How AI is Decoding the Carbonized Scrolls of Herculaneum

For more than two centuries, one of the greatest literary treasures of the ancient world has sat in plain sight, completely unreadable. The Herculaneum papyri represent the only large-scale library from classical antiquity that has survived to the present day in its entirety. Yet, until recently, the hundreds of scrolls housed in museum vaults were little more than tantalizing mysteries—fragile lumps of carbonized ash that guarded the lost philosophies, histories, and poetry of the Greco-Roman world. Now, an unprecedented convergence of particle physics, computer vision, and crowdsourced machine learning is finally unlocking the invisible library. Without ever physically opening the delicate artifacts, researchers are peering deep inside their crushed layers to read words that have not been seen by human eyes in nearly two millennia.[7]

The story of the scrolls begins with a cataclysm. In 79 AD, the catastrophic eruption of Mount Vesuvius obliterated the Roman towns of Pompeii and Herculaneum. While Pompeii was famously buried under a steady rain of pumice and ash, Herculaneum was hit by a blistering pyroclastic flow—a superheated avalanche of gas and volcanic debris. This intense, oxygen-deprived heat instantly carbonized the organic materials in the town. Inside a sprawling luxury estate known today as the Villa of the Papyri, thought to have belonged to Julius Caesar's father-in-law, a vast collection of papyrus scrolls was flash-fried into compact, coal-like briquettes. Preserved beneath twenty feet of volcanic rock, the library slept undisturbed for centuries.[2][6]

The modern saga of the Herculaneum library began in 1752, when workmen digging a well for the Bourbon royal family accidentally struck the buried villa. As excavators tunneled through the hardened volcanic mud, they unearthed stunning bronze statues, intricate frescoes, and hundreds of blackened, cylindrical objects. Initially, many of these artifacts were thrown away, mistaken for lumps of charcoal or petrified tree branches. It was only when someone noticed faint traces of lettering on a broken fragment that the excavators realized they had stumbled upon a massive ancient library, containing upwards of 1,800 individual scrolls.[2][4]

Discovering the library was a triumph; attempting to read it was a disaster. For generations, scholars and antiquarians tried desperately to unroll the carbonized papyri. Early methods were brutal, involving slicing the scrolls in half with butcher knives or dousing them in harsh chemicals. Even the most delicate mechanical attempts, such as a specialized machine invented by a Vatican monk that used silk threads to slowly pull the papyrus apart, resulted in the scrolls crumbling into flakes of black dust. While some fragmented texts were recovered, the physical unrolling process invariably destroyed the very artifacts it sought to preserve.[2][4]

By the late 20th century, physical unrolling was entirely abandoned. The remaining collection—more than 300 completely intact, sealed scrolls—was placed in protective storage, mostly at the National Library of Naples and the Bodleian Libraries in Oxford. Scholars resigned themselves to the bitter reality that the technology required to read the scrolls simply did not exist. The invisible library became a symbol of lost knowledge, containing works by classical authors that had otherwise vanished from history, locked away in a format that was too fragile to touch.[4][6]

The paradigm shifted with the pioneering work of Dr. Brent Seales, a computer scientist at the University of Kentucky. Seales envisioned a process called "virtual unwrapping," which would use high-resolution medical imaging to see inside an object without opening it. In 2015, Seales and his team achieved a monumental proof-of-concept when they successfully read the En-Gedi scroll, a charred parchment from the Dead Sea region. By tracing the layers of the scroll in a 3D scan and digitally flattening them, they revealed the text of the Book of Leviticus. It was a watershed moment for digital archaeology, but Herculaneum presented a vastly more difficult challenge.[3][6]

The En-Gedi scroll was written with a metal-based ink, which showed up brightly on an X-ray against the carbonized parchment. The ancient Romans at Herculaneum, however, used a carbon-based ink made from soot and water. Because the papyrus itself had been turned into carbon by the volcano, the ink and the paper had the exact same radiodensity. On a standard X-ray or CT scan, the ink was completely invisible. The scrolls appeared as a dense, chaotic maze of crushed carbon, with no discernible lettering to guide the virtual unwrapping process.[1][3]

To find the invisible ink, researchers needed unprecedented resolution. They took the delicate scrolls to national synchrotron facilities—massive particle accelerators like the Diamond Light Source in the UK and the European Synchrotron Radiation Facility (ESRF) in France. These facilities produce X-ray beams billions of times brighter than the sun, capable of capturing 3D scans at a microscopic resolution of just a few micrometers. The resulting datasets were massive, providing a highly detailed topological map of the crushed papyrus layers, but the carbon ink still refused to show itself clearly to the naked eye.[1][5]

Realizing that the sheer scale of the data required a global effort, Seales partnered with Silicon Valley entrepreneurs Nat Friedman and Daniel Gross to launch the Vesuvius Challenge in March 2023. The project released thousands of 3D X-ray scans and open-source software tools to the public, offering over $1 million in prize money to anyone who could successfully decode the text. The competition galvanized a worldwide community of computer vision experts, machine learning engineers, and student researchers, turning a niche archaeological puzzle into a high-stakes technological race.[1][4]

The breakthrough came from an unexpected realization about the physical properties of the ink. While the carbon ink didn't show up as a bright spot on an X-ray, it did subtly alter the surface texture of the papyrus. When the water-based ink dried, it caused the plant fibers to warp slightly, creating a microscopic "crackle" pattern. Contestants began training deep learning models to recognize these minute topological variations. By treating the 3D scans like video frames and feeding them through neural networks, the AI learned to detect the exact locations where ink had been applied, even when it was invisible to human researchers.[3][4]

In late 2023, the world witnessed the first fruits of this labor. Two students—Luke Farritor, a SpaceX intern, and Youssef Nader, a biorobotics graduate student in Berlin—independently trained machine learning models that revealed the first complete word from inside a sealed Herculaneum scroll. The word was "porphyras," the ancient Greek term for purple dye. The discovery sent shockwaves through the academic community, proving definitively that the carbonized ink could be detected and that the invisible library could, in fact, be read.[3][6]

The momentum accelerated rapidly. In early 2024, a team of students claimed the Vesuvius Challenge Grand Prize by decoding over 2,000 characters across 15 columns of text from a single scroll. Papyrologists translated the newly revealed Greek text, identifying it as a previously unknown philosophical work by Philodemus of Gadara, an Epicurean philosopher who lived in Herculaneum. The text discussed the nature of pleasure, specifically analyzing how the scarcity or abundance of food and music affects human enjoyment. For the first time in two millennia, a lost voice from antiquity was speaking again.[1][6]

The technological pipeline continued to yield astonishing results into 2025. Researchers turned their attention to a scroll known as PHerc. 172, housed at Oxford. By refining their segmentation and ink-detection algorithms, they managed to read the colophon—the final section of the scroll where the author and title are typically recorded. Without ever unrolling the papyrus, they identified the work as "On Vices," another text by Philodemus. It marked the first time the title of a still-rolled Herculaneum scroll had ever been recovered non-invasively, providing scholars with a direct map of the library's contents.[2][5]

As the project moves through 2026, the focus has shifted from proving the concept to scaling the operation. The Vesuvius Challenge community is currently working to fully automate the virtual unwrapping pipeline, minimizing the need for painstaking manual segmentation of the 3D meshes. Armed with faster scanning protocols developed at the ESRF, the team aims to image and read the remaining 300 intact scrolls over the next few years. What began as a painstaking, decades-long quest by a single laboratory has blossomed into a highly efficient, open-source factory for recovering ancient literature.[4][7]

The implications of this technology extend far beyond the scrolls currently sitting in museum vaults. The Villa of the Papyri has only been partially excavated, and archaeologists believe that the main library—potentially containing thousands of additional scrolls, including lost Latin masterpieces—still lies buried beneath the volcanic rock of Herculaneum. For centuries, the high cost of excavation and the inability to read the carbonized papyri kept the shovels at bay. Now, with the digital key to the library finally in hand, the push to unearth the rest of the villa has gained unprecedented urgency, promising a new Renaissance of classical knowledge.[4][7]