Researchers Asked AI to Decipher Ancient Texts Lost for 2,000 Years

 

Researchers worldwide are using AI to decipher ancient texts, opening up libraries that have been lost for thousands of years. The technology is advancing so quickly that papyrologists say their field will never be the same.

What Happens When We Ask AI to Read the Unreadable

When researchers ask AI to examine ancient texts, they're getting results that seemed impossible just years ago.

The breakthrough came in October 2023, when papyrologist Federica Nicolardi received an email with an image that changed everything.

The image showed Greek text from a scroll burnt in Mount Vesuvius's eruption in 79 CE. After 2,000 years of silence. "It was incredible," Nicolardi said. "I thought, 'So this is really happening.'" She knew papyrology would never be the same. She was reading lines from a text that had been completely inaccessible since Roman times.

The Herculaneum scrolls were discovered in a luxury Roman villa in the 1700s.  For centuries, attempts to open them destroyed many. The rest couldn't be touched without turning them to dust.

Enter the Vesuvius Challenge in March 2023. Computer scientist Brent Seales from the University of Kentucky teamed up with Silicon Valley entrepreneur Nat Friedman. They offered $700,000 in prizes for anyone who could read the scrolls without opening them.

More than 1,000 teams participated in the challenge. The Discord channel was full of people talking about their progress every day. Luke Farritor, a college student, heard about it on a podcast while driving to his SpaceX internship. "You're going to potentially discover a new library from the ancient world, and that's a big deal," he said.

The problem seemed impossible. The scrolls couldn't be unwound. The carbon-based ink was the same density as papyrus, so it was invisible on CT scans. But maybe AI could spot small differences in the texture between the bare papyrus and the areas with ink.

By the way, if you're curious about how to ask AI about ancient texts and languages, try Overchat AI. It offers access to different AI models for you to try, including Qwen 3, a powerful model for statistical analysis and coding. It can also help you with translation, historical context, and linguistic analysis.

How the Technology Works

Reading ancient texts with AI involves several steps. First, researchers take detailed pictures of the inside of the scrolls using a special machine. These scans capture hundreds of paper-thin layers wound together.

Then comes virtual unwrapping. Algorithms map the surfaces visible in each cross-section frame. They make these surfaces into flat images using a computer. Then, the winning Vesuvius Challenge team used a TimeSformer model. This version of the transformer model usually analyzes videos. Here, it separated the depth of the papyrus from how it looked on the surface.

While we only see charred black material, the neural network detects small differences that reveal letters, which allowed the researchers to virtually unwrap the scroll.

In China, researchers use AI to read Oracle Bone Script

The Vesuvius Challenge is just one example of AI helping to study ancient texts. Neural networks are being used all over the world to analyze texts.

In China, researchers use AI to read Oracle Bone Script. These are divination texts written on cow bones and turtle shells. The AI fills in eroded lettering and puts together broken fragments. At the same time, researchers from Greece created a model called Ithaca. It can restore missing text from ancient inscriptions 62% of the time. Humans alone achieve 25%. However, humans working with Ithaca can achieve 72% accuracy. The AI can also date texts within 30 years and identify where they are from.

Meanwhile, south Korean teams face different challenges. They're translating one of the world's largest historical records. These records are from the daily lives of 27 Korean kings from the 1300s to the early 1900s. The texts are written in Hanja, an ancient system different from modern Korean or Chinese. Human translators would need decades to finish manually. AI is doing it in a matter of weeks right now.

This is just the beginning.

There are hundreds of unopened scrolls in Naples, Paris, London, and Oxford. "That's going to be more text for papyrologists that's new from the ancient world than they've seen in a hundred years," says Seales.

The technology opens what Seales calls "the invisible library." Texts hidden inside medieval book bindings. Ancient Egyptian writings found on mummies. Papyrus documents from Petra were burned in the seventh century. We held these objects but couldn't read them.

The Future of History

By next year, researchers expect to read 90% of four complete scrolls. Within five years, entire libraries thought lost forever will be accessible again. AI isn't just reading texts. It's about finding patterns in huge archives that no one could find on their own. This changes the questions historians can ask. It's not just about understanding what a text says, but also looking for patterns that appear in many different texts. AI can track how languages evolved, how ideas spread, and how civilizations connected.

The invisible library is becoming visible. After two thousand years, it's time to speak up. And we're just getting started.

 

---