AI Unlocks Secrets of 2,000-Year-Old Scroll Charred in Mount Vesuvius Eruption

A group of student researchers has successfully deciphered ancient Greek script on a scroll that was charred during the catastrophic eruption of Mount Vesuvius in 79 CE. This breakthrough could not only unveil a text that was once unreadable but also herald new uses of artificial intelligence in archaeology.

Originating from the opulent Roman villa of Herculaneum, the scroll is part of the Herculaneum papyri collection, which includes over 1,800 papyri turned to carbonized remains, discovered in the 18th century. Previously deciphered scrolls from this collection have provided insights into the philosophy of Epicurus, who lived from 341 to 270 BCE. This collection represents the only known library from the ancient world that has survived, although the majority of the scrolls are too delicate to be opened.

The decipherment came about through the Vesuvius Challenge competition, where the winning team employed machine learning techniques on scanned images of the tightly rolled papyrus. They revealed an unknown philosophical manuscript discussing the nature of senses and pleasure, touching on topics like music, the flavor of capers, and the hue of purple, and even mentioning Xenophantus, a flautist referred to by ancient writers Seneca and Plutarch.

The student team managed to decode hundreds of words over more than 15 columns of text, which represents roughly 5 percent of the entire scroll. For their groundbreaking work, the trio from Egypt, Switzerland, and the USA was awarded a $700,000 grand prize.

Brent Seales, a computer scientist at the University of Kentucky, Lexington, and co-founder of the prize, expressed that the contest has dispelled any skepticism about the feasibility of such technological applications. Seales has spent two decades attempting to read these hidden texts by developing software that uses 3-D computed tomography (CT) to map the surfaces of the rolled papyri. However, the challenge of distinguishing the carbon-based ink from the papyrus in CT scans has been a significant obstacle.

The Vesuvius Challenge, initiated by tech entrepreneur Nat Friedman with a $125,000 donation and additional funds raised online, released Seales’ software and high-resolution scans for competitors to use. The challenge tasked participants with creating machine-learning models capable of unwrapping the scrolls and identifying the text, offering a grand prize for deciphering four passages of at least 140 characters each by the end of 2023. The competition encouraged collaborative progress by releasing the winning machine-learning code in stages and providing smaller prizes for ongoing advancements.

In the midst of the competition, Casey Handmer, a US entrepreneur and former physicist, observed a texture resembling cracked mud in the scans that spelled out Greek letters. Luke Farritor, a computer science undergraduate at the University of Nebraska-Lincoln, utilized this texture to train a machine-learning algorithm that successfully identified the word for “purple,” earning him a prize. Youssef Nader, a PhD student based in Berlin, subsequently produced clearer images of the text.

Out of eighteen entries, a jury vetted the code and presented twelve to a panel of papyrologists who evaluated the clarity and transcribed the content. Only the team comprising Farritor, Nader, and Julian Schilliger, a Swiss robotics student at the Swiss Federal Institute of Technology Zurich, fulfilled the criteria to win the prize.

Federica Nicolardi, a judge and papyrologist at the University of Naples Federico II, along with her colleagues, is currently analyzing the text and was astonished by the clarity of the images provided by the team.