AI Cracks 2,000-Year-Old Herculaneum Scrolls Buried by Mount Vesuvius

In October, three computer science students sitting at their screens discovered the Greek word for purple buried inside a scroll that had been sealed in ash for two thousand years. The word emerged from layers of carbonized papyrus that no human eye had ever seen, revealed by machine-learning software trained to detect the faintest trace of ancient ink. It was a small victory—two characters pulled from the darkness—but it announced something larger: the Herculaneum scrolls, thought to be permanently closed, were finally opening.

These scrolls had been waiting since 79 A.D., when Mount Vesuvius erupted and buried a villa near the Bay of Naples under sixty-five feet of superheated ash. The villa belonged to Julius Caesar's father-in-law and housed a library of papyrus texts. The heat that should have destroyed them instead carbonized them, turning the scrolls into brittle, blackened objects that looked like charcoal and could not be touched without crumbling. For nearly two thousand years, they remained sealed. A farmer digging a well in the 1700s found more than six hundred of them. Scholars unwrapped a few using mechanical devices, discovering they contained primarily Epicurean philosophy. The rest sat in European libraries, unread and seemingly unreadable.

Brent Seales, an imaging specialist from Western New York, spent two decades trying to change that. He began in the mid-1990s working on digitizing medieval manuscripts, then moved to the idea of virtually unwrapping damaged texts using computed tomography—the same scanning technology used in hospitals. By 2004, a classics scholar named Richard Janko suggested the Herculaneum scrolls as the ultimate test. Seales spent four years building relationships with conservators and museum directors before gaining permission to scan them in 2009 at the Institut de France. The scrolls were smaller than he expected, shrunken by carbonization, and looked impossibly fragile. But once scanned, they existed as digital data—three-dimensional maps of their internal structure.

The challenge was reading them. The scrolls had been written with soot from oil lamps, which is chemically almost pure carbon. The papyrus itself is also carbon. When X-rays passed through, they saw gray on gray—no contrast, no visible ink. For years, Seales' team tested machine-learning models on fragments and replicas, training algorithms to detect the faintest difference between ink and papyrus. By 2023, they had read two characters from inside a sealed scroll. It was proof the method could work.

In March 2023, Seales, former GitHub CEO Nat Friedman, and entrepreneur Daniel Gross launched the Vesuvius Challenge—a global competition offering $700,000 in prize money to anyone who could read the scrolls using the CT scans and open-source software the team had developed. They released the data to the world. More than a thousand people worked on the problem over the next nine months. In December, three students—Youssef Nader, Luke Farritor, and Julian Schilliger—submitted fifteen columns of readable text. Preliminary analysis suggested the writing discussed whether scarcity or abundance of goods affects how much pleasure humans find in them. It was Epicurean philosophy, exactly as scholars had predicted. It was also proof that the entire library could be read.

Michael McOsker, a researcher who has studied the scrolls, estimates the collection could yield approximately two hundred new books. This is the only surviving library from antiquity. We possess less than one percent of all literature written in the ancient world. Every text recovered matters. Seales imagined reaching back two thousand years to a conversation stripped of modern boundaries—about love, war, music, rhetoric, poetry, and food. The scrolls talk about what it means to be human. "We have to read it," he said. "We can't forget it." The work continues. The goal for 2024 is to read ninety percent of the first scroll. There are still seventeen hundred more waiting in the dark.