The AI Breakthrough That’s Changing How We Study Fossils — Faster and Smarter

The Tools My Paleontologist Wife Wished She Had During Her PhD

cientists from Queensland Museum and James Cook University have figured out a way to speed up the painstaking process of analyzing fossils using artificial intelligence (AI). Here’s the thing: fossils, especially fragile ones embedded in dense rock, have always presented a unique challenge for paleontologists. Removing them from the rock would certainly damage them, but you really want to know what lies underneath.

So, “traditionally,” and by that we mean the last two decades, examining these fossils without damaging them involved using CT scans. However, the real challenge lies in digitally separating the fossil from the rock in those scans. This process could take months of manual work—until now. As it turns out, with AI, what used to take months can now be done in days.

You did your PhD a decade too early, honey!

But, what exactly are we talking about here? A recent study published in Scientific Reports has some answers for us. The researchers, Dr. Espen Knutsen and Dr. Dmitry Konovalov, found a way to use AI to automate much of the fossil segmentation process.

In other words, AI helps identify where the fossil ends and the surrounding rock begins, significantly reducing the time needed to extract and study these ancient remains digitally.

Trust me. I spent days at end digitally editing figures, only for her computer to crash right when she was about to finish a four-hour task. 

But moving on, what did they do exactly?

Dr. Knutsen and Dr. Konovalov trained a Deep Learning model—a type of AI—to segment fossil CT scans. This process involved training the model on a small portion (just 2%) of a larger dataset consisting of CT slices, each slice essentially a high-resolution X-ray image. These scans allow paleontologists to see inside a fossil without breaking its encased rock.

The AI model learned how to spot the difference between rock and fossil material from just a few manually labeled slices, and from there, it took over the rest of the dataset on its own. With the help of advanced image processing algorithms, the model created a 3D representation of the fossil, which was then compared to manual segmentations to estimate its accuracy.

And what did they find?

The AI model delivered. It could create highly precise 3D models of the fossils in just a few days — something that would normally take months of careful, manual work. As Dr. Knutsen explained, “We achieved a highly precise 3D representation of a tiny Triassic reptile from Queensland that was around 240 million years old. This was completed in days rather than months.”

Kudos guys! Faster data collection also means faster discoveries!

This particular fossil came from Queensland, Australia, and dates back to the Early Triassic period, not long after the largest mass extinction in Earth’s history. The researchers tested their method on a small skull and limb bone.

Although the AI didn’t get every detail perfect — there were some minor inaccuracies, such as in the internal structure of teeth and some limb bones — the results were remarkably close to manual segmentations. In most cases, the AI’s results required only minor adjustments, drastically cutting down the human effort involved.

A little step for paleontologists, a bigger step for Science!

However, this new AI technique is more than just a time-saver. It represents a significant step toward making fossil data more accessible and available for research. Segmenting fossils from CT scans has traditionally been a major bottleneck in paleontological research. Indeed, I have seen many dissertations fall short of initial expectations because of this.

However, this approach allows researchers to process far more data in less time, opening up opportunities to study a wider range of fossils and ask bigger questions about deep time.

One of the most exciting aspects of this development is its potential scalability. Although the study focused on fossils from Queensland, there’s every reason to believe this method can be adapted to fossils from other parts of the world.

This AI-based technique could be used with further refinement on various fossil species, from tiny invertebrates to massive dinosaur bones. When I read this paper, multiple ideas popped up within an hour. I had a fun time brainstorming possibilities!

As Dr. Knutsen points out, this work “removes a huge bottleneck in modern paleontology, allowing researchers to rapidly generate more data from fossil collections and spend more time exploring our ancient past.” In other words, AI might soon allow us to unlock the secrets of Earth’s history faster than ever before.

However, the key takeaway here is that AI also has the potential to revolutionize how we study fossils. While fossil hunting will always involve a bit of luck and adventure, the time-consuming task of analyzing them could soon be streamlined thanks to technology.

Let me tell you, no paleontologist I know will complain about being able to spend more time in the field while machines do the analyses.

The marriage of paleontology and AI could make it easier for scientists to study fossilized remains and answer questions about ancient life on Earth, from the behavior of tiny creatures to the environmental changes that shaped their worlds. Just like my marriage with this amazing human, which has brought me the most amazing life I could imagine!

By making fossil analysis faster and more efficient, AI can free researchers to focus on their work’s more creative and interpretive aspects. It also opens the door for more collaboration and data-sharing between institutions, accelerating the pace of discoveries in paleontology.

In the coming years, AI may play an even bigger role in this field, transforming how we analyze fossils and interpret the history of life on Earth. And who knows? An AI model may process and understand the next big fossil discovery before it even hits the museum floor. I’ve seen this almost happen already…