Artificial intelligence has recently ignited fresh discussions surrounding one of the most enduring debates in paleontology. At the heart of this controversy lies a remarkable collection of fossilized footprints that strikingly resemble modern bird tracks, yet date back millions of years before the earliest known birds ever existed.
A research team from Europe has developed an innovative AI system that sheds new light on how we analyze these ancient tracks. Unlike traditional methods that depend on human classification, this software directly identifies patterns based on raw shape data, which could potentially upend our established understanding of how avian species evolved over time.
Initial findings from this groundbreaking model have revealed surprising similarities between tracks from the Triassic and early Jurassic periods—timeframes that predate Archaeopteryx, historically recognized as the first true bird, by a staggering margin.
In a study published in January 2026 in the Proceedings of the National Academy of Sciences, scientists from institutions including the University of Tübingen, the University of Manchester, and the Museum für Naturkunde in Berlin detailed their findings. They used an unsupervised machine learning approach to analyze 2,000 tridactyl (three-toed) dinosaur footprints, categorizing tracks according to eight morphological factors such as toe orientation, heel size, toe spacing, and how weight was distributed on the ground.
Importantly, the tracks analyzed were not pre-labeled; the model categorized them purely based on geometric features, creating what the researchers refer to as an "eight-dimensional morphospace." After the AI organized the data, human experts compared its classifications with traditional taxonomic labels and found an impressive alignment, with matches occurring 80 to 93 percent of the time for well-preserved specimens.
This tool also powers a public application named DinoTracker, which invites users to upload images or drawings of fossil tracks. The app evaluates these submissions by comparing them against a database of previously classified samples, displaying where each entry falls within the AI’s morphospace. This innovation aims to standardize a process that has historically relied heavily on subjective assessments by experts.
Dr. Gregor Hartmann, the lead author of the study, explained in an interview with The Guardian that this unsupervised technique helps overcome a significant issue in the field—namely, the reliance on potentially erroneous previous labels. He pointed out, "Most likely, some of these labels are wrong."
To ensure the AI could perform effectively under real-world conditions, the research team generated over 10,000 synthetic variations of the original track samples. These simulations accounted for common fossil deformations like partial preservation, erosion, and compression, enabling the AI to discern meaningful differences even amidst noise and damage.
Among the most astonishing findings were a group of tracks dating from the Late Triassic and Early Jurassic periods. These footprints displayed narrow spacing between toes, pronounced central symmetry, and a structure that closely resembles those of contemporary birds. It’s particularly intriguing because these tracks predate the earliest known bird fossils by approximately 60 million years.
Dr. Stephen Brusatte, a co-author of the study from the University of Edinburgh, remarked to The Guardian that these discoveries align with suspicions held by some paleontologists for quite some time. He stated, "If these tracks were made by birds, that would mean that birds have a much older, much deeper ancestry than we used to think." However, he also acknowledged the possibility that the tracks might have been made by non-avian theropods with convergent evolution leading to bird-like feet, as the AI does not assign specific species but merely analyzes the geometry of the footprints without any preconceived taxonomic biases.
Despite these exciting revelations, skepticism remains among some researchers. Dr. Jens Lallensack from Humboldt University, who did not participate in the study, expressed doubts in The Guardian, suggesting that certain similarities might arise from the interaction of dinosaur feet with soft or uneven surfaces. He cautioned that mere morphological resemblance is not enough to confirm evolutionary connections, especially when no corresponding body fossils exist.
The introduction of DinoTracker represents a significant advancement in the study of trace fossils, paving the way for a scalable approach to ichnological analysis worldwide. The app allows individuals—from amateur fossil enthusiasts to professional field researchers—to contribute new findings from excavation sites, thereby enriching the central database with a wide array of data.
Given the global shortage of ichnologists in many areas, this tool aims to bolster efforts in fossil documentation, particularly in regions that lack resources. Once a track is uploaded, it is automatically analyzed and situated within the AI’s morphospace, allowing users to compare it with known samples and identify structural similarities.
According to a report from Phys.org, the AI has unearthed clusters of footprints that were previously overlooked by conventional classification methods, including forms that could suggest instances of convergent evolution between early dinosaurs and birds.
Currently, the focus of this research is on tridactyl dinosaur tracks, but there are plans to broaden the system's capabilities to include other types of trace fossils, such as plant impressions and movement patterns of invertebrates. The ultimate vision is to create an automated, geometry-based framework for classifying fossils across various domains within paleontology.
With public access to this innovative tool now available as a free mobile application, the reach of this technology is expanding. Its implications extend beyond academic circles, inviting broader participation in a field that has often been dominated by specialized interpretations.
