A New App Can Match Footprints To The Dinosaurs That Made Them
Source: Smithsonian Magazine, Mary Randolph
Photo: DinoTracker was trained on almost 2,000 dinosaur fossils to classify new tracks. It is especially helpful for three-toed dinosaurs, as so many tracks fall under this umbrella, co-author Paige dePolo writes in The Conversation. (James St. John/Wikimedia Commons)
Using artificial intelligence, DinoTracker can accurately classify dinosaur tracks around 90 percent of the time
New inventions can come from unexpected places. After reading a book on the history of dinosaurs to his son, physicist Gregor Hartmann realized that the artificial intelligence methods he worked on in photon science might have an application in paleontology. He reached out to the book’s author, paleontologist Steve Brusatte, with the idea, and DinoTracker was born. The new model uses A.I. to classify fossilized tracks to different dinosaur species.
Hartmann’s team published the model’s findings in the Proceedings of the National Academy of Sciences last week, and the DinoTracker app is now available on GitHub.
Until now, most paleontologists have identified dinosaur tracks in fossils manually or based on previous classifications, which can sometimes lead to biases or misclassifications, Hartmann tells the Guardian’s Nicola Davis.
“You never find a footprint and alongside [it] the dinosaur that had made this footprint,” Hartmann says. “So, no offense to palaeontologists and such, but most likely some of these labels are wrong.”
Unlike previous models, DinoTracker was trained on almost 2,000 unlabeled dinosaur footprints, which it sorted into classifications based on eight key characteristics. These included spread of toes, amount of ground contact and heel position. After over a year of training the neural network, the model aligned with human classifications around 90 percent of the time.
“I wish there was an app like this when I first started studying dinosaur tracks,” Brusatte, who co-authored the report in PNAS, says to Rachael Funnell at IFL Science. “It really is challenging to understand the variation among tracks that were made by different dinosaurs and preserved in different environments, and this app now makes everything more objective.”
Because the model’s only input is the track itself, human experts still oversee and confirm the classification based on additional context of age and location.
“This app certainly isn’t the end of the story when it comes to puzzling over the mysteries of dinosaur footprints,” Paige dePolo, the report’s co-author, writes in the Conversation. “It’s a useful research resource for figuring out what tracks any footprint is most similar to in terms of shape, and what features are driving that similarity.”
DinoTracker also supported paleontologists’ previous observations that some three-toed dinosaur tracks are remarkably similar to bird tracks. This confirmation from an unbiased, computational source is significant, Brusatte tells IFLScience
“Our dinosaur footprint A.I. model shows that some of these mysterious, controversial three-toed Triassic tracks really do resemble those of birds,” he says. “The humans studying them were correct. It wasn’t just wishful thinking, or that they were seeing a bird shape in the tracks in the same way somebody imagines the face of Jesus on a slice of toast.”
This raises the hypothesis that birds or bird ancestors originated nearly 60 million years before scientists previously thought. But tracks are not a perfect representation of the creature that made them, dePolo writes in the Conversation.
Fun fact: Record-setting site
Bolivia’s Torotoro National Park is home to 16,600 exposed three-toed dinosaur footprints—the most ever found in one location.
“Dinosaur footprints are not perfect snapshots of the feet that made them,” she writes. “They reflect the shape of the foot, how the dinosaur was moving, and how soft or hard the ground was at the time.”
Jens Lallensack, a geoscientist at Humboldt University of Berlin, who has also used A.I. to help identify dinosaur tracks but was not involved in the study, tells the Guardian that the birdlike tracks do not directly reflect the shape of the foot and “are not evidence for an early appearance of birds.”
The scholars have made the model publicly available as an app on GitHub for both expert and amateur dinosaur lovers.
“I think A.I. has a bright future in paleontology,” Brusatte tells IFLScience. “It’s not that A.I. will become some all-knowing god that can identify every single fragment of dinosaur bone or tell us exactly where to find every fossil in the rocks. But what excites me most is that A.I. can become a new type of paleontologist, one that compiles and observes and filters through and classifies data and does so in a way that is free from the usual human biases.”