An artificial intelligence model trained on millions of fossil records suggests that three of Earth’s “Big Five” mass extinctions may have been gradual declines rather than sudden catastrophes. The AI fossil model, called OneStrata, was presented by Nanjing University palaeontologist Shuzhong Shen at the Geological Society of America meeting in Denver on Oct. 12, and Nature reported on the work on Oct. 20, 2026.
Image: An ammonite fossil preserved in limestone, used for illustration. Photo: James St. John, via Wikimedia Commons (CC BY 2.0).
How the OneStrata model works
Scientists read the history of life from layers of sedimentary rock, called strata. The depth at which a species’ fossils appear corresponds roughly to when it lived. The hard part is stitching together sequences from thousands of sites and tens of thousands of species into one global timeline.
Traditional methods split time into blocks of millions of years to keep the math manageable, which blurs short-term changes. A correlation method called CONOP can be more precise, but according to the team’s conference abstract, its computing cost becomes prohibitive once a dataset tops about 10,000 fossil species, even on modern supercomputers.
OneStrata takes a different approach. It is a transformer-based model, the same broad type of AI behind chatbots such as ChatGPT, but trained on fossil records instead of text, Nature explained. Per the abstract, it learns a shared probabilistic picture of geological time by combining the order of layers within each rock section, which species occur together, time windows and radiometric age dates.
A huge fossil dataset
According to the abstract, the team applied OneStrata to a global dataset of:
- 31,044 stratigraphic sections
- 183,514 species
- 2,845,815 fossil occurrences
- 449 high-precision ID-TIMS radiometric dates
The result is a reconstruction of marine biodiversity spanning about 600 million years, from the Ediacaran period to the early Neogene, at an average resolution of thousands of years. Nature described the training data as nearly four million fossils.
The project builds on two decades of work by Shen and colleagues compiling a database called OneStratigraphy. In a 2020 study in Science, the group used a supercomputer to reconstruct a timeline for 11,000 marine species over about 300 million years at a resolution of 26,000 years. That analysis found no sharp mass extinction in the Late Devonian, New Scientist reported at the time, but rather a long decline.
What it says about the Big Five mass extinctions
The Big Five are the end-Ordovician, Late Devonian, end-Permian, end-Triassic and end-Cretaceous extinctions, the last of which wiped out the non-bird dinosaurs. In OneStrata’s reconstruction, only the end-Ordovician and end-Permian events stand out as profound global collapses in marine biodiversity, the abstract says. The Late Devonian, end-Triassic and end-Cretaceous extinctions appear “comparatively weak or diffuse.”
That does not mean those events did not happen. The analysis looks at overall marine biodiversity, and a weaker signal there could reflect a slower decline spread over a longer stretch of time.
How other scientists reacted
Shen called the model “a real revolution for studying stratigraphy and past diversity change,” Nature reported. Other researchers were more cautious. They said OneStrata appears to be the most detailed global record yet, but that scientists already have high-resolution reconstructions for specific locations, including many that shed light on mass extinctions.
“We could do some extra things with it, but I don’t know if it would be Earth shattering,” Matthew Clapham, a palaeontologist at the University of California, Santa Cruz, told Nature. The work was presented at a conference; findings like these typically face further scrutiny when published in a peer-reviewed journal.
The team says OneStrata is meant as a general framework for studying Earth’s deep past, not just biodiversity, and could be expanded with environmental and geochemical data. The Boston Report has covered other AI-in-science stories, including a $15 million math and AI effort backed by the Simons Foundation, OpenAI and Anthropic and Japan’s MMX mission to bring back samples from Mars’ moon Phobos.
Sources
- Nature: Vast new fossil record challenges story of Earth’s ‘big five’ mass extinctions
- Geological Society of America: OneStrata conference abstract
- New Scientist (2020): AI suggests Earth has had fewer mass extinctions than we thought
- Wikimedia Commons: Ammonite in limestone (CC BY 2.0)
Read next in Artificial Intelligence
Like this story? Get the newsletter
The day's Boston news, weather and local business in one short email, with links to the full stories.
Most read this week
- Iran–U.S. War Update: Trump Warns of ‘Finishing the Job’ as Hormuz Attacks Climb and Lavrov Visits Tehran
- Keen Sword 27: U.S. and Japan Open Largest-Ever Island Defense Drills
- Fire Hits St. Petersburg Oil Terminal in Massive Drone Attack on Russia
- 100 Summer Street Auction Moved to Oct. 27
- Keen Sword 27 Begins: U.S. and Japan Open Island Defense Drills Near Taiwan
Latest
- ‘Glee’ Revival Ordered at Disney+ and Hulu: What We Know
- DHS Orders Federal Agencies to Remove DeepSeek AI From Their Systems
- Anthropic S-1 Draft Shows $16.3B Revenue and an $80.7B Loss
- OpenAI Models Come to Google Cloud as Microsoft Exclusivity Fades
- O Ya Boston: Leather District Omakase Guide


