How AI is Finally Reading the 'Unreadable' Herculaneum Scrolls

In the autumn of 79 AD, Mount Vesuvius erupted, burying the Roman resort town of Herculaneum under a devastating pyroclastic surge. Unlike nearby Pompeii, which was slowly buried under falling ash, Herculaneum was hit by a 500-degree Celsius wave of volcanic gas and mud.[1]

This intense, oxygen-deprived heat instantly carbonized organic material, preserving wooden beams, food, and most importantly, an entire library. Discovered by well-diggers in the 1750s, the Villa of the Papyri yielded hundreds of tightly rolled papyrus scrolls.[1][3]

For more than two and a half centuries, these scrolls represented one of archaeology's most tantalizing and frustrating mysteries. They physically survived, but they resembled brittle lumps of charcoal.[2]

Early attempts to physically unroll them were disastrous. When researchers tried to pry the layers apart, the carbonized papyrus would shatter and crumble into dust, destroying the very history they sought to uncover.[2][4]

Today, a global coalition of computer scientists, physicists, and classicists is finally reading the unreadable. Through a combination of high-resolution particle accelerators and advanced machine learning, the "Vesuvius Challenge" is successfully extracting lost ancient literature without ever opening the physical artifacts.[1][7]

The breakthrough relies on a technique called "virtual unwrapping," pioneered by Dr. Brent Seales, a computer scientist at the University of Kentucky. The process begins by taking the charred scrolls to facilities like the Diamond Light Source in the UK, where powerful X-ray computed tomography (micro-CT) scans create massive, three-dimensional digital models of the scrolls' internal structures.[4][6]

But scanning the scrolls was only half the battle. The scribes of Herculaneum used carbon-based ink. Because the ink and the carbonized papyrus have the exact same density, the text is completely invisible on a standard X-ray.[3]

To solve this, Seales and his team hypothesized that the ink, even if invisible, might leave microscopic physical signatures on the papyrus—subtle changes in texture or thickness where the liquid once soaked into the plant fibers.[3][6]

In March 2023, Seales partnered with technology entrepreneurs Nat Friedman and Daniel Gross to launch the Vesuvius Challenge, offering a $1 million prize pool to anyone who could build an artificial intelligence model capable of detecting these microscopic ink signatures.[1][6]

The organizers released terabytes of 3D X-ray data and open-source virtual unwrapping software, inviting citizen scientists and machine learning experts worldwide to tackle the bottleneck. The first major milestone fell in October 2023, when Luke Farritor, a 21-year-old computer science student at the University of Nebraska-Lincoln, trained a neural network that successfully detected the ancient Greek word "porphyras," meaning purple.[1][3][5][6]

The pace of discovery rapidly accelerated. In February 2024, the $700,000 Grand Prize was awarded to a trio of student researchers—Farritor, Youssef Nader, and Julian Schilliger—who managed to decode 15 columns of text containing more than 2,000 characters.[1][6]

Papyrologists translating the Grand Prize submission revealed it to be a previously unknown philosophical treatise. The text, likely written by the Epicurean philosopher Philodemus, discusses the nature of pleasure, specifically examining how the scarcity or abundance of food and music affects human enjoyment.[5][6]

The momentum has continued into recent years. In early 2025, the Vesuvius Challenge team successfully generated the first internal images of a scroll held at Oxford's Bodleian Libraries, identifying the Greek word "diatrope," meaning disgust.[4]

Because AI models are prone to "hallucinating" data, the evidence-gathering process is rigorously transparent. The competition requires independent teams using entirely different machine learning architectures—such as TimeSformer and ResNet models—to process the same digital segments.[7]

When multiple, distinct neural networks independently highlight the exact same geometric patterns of Greek letters, papyrologists can be highly confident that the AI is detecting genuine physical ink rather than generating digital artifacts.[7]

The implications for classical history are staggering. Scholars estimate that more than 99 percent of ancient Greek and Roman literature has been lost to time. The Villa of the Papyri is the only intact library to survive from antiquity, and its unread scrolls could contain lost plays by Sophocles, missing dialogues of Aristotle, or early texts that reshape our understanding of the ancient world.[1][7]

Furthermore, archaeologists believe that the roughly 800 scrolls recovered so far may only represent a fraction of the villa's collection. There is widespread speculation that a larger, primary library remains buried in the unexcavated lower levels of the complex.[1][7]

As the Vesuvius Challenge moves toward its next goal of fully automating the segmentation and unwrapping pipeline, the focus is shifting from proving the concept to scaling it. If the software can be refined to read entire scrolls in days rather than months, the "invisible library" of Herculaneum will finally be open to the world.[1][6][7]