Skip to main content

Scientists asked A.I. to analyze the fossil record. This is what it found

There have been a total of five extinction events in Earth’s history. The first, the Ordovician-Silurian extinction, took place 440 million years ago, wiping out small marine organisms. A second, the Devonian extinction, occurred 365 million years ago, killing off an assortment of tropical marine species. After that, 250 million years ago, was the Permian-Triassic extinction, followed by Triassic-Jurassic extinction 210 million years ago, and the Cretaceous-Tertiary extinction 65 million years ago. Things have been just about OK in the years since then — even if 2020 has done its very best to make us feel otherwise.

Scientists have long hypothesized that mass extinctions result in productive periods of evolution, a process sometimes referred to as creative destruction or “radiations.” The idea is that these periods in which large numbers of species disappear correlates with the arrival of new species.

To explore how well the data supports these theories, an international group of researchers applied machine learning technology to the immense fossil record of life’s history, visualizing their structure with an algorithm that embedded 171,231 species within a multidimensional virtual space. The results helped the researchers work out whether certain species coexisted in time or never co-occurred.

The machine learning algorithm was created by Nicholas Guttenberg of GoodAI and Cross Labs/Cross Compass in collaboration with with Tokyo Institute of Technology.

“Five exceptionally bad extinctions were known to have occurred during the last eon of life’s history, and it’s hypothesized that we may be entering another,” Jennifer Hoyal Cuthill, an evolutionary biologist at the Institute of Analytics and Data Science and School of Life Sciences at the University of Essex in the U.K., which is affiliated with the Tokyo Institute of Technology’s Earth-Life Science Institute (ELSI), told Digital Trends. “Our machine learning method … picks up these ‘big five’ mass extinction events and allows us to visualize their significant impacts. But our study also puts the big five mass extinctions in the context of all the detectable disturbances in the fossil record. This shows the range of evolutionary events that have impacted life’s history.”

Hoyal Cuthill said that the work has opened up some exciting areas of possible research. In particular, it has helped counter some of the popular wisdom about mass extinctions and mass radiations, showing that the creation of large numbers of species and the extinction of others frequently take place at different times.

“Our new methods have potential applications across a wide range of data, and we have more projects planned,” she said. “Overall, there are so many areas of human knowledge where machine learning has not yet been applied and so, potentially, many great insights are waiting to be made right now. I think this is a very exciting time.”

A paper describing the work was recently published in the journal Nature.

Editors' Recommendations