Dinosaur tracks have always been a bit of a riddle. Now an artificial intelligence app called DinoTracker is turning those ancient footprints into data that scientists can finally read in detail, and the results might push the origin of birds tens of millions of years further back in time than anyone expected.
Developed by researchers at Helmholtz-Zentrum Berlin and University of Edinburgh, the system lets users upload a photo or simple outline of a fossil footprint. In seconds, the app suggests what type of dinosaur likely made it, with accuracy that rivals expert paleontologists in most tests.
So what is the app actually looking at when it studies a three‑toed print in stone?
How DinoTracker reads stone footprints
Behind the simple interface is an unsupervised neural network that was trained on 1,974 dinosaur and bird footprints, spanning more than 200 million years of evolution. Instead of being told in advance which tracks belong to which group, the network was left to sort the shapes on its own and find the patterns that best separate them.
The model learned eight key ways that footprints differ. Among them are how widely the toes spread, how the toes attach to the rest of the foot, where the heel sits, how much of the foot actually touched the ground, and how the weight was distributed from left to right.
In trials, its classifications matched human expert judgments roughly 80 to 93% of the time, depending on the kind of track.
Physicist Gregor Hartmann, who led the study, describes the method as an objective way to recognize subtle variation in footprints that people tend to miss. He notes that it reduces the risk that long‑standing assumptions will simply be baked into new digital tools.
Oldest “birds” hiding in plain sight
Once the network had mapped this footprint “landscape,” the team added labels based on previous expert work, including tracks from meat‑eating theropods, plant‑eating ornithopods, large quadrupeds, and both fossil and modern birds. Then they fed in some of the most controversial prints in dinosaur research.
Several tiny three‑toed tracks from the Late Triassic and Early Jurassic, including a set from what is now South Africa, had long looked uncannily bird‑like. DinoTracker placed most of them squarely among bird footprints, not with non‑avian theropods.
The tracks are around 210 million years old, roughly 60 million years older than the earliest known bird skeletons.
Does that mean real birds were already walking around that early? Researchers are cautious. The prints might represent true early birds whose bones have not yet been found, but they could also come from small dinosaurs that just happened to have very bird‑like feet or that made bird‑like impressions in soft, wet sediment.
As one of the study’s senior authors, paleontologist Steve Brusatte, puts it, the resemblance is so strong that scientists now have to take this possibility seriously and look for an explanation that fits both bones and tracks.
The app also revisited puzzling Middle Jurassic footprints from the shores of a long‑vanished lagoon on the Isle of Skye.
Many of those prints cluster with early relatives of duck‑billed dinosaurs, suggesting some of the oldest known members of that group were already roaming what is now Scotland, even though a few tracks still look closer to classic theropods.
From ancient mud to modern climate questions
At first glance, all this may sound like a very specialized debate. Yet dinosaur footprints are not just curiosities pressed into slabs in a museum hallway.
They form in real environments such as riverbanks, floodplains, and tidal flats, and they often preserve how animals moved, which paths they followed, and which habitats they preferred. That makes them powerful clues to the structure of ancient ecosystems and to how life responded when climate and sea level shifted in the deep past.
By sorting thousands of tracks in a more consistent way, AI tools like DinoTracker can help scientists better map when different dinosaur groups appeared, where they lived, and how communities changed across mass extinctions.
Those long‑term patterns matter when we ask how today’s species might react to rapid warming or habitat loss. The rock record becomes a kind of stress test for planetary change.
A new tool, not a crystal ball
The team is clear that DinoTracker does not replace human fieldwork. The shape of a footprint depends on what the animal was doing, how wet the ground was and how much the print has been eroded or distorted during fossilization.
The app focuses on shape alone, so researchers still need to verify the age and geology of each site before any bold claim about bird origins or dinosaur behavior can stand.
The good news is that the software is free and meant for both professionals and curious members of the public.
A paleontologist working in a remote desert, a student on a class trip, or a hiker who stumbles across a suspicious three‑toed impression can all use the same tool to get a first, standardized comparison with known tracks.
That first pass may be wrong in some cases, but it gives experts a common starting point instead of a stack of competing gut feelings.
In the end, DinoTracker acts less like an oracle and more like an extra colleague at the table, one that sees patterns in stone that human eyes easily overlook.
The study was published in Proceedings of the National Academy of Sciences.
