How AI Helped Decipher an Ancient Roman Scroll

By Gina Wynn

In 79 A.D., a massive volcanic eruption covered the Roman city of Herculaneum in over 65 feet of ash, hiding nearly 2,000 valuable scrolls for centuries. When they were discovered in the late 1700s in a buried luxury villa, the charred documents were thought to be unreadable. Thanks to technology and a collaborative scientific contest, however, a 21-year-old student has been able to identify the first whole word.

Modern Interprets Ancient

To speed up the process of finding a way to read the ancient texts, Brent Seales and his team at the University of Kentucky in Lexington set up the Vesuvius Challenge—a competition designed to tap into the creativity and expertise of researchers worldwide. Seales and his colleagues posted X-ray computed tomography (CT) scans of the scrolls and their codes on the Challenge website scrollprize.org and offered prize money to teams or individuals who met certain criteria for decoding letters and passages.

So far, Challenge participants are contributing to a mood of “unbridled optimism” for Seales and his colleagues, according to the Nature article “AI reads text from ancient Herculaneum scroll for the first time.” A computer science major at the University of Nebraska-Lincoln, Luke Farritor, has successfully deciphered the first whole word in the scrolls—purple, written in Greek. He was able to view the letters through a machine-learning algorithm he created, an application of advanced artificial intelligence (AI).

Great Minds Think Alike

Shortly after Farritor, Free University Berlin graduate student Youssef Nader independently discovered the same word. Both received part of the $50,000 “First Letters” prize. The first person or team to decipher four continuous passages of at least 140 characters will win $700,000.

Competitors use a technique called “virtual unwrapping,” developed by Seales, to help decode the scrolls, according to the CNN article “AI reads text from famously inscrutable ancient scroll for the first time.” This process virtually flattens the pages using the curved layers in the CT scans as guides. Advanced AI trained to look for ink on the pages is then applied.

Crackle and Pop

Because the ink on the scrolls is mostly made of charcoal and water—like the carbonized, plant-based papyrus to which it was applied—it doesn’t show up on the CT scans. So instead of analyzing the ink properties, Farritor decided to focus on the minute differences in the papyrus surface texture under the ink, Nature reported. Another Vesuvius Challenge competitor first drew attention to the differing texture or “crackle.” Farritor’s algorithm highlighted the crackle that ended up revealing the hidden letters.

Peeking into the Past

Farritor’s breakthrough is scientifically and culturally important because similar algorithms might be able to help unlock knowledge in other unreadable ancient texts, including the remaining 600 scrolls recovered from the villa library. Unfortunately, attempts to physically unroll a large portion of the excavated scrolls left them in crumbled pieces and fragments.

Farritor’s and Nader’s ability to decipher parts of the scrolls while leaving them intact not only sheds light on ancient history, but it also shows how technology and teamwork can uncover secrets lost for centuries. Seales and his team’s dedication to understanding these forgotten manuscripts is opening a new door to the past, making history come alive in new ways.