the benefits of Machine Learning, especially in automating tasks, enhancing modeling strategies, and accelerating knowledge discovery. However, the integration of Machine Learning into scientific research also presents significant challenges. Key concerns include understanding what Machine Learning models actually learn, ensuring transparency, reproducibility, and improving model interpretability. These challenges become particularly critical in high-risk contexts such as volcanic hazard assessment, risk mitigation, and crisis management, where reliance on Machine Learning outputs can have profound consequences for human lives. Additionally, we address ethical considerations, such as the potential for over-reliance on Machine Learning models and the broader implications of geopolitical development plans, laws, and regulations in the EU, China, and the United States.