Ghost in the Stone: AI Uncovers Prehistoric Clues That Could Rewrite Bird Evolution

The Digital Hunter: A New Way to See the Past

DinoTracker isn’t your average mobile tool. Developed by researchers at the Helmholtz-Zentrum Berlin and the University of Edinburgh, this open-source application—available on GitHub—utilizes neural networks trained on thousands of fossilized outlines. Instead of relying on potentially flawed human labels, the team used an “unsupervised” approach. This means the algorithm independently identifies geometric patterns and clusters footprints by similarity before comparing them to expert interpretations.

In other words, the DinoTracker app is designed to scale a task that has traditionally depended on rare expertise and subjective judgment.

How Does AI Dinosaur Track Identification Work?

Whether used by a seasoned paleontologist in the Gobi Desert or an amateur hiker in the Rockies, the process is seamless. Users upload a photo or sketch of a footprint, and the AI immediately dissects 8 critical features. These include toe spread, heel position, surface contact area, and weight distribution.

The algorithm then compares these metrics against a database of thousands of real-world tracks and millions of simulated variations. These simulations are crucial; they account for how a track might deform over millions of years due to geological pressure or weathering. Ultimately, the system suggests the most likely “author” of the print from the dinosaur kingdom—an approach many researchers describe simply as dinosaur footprint identification AI, but at far greater speed and consistency than traditional sorting.

Matching the Experts: A 90 percent Success Rate

In rigorous head-to-head tests, DinoTracker matched the classifications of human specialist teams about 90 percent of the time. It maintained this accuracy even when faced with controversial or degraded tracks that often baffle the human eye. This high performance stems from the AI’s ability to “see” through variations caused by the substrate—whether the dinosaur was walking through slippery mud or shifting sand.

When we find a dinosaur track, we’re trying to play Cinderella and find the foot that fits the slipper. But it’s not that simple. The shape of a track depends not just on the shape of the foot, but on the type of mud or sand it stepped in, and the way the foot was moving,

– explains Professor Steve Brusatte of the University of Edinburgh in a recent interview with The Guardian.

However, researchers stress that this 90 percent mark doesn’t mean the AI is infallible. Instead, it serves as a high-speed “second opinion” for questionable cases, allowing experts to process findings at a scale previously impossible. Put plainly: the AI identifies dinosaur tracks quickly, but final classification still belongs to the scientific process.

The Essential Human Element

Despite its brilliance, AI dinosaur track identification does not replace the formal scientific process. Computers analyze shapes, but they lack the ability to read the landscape. Human researchers must still verify the age of the rock layer and the type of sediment before confirming a find.

Furthermore, machine learning models can be sensitive to data gaps. If the training set lacks enough tracks from a specific group, the algorithm’s performance will inevitably dip.

Our method provides an objective way to recognize track variation and test hypotheses about their makers. It is an excellent tool for research, education, and field work,

– says Dr. Gregor Hartmann of the Helmholtz-Zentrum.

Ancient Prints, Modern Revolutions

Because DinoTracker is free, it democratizes paleontology. It allows hobbyists to instantly see if their backyard discovery is worth reporting to a museum or beginning a formal excavation. Scientifically, however, the real power lies in its ability to scan massive, scattered photo databases to find “hidden” patterns.

Are Birds Older Than We Thought?

The system has already delivered its first bombshell. DinoTracker identified tracks older than 200 million years that bear a striking resemblance to modern birds. These narrow, three-toed prints feature a distinct, separate rear “toe” and minimal heel contact—a signature consistent with light, bipedal, bird-like locomotion rather than the heavy gait of massive dinosaurs.

These tracks predate the oldest widely accepted bird body fossils by roughly 60 million years. That leaves scientists with two plausible interpretations: either early birds (or close avian relatives) appeared much earlier than the current body-fossil record suggests, or a specific dinosaur lineage evolved bird-like feet long before true birds ever took flight.

If birds left these tracks, it would mean that birds have much older, deeper roots than we ever suspected,

– comments Brusatte.

Regardless of the final verdict, AI dinosaur track identification has proven that the stones beneath our feet still have plenty of stories left to tell—if only we have the right eyes to see them.

Read this article in Polish: AI tropi pradawne zwierzęta. Odkryła coś ważnego