How AI is 'Unburning' the Lost Ancient Library of Herculaneum

In the autumn of 79 AD, the catastrophic eruption of Mount Vesuvius obliterated the Roman cities of Pompeii and Herculaneum, burying them under meters of superheated volcanic ash and pumice. Amidst the apocalyptic destruction, the eruption inadvertently preserved a staggering treasure of antiquity: the Villa of the Papyri. Believed by historians to have been the luxurious seaside estate of Lucius Calpurnius Piso Caesoninus, the father-in-law of Julius Caesar, the villa housed a massive private collection of texts. It remains the only intact library from the classical world to survive into the modern era. However, the intense heat of the volcanic pyroclastic flows flash-fried the library's contents. Roughly 1,800 papyrus scrolls were transformed in an instant into brittle, carbonized lumps that closely resembled charred logs [2, 4].[2][4]

For centuries following their initial discovery by Italian farmworkers digging a well in 1752, these artifacts remained one of archaeology's most tantalizing and frustrating mysteries. Excavators quickly realized that the blackened lumps were actually ancient books, but accessing the knowledge inside proved nearly impossible. Early attempts to physically unroll the fragile papyri were disastrous. Well-meaning researchers sliced them in half with knives, soaked them in rose water, or subjected them to a specialized unwinding machine invented by a Vatican conservator named Antonio Piaggio. While Piaggio's machine managed to peel off a few layers, the process severely damaged the scrolls, often shattering the carbonized papyrus and turning the ancient text into useless dust [2, 8].[2][8]

Recognizing the catastrophic damage caused by physical intervention, conservators eventually halted all attempts to open the remaining collection. For more than 270 years, the intact Herculaneum scrolls sat in museum vaults in Naples, Paris, and Oxford, widely considered a lost cause [2, 6]. They became the ultimate "invisible library"—a vast repository of human knowledge permanently locked away by the forces of nature. Scholars knew the scrolls contained significant philosophical and literary texts from ancient Greek and Roman thinkers, but the artifacts were entirely inaccessible. The consensus in the archaeological community was that the secrets of the Villa of the Papyri would remain hidden forever, barring a miraculous leap in technology [2, 8].[2][6][8]

That miraculous leap began to take shape in the 21st century, spearheaded by computer scientist Brent Seales and his research team at the University of Kentucky. Seales had spent decades pioneering the field of "virtual unwrapping," a non-invasive digital technique designed to read damaged manuscripts. His team achieved a landmark victory in 2015 when they successfully used X-ray tomography to read the En-Gedi scroll, a charred Hebrew parchment from the Dead Sea region, without opening it [5]. However, when Seales attempted to apply this same methodology to the Herculaneum scrolls, he hit a massive technological roadblock. The ink used on the Dead Sea scrolls contained heavy metals, which glowed brightly on X-ray scans. The ancient Romans at Herculaneum, by contrast, used a carbon-based ink made from soot and water [2, 5].[2][5]

To a standard medical or industrial X-ray machine, the carbon-based ink is virtually indistinguishable from the carbonized papyrus it sits upon. To overcome this invisibility problem, researchers realized they needed to look closer than ever before. They transported several of the fragile, charred scrolls to the Diamond Light Source in Oxfordshire, England [1, 7]. This facility houses a synchrotron—a massive circular particle accelerator where electrons are propelled to near light speeds, emitting X-ray beams 10 billion times brighter than the sun [1, 7]. By bombarding the scrolls with these intense beams, scientists captured incredibly detailed 3D tomographic scans of the artifacts' internal structures, mapping the microscopic, tightly wound layers of papyrus at a resolution of just a few micrometers [5, 7].[1][5][7]

Even with these unprecedented high-resolution 3D volumes, the text remained hidden. The scans provided a perfect geometric map of the crumpled, fused layers of the scroll, but the carbon ink still did not show up as a distinct material. The breakthrough came when researchers hypothesized that the ink, while chemically identical to the burned paper, might have altered the physical topography of the papyrus. They theorized that the application of the water-based soot ink would have caused the plant fibers to swell slightly, leaving behind microscopic textural anomalies—faint ridges and patterns that the team dubbed "crackles" [8]. Finding these crackles manually within terabytes of 3D data, however, was a task far beyond human capability.[8]

To solve this massive data problem, Seales partnered with Silicon Valley entrepreneurs Nat Friedman and Daniel Gross to launch the Vesuvius Challenge in March 2023 [3, 5]. Rather than keeping the research siloed in a single university lab, they open-sourced the high-resolution synchrotron scans and the custom virtual unwrapping software, releasing them to the public. They attached over $1 million in prize money to a series of specific milestones, effectively crowdsourcing the millennia-old puzzle to the global technology community [2, 5]. The initiative invited tech-savvy sleuths, computer vision experts, and machine-learning engineers worldwide to develop algorithms capable of detecting the invisible ink and coaxing the long-lost texts from the digital ashes [3, 5].[2][3][5]

The competitors were tasked with training artificial intelligence models to hunt for the microscopic crackles. The AI employed for this project does not understand Greek or Latin, nor does it attempt to guess missing words; rather, it functions purely as an advanced pattern-recognition tool [6]. The machine-learning algorithms were trained on small, manually verified fragments where ink was faintly visible, learning to map the physical presence of the ink in three-dimensional space [1, 6]. Once the AI detects the ink, specialized segmentation software virtually "unrolls" the crumpled layers of the 3D scan, flattening them out on a computer screen to reveal the hidden characters in their original sequence [1, 5].[1][5][6]

The global competition yielded spectacular results almost immediately. In late 2023, an American computer science student named Luke Farritor won a progress prize by becoming the first person in two millennia to read a word from a sealed Herculaneum scroll: "porphyras," the ancient Greek word for purple [2, 8]. Months later, in early 2024, Farritor teamed up with Egyptian PhD student Youssef Nader and Swiss robotics student Julian Schilliger to claim the $700,000 Grand Prize [5, 8]. Their combined, highly refined pipeline successfully decoded more than 2,000 characters across 15 columns of text from a single sealed scroll, far exceeding the competition's baseline requirements [2, 8].[2][5][8]

When expert papyrologists examined the virtually extracted images, they translated a previously unknown philosophical treatise. The text discussed the nature of pleasure, the availability of goods, and the impact of music on the human senses [2, 8]. Scholars strongly believe the work was authored by the Epicurean poet and philosopher Philodemus of Gadara, who is thought to have been the resident philosopher at the Villa of the Papyri [2, 8]. This discovery was particularly thrilling for classicists because Epicurean texts were rarely recopied by scribes during the Middle Ages, as their materialistic philosophy heavily conflicted with prevailing Christian doctrine [3]. The AI had effectively resurrected a voice that history had actively tried to silence.[2][3][8]

Since that initial Grand Prize, the momentum of the Vesuvius Challenge has rapidly accelerated. In 2025, the Bodleian Libraries at the University of Oxford announced a historic breakthrough regarding scroll PHerc. 172, one of three Herculaneum artifacts housed in their collection [1, 6]. This particular scroll had proven uniquely challenging due to its severe carbonization, but refined AI segmentation models managed to reveal substantial parts of the papyrus, including several columns containing about 26 lines of text each [6]. One of the first words deciphered from this new scroll was the Ancient Greek term "διατροπή," meaning "disgust," further confirming the text's philosophical nature [1, 6].[1][6]

The technology is now moving from bespoke, manual extraction to automated, large-scale recovery. Researchers are deploying generalist AI models that can recognize text across multiple different scrolls without needing to be painstakingly retrained for each artifact's unique damage patterns [5, 6]. The Vesuvius Challenge has launched a new phase aimed at "Unwrapping at Scale," incentivizing the community to automate the segmentation process so that entire scrolls can be read simultaneously [5]. The ultimate goal is to process the remaining 800 unread scrolls currently sitting in museum vaults, transforming a trickle of recovered words into a flood of ancient literature [5, 8].[5][6][8]

The implications of this technology for the study of classical antiquity cannot be overstated. Historians estimate that successfully decoding the entire Herculaneum library could more than double the amount of surviving literature from the ancient Greco-Roman world [2, 4]. Classicists are holding out hope that the unread scrolls might contain lost masterpieces: missing plays by Sophocles or Aeschylus, the lost volumes of Livy's monumental history of Rome, or foundational texts of early Western science and philosophy [2, 4]. The Villa of the Papyri is no longer a graveyard of burned books, but a time capsule waiting to be fully downloaded.[2][4]

Beyond the ashes of Mount Vesuvius, this AI-driven virtual unwrapping technology promises to illuminate history's darkest corners worldwide. Leading papyrologists and conservators are already exploring how these machine-learning techniques can be adapted for other endangered heritage materials [7]. The same algorithms that detect carbon ink on carbonized papyrus could theoretically be used to read fire-ravaged archives, water-damaged medieval manuscripts, or ancient texts fused together by time and decay [2, 7]. By proving that open-source collaboration between computer scientists and humanities scholars can solve millennia-old mysteries, the project has established a revolutionary new blueprint for archaeological discovery [6].[2][6][7]