An international team of researchers has, for the first time, used artificial intelligence to analyze dinosaur tracks, and AI has come out on top – beating trained paleontologists at their own game.
“In extreme examples of theropod and ornithopod footprints, it is easy to distinguish the shapes of their footprints — a theropod with long, narrow toes and an ornithopod with short, narrow toes. The pathways between these shapes are not entirely clear from the toes,” said one of the researchers, University of Queensland paleontologist Dr Anthony Romelio. where it was made. Cosmos.
“We wanted to see if the AI could learn these differences, and if so, it could be tested in distinguishing between the three most difficult fingerprints.”
Theropods are meat-eating dinosaurs, while ornithopods eat plants, and doing this analysis wrong could change the data showing the diversity and abundance of dinosaurs in the area, or even could change what we think of as the behaviors of some dinosaurs.
One set of dinosaur prints in particular has been a struggle for the researchers to analyze. Large footprints at Queensland’s Dinosaur Stampede National Monument have divided Romelio and his colleagues. The mysterious relics are believed to have been left during the middle Cretaceous period, about 93 million years ago, and could be from either a flesh-eating theropod or a plant-eating ornithopod.
“I consider them to be the footprints of an herbivore while my colleagues share the broader consensus that they are the tracks of a theropod.”
Therefore, an artificial intelligence system called a convolutional neutral network was introduced to be a decisive factor.
“We were pretty stuck, so thank God for modern technology,” he says. Dr. Jens Lalensaclead author from Liverpool John Moores University in the UK.
“In our research team of three, there was one person who was pro-carnivore, one person who was reluctant, and one who was pro-plant.
“So — to really check out our science — we decided to go to five experts to demonstrate, as well as use artificial intelligence.”
The AI was given nearly 1,500 already known paths to figure out which dinosaur it was. The tracks were simple line drawings to facilitate AI analysis.
Then they started testing. First, 36 new pathways were awarded to a team of experts, artificial intelligence, and researchers.
“Each one of us had to sort these footprints into the categories of footprints left by meat-eaters and plant-eaters,” says Romelio.
“In this AI was the clear winner with 90% correctly identified. One of my colleagues and I came next with about 75%.”
Then they went for the crown jewel – Dinosaur Trails National Monument. When the AI analyzed this, it came back with a very strong result that they are plant eating ornithopod tracks. It’s not entirely certain, though, the data suggests that there’s a one in five million chance it could have been a theropod instead.
It’s still early days for AI to be used in this way. In the future. The researchers hope to fund a FrogID style app Anyone can use it to analyze dinosaur tracks.
“We hope to develop an app where anyone can take a picture on their smartphone and use the app and it will tell you what kind of dinosaur track it is,” Romelio says.
“It will also be useful in drone work surveying dinosaur tracks, collecting and analyzing image data and remotely locating fossil footprints.” The paper was published in Front of the Royal Society.