Seminar Johnny ZhangZhou

In this talk, I will present three related themes. First, I will discuss when AI can lead to scientific discovery and when it is likely to be misleading, with an emphasis on the complementary relationship between machine learning and thermodynamics. Second, I will present a petrological case study on sub‑arc mantle oxygen fugacity, using machine learning to distinguish arc from non‑arc samples and to extend this framework to older samples with less certain tectonic settings. Third, I will introduce recent work on structural similarities between minerals and protein active sites using AlphaFold‑derived structures, and discuss possible temporal parallels between mineral evolution and the emergence of biological metal‑binding sites, particularly around major transitions such as the last universal common ancestor and the Great Oxidation Event.