[ESS] Sophie Giffard-Roisin (ISTerre)

“AI in Geosciences: what’s left to do for academic research?”

Generative AI (GenAI), including large language and image models, is rapidly advancing and can produce text, images, and scientific content. However, many geoscience problems require specialized data, models, and physics-based reasoning that current foundation models cannot fully capture. I will explore how some foundation models, particularly for images and remote sensing, can be derived to geoscience tasks that are too specialized for commercial development. I will also show how problems without available ground truth can be addressed using realistic, physics-based simulated datasets. Concrete examples will be presented from geomorphology, remote sensing of Earth deformation, and volcanology.  This will open the discussion on how academic research could harness AI advances while staying open-access and domain-focused.

Earth Science Seminar —

Lecture portion of the EAPS graduate-level class 12.571, covering current research in geophysics, geology, geochemistry, and geobiology. All members of the MIT community are welcome to join for presentations by guest speakers, held approximately every two weeks during the term.

Contact: earth-science-seminar-info@mit.edu