How AI is Deciphering the Carbonized Herculaneum Scrolls

Deep beneath the modern Italian town of Ercolano lies the Villa of the Papyri, a sprawling ancient estate believed to have been owned by the father-in-law of Julius Caesar. When Mount Vesuvius erupted in 79 AD, a superheated pyroclastic flow buried the villa in twenty meters of volcanic mud and ash. Inside the estate was a vast library of papyrus scrolls. The intense heat instantly carbonized the manuscripts, flash-frying them into brittle, blackened cylinders. For nearly two millennia, the only intact library to survive from classical antiquity sat frozen in the dark, preserved but entirely unreadable.[1][6][8]

Workmen digging a well in 1752 accidentally struck the villa's marble pavement, leading to the discovery of more than 1,800 carbonized scrolls. Initially mistaken for lumps of coal or charred logs, some were discarded or burned as torches before their true nature was realized. But the discovery quickly turned into a preservation nightmare. The scrolls were so fragile that early attempts to physically unroll them caused the papyrus to crumble into dust, permanently destroying the ancient texts they held.[6][8]

For centuries, the surviving unopened scrolls were considered a lost cause. Modern imaging techniques offered little hope, thwarted by a cruel chemical coincidence: ancient Roman scribes used a carbon-based ink. Because the volcanic eruption had turned the papyrus itself into carbon, the ink and the paper possessed the exact same density. When researchers subjected the scrolls to standard X-rays, the ink was completely invisible against the charred pages. The words were there, but human eyes and traditional cameras could not distinguish them from the void.[2][8]

The path to reading the unreadable began with a technique called "virtual unwrapping," pioneered by Dr. Brent Seales, a computer scientist at the University of Kentucky. Seales realized that if a scroll could be scanned at an incredibly high resolution, software might be able to digitally flatten the layers without anyone ever touching the physical artifact. To achieve this, researchers took the Herculaneum scrolls to particle accelerators, such as the Diamond Light Source in the United Kingdom, bombarding them with ultra-bright X-rays to create 3D micro-CT scans with a resolution of just a few micrometers—a fraction of the width of a human hair.[1][3][5]

The resulting 3D scans contained terabytes of data, but they were a chaotic mess of crushed, warped, and fused layers of papyrus. To make sense of the volume, researchers had to perform "segmentation"—painstakingly tracing the geometry of a single sheet of papyrus as it spiraled through the carbonized block. Once a continuous surface was mapped, the software could digitally extract that layer and flatten it into a 2D image. Yet, even after this monumental geometric reconstruction, the flattened pages still appeared blank. The carbon ink remained hidden.[2][8]

The final key to unlocking the library was artificial intelligence. In March 2023, Seales teamed up with tech entrepreneurs Nat Friedman and Daniel Gross to launch the Vesuvius Challenge, a global crowdsourcing competition offering over $1 million in prizes to anyone who could detect ink in the 3D scans. The breakthrough came when a contestant named Casey Handmer noticed a microscopic "crackle" pattern in the CT scans—a subtle textural difference where the ink sat on top of the papyrus fibers.[1][2]

Building on that observation, Luke Farritor, a 21-year-old college student, spent his evenings training a machine learning model to recognize the crackle. By feeding the neural network examples of the subtle texture, the AI learned to identify the invisible ink. In late 2023, Farritor's model successfully revealed the first complete word from a sealed Herculaneum scroll: πορφύρα (porphyras), the ancient Greek word for "purple." It was the first time in 2,000 years that a word had been read from the unopened library.[1][2][8]

The floodgates opened. In early 2024, a team comprising Farritor, Youssef Nader, and Julian Schilliger won the Vesuvius Challenge's $700,000 Grand Prize. By combining advanced segmentation software with refined AI ink-detection models, they recovered 15 columns of text containing more than 2,000 characters. The deciphered passages belonged to a previously unknown philosophical work, likely by the Epicurean philosopher Philodemus, discussing the nature of pleasure, music, and food.[1][3]

The technology is now yielding profound historical discoveries. In April 2024, Graziano Ranocchia, a papyrologist at the University of Pisa, announced that newly deciphered passages from a Herculaneum scroll had solved a long-standing archaeological mystery: the exact burial location of Plato. The AI-enhanced text revealed that the legendary Greek philosopher was interred in a private garden near a sacred shrine to the Muses within his Academy in Athens. The scroll also contained previously unknown details about Plato's life, including evidence that he may have been sold into slavery earlier than historians previously thought.[4][7]

Institutions worldwide are now applying the pipeline to their own collections. In 2025, the Bodleian Libraries at the University of Oxford successfully generated the first internal images of scroll PHerc. 172, which had been donated to the university in the 19th century. Using the AI segmentation and ink-detection models, researchers revealed multiple columns of text, identifying the Greek word διατροπή (diatropē), meaning "disgust." The Bodleian scroll proved to have denser ink than others, offering some of the clearest text recovered to date.[3][5]

As of 2026, the Vesuvius Challenge has entered a new phase focused on massive scale. The manual segmentation of the crushed papyrus layers remains a severe bottleneck, taking hundreds of hours per scroll. To solve this, researchers are developing self-supervised 3D foundation models, adapting architectures like Meta's DINO to work directly on volumetric images. These generalist models are learning to track the complex geometry of the papyrus fibers automatically, aiming to eliminate the need for human tracing.[1][8]

The stakes for classical history are difficult to overstate. The vast majority of literature from antiquity has been lost to decay; the surviving works of authors like Sophocles and Aristotle represent only a fraction of their output. With hundreds of Herculaneum scrolls still waiting to be scanned, and the possibility that a lower level of the Villa of the Papyri remains unexcavated, the AI-driven virtual unwrapping pipeline stands to dramatically expand the corpus of human knowledge. For the first time, technology is not just preserving the past—it is actively resurrecting it.[4][6][8]