A new artificial intelligence model, OneStrata, has analyzed millions of fossils spanning 540 million years and suggests that three of Earth's five major mass extinctions may have been gradual declines rather than sudden catastrophes. The research, presented on October 12, 2026, at a meeting of the Geological Society of America and covered by Nature, challenges long-held assumptions about the nature of these events.
OneStrata, developed by a team led by Shuzhong Shen of Nanjing University, uses a transformer architecture similar to large language models (LLMs) to analyze fossil distribution patterns in rock layers. The model was trained on a global database called OneStratigraphy, which includes 31,044 stratigraphic sections, 183,514 species, and approximately 2.85 million fossil records, along with 449 high-precision dates. This allowed the AI to reconstruct the evolution of over 110,000 marine species with a temporal resolution estimated at around a thousand years.
The findings are preliminary but raise important questions about how scientists reconstruct Earth's past. The model's ability to process vast datasets more efficiently than traditional methods could transform paleontological research. However, the results highlight that some catastrophic events may have been more complex than simple, sudden disasters.
