Palaeontologists believe they have shed new light on a debate over what kind of dinosaur may have created the ancient tracks at the Lark Quarry Conservation Park in Australia – by analyzing the footprints using AI.

The imprints on the ground have been interpreted by experts as fossilized footprints, left when a group of 150 or so dinosaurs scattered to escape a predator some 93 million years ago – during the mid-Cretaceous Period. They have been preserved, protected, and turned into a tourist attraction.

"Large dinosaur footprints were first discovered back in the 1970s at a track site called the Dinosaur Stampede National Monument, and for many years they were believed to have been left by a predatory dinosaur, like [1]Australovenator , with legs nearly two meters long," [2]explained Anthony Romilio, a research associate at the University of Queensland, Australia, who has been studying the tracks.

[3]

The patterns consist of a series of small impressions surrounded by a set of larger ones that look like they were produced by the footsteps of dinosaurs. Palaeontologists, however, disagree on what kind of creature may have produced the bigger track. The stampede idea envisions a larger, carnivorous theropod dinosaur preying upon the smaller ornithischian herbivorous species.

[4]

[5]

But that may be incorrect, according to a [6]paper published in the Journal of the Royal Society Interface .

Interpreting dinosaur tracks is difficult. The outlines of imprints are often fuzzy, and evidence is subjective. A team of researchers at the University of Queensland and the UK's Liverpool John Moores University, along with colleagues in Germany, turned to machine learning for assistance.

[7]

"Some very different dinosaurs – meat-eating theropods and plant-eating ornithopods – had feet with three toes," Romilio told The Register . "Distinguishing which dinosaur made the three-toed dinosaur tracks can be a source of frustration. Typical approaches use very general identifiers (theropods with long, narrow toes and ornithopods with short, dumpy toes) and measurements of footprint landmarks (footprint is longer than wide for theropods, wider than long for ornithopod). But these still have problems distinguishing tracks between these extreme shapes."

[8]Intel says it can sort the living human beings from the deepfakes in real time

[9]Nvidia, Lockheed team up to build digital twin of the Earth for climate researchers

[10]A next-gen AI protein folder that could help science? Meta's good for something

[11]Microsoft said to be in talks to invest more in OpenAI

The researchers trained a convolutional neural network on 1,500 dinosaur footprints, which were split into theropods or ornithopods – a type of dinosaur closely related to ornithischians. The model only looked at the overall shape of the footprint images, and did not consider other types of information – like their sizes or surface features. It was then tested on a set of 36 tracks.

The neural network spat out a score between 0 and 1 – a measurement of how confident it was that an image should be classified as a theropod or an ornithopod. The researchers admitted in their paper that the small sample sizes of both the training and testing datasets limit the performance of the system.

Still, they claimed the model outperformed human experts during testing. "Eighty-six per cent of tracks were correctly classified while 14 [percent] were incorrectly classified. If all neural network decisions above 0.4 and below 0.6 are treated as ambiguous (22 [percent] of tracks), the neural network classified 67 [percent] of tracks correctly and 11 [percent] incorrectly. The human experts, on average, classified 57 [percent] correctly, 20 [percent] incorrectly and 24 [percent] as ambiguous," they wrote.

When they applied the model to the images from the Dinosaur Stampede National Monument, the software classified the footprints as belonging to an ornithopod. "All but one of these tracks was confidently classified as left by an ornithopod dinosaur – our prehistoric 'predator'," said Jens Lallensack, lead author from Liverpool John Moores University.

[12]

In other words, the larger dinosaur was probably not predatory. The computer vision software isn't perfect and has its own subjectivity based on its training. The authors warned "it has to be the job of the ichnologist" – a scientist who studies trace fossils like footprints, rather than bones – "to combine the neural network evaluation of the shape with all relevant context information to arrive at a meaningful interpretation of the track."

Of course the only way to answer the question definitively would be to restore the various dinosaur species using DNA trapped in amber and compare the tracks against contemporary exemplars. Where's John Hammond when you need him? ®

Get our [13]Tech Resources



[1] https://www.australianageofdinosaurs.com/page/48/australian-age-of-dinosaurs-australovenator-wintonensis

[2] https://science.uq.edu.au/article/2022/11/prehistoric-predator-artificial-intelligence-says-no

[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Y3TCzdr1O6B47AVSynhU4wAAAIo&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y3TCzdr1O6B47AVSynhU4wAAAIo&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y3TCzdr1O6B47AVSynhU4wAAAIo&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[6] https://royalsocietypublishing.org/doi/10.1098/rsif.2022.0588

[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y3TCzdr1O6B47AVSynhU4wAAAIo&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[8] https://www.theregister.com/2022/11/15/intel_fakecatcher/

[9] https://www.theregister.com/2022/11/14/nvidia_lockheed_noaa_digital_twin/

[10] https://www.theregister.com/2022/11/02/meta_ai_protein/

[11] https://www.theregister.com/2022/10/21/microsoft_openai_investment/

[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/science&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y3TCzdr1O6B47AVSynhU4wAAAIo&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[13] https://whitepapers.theregister.com/