AI Unlocks the 2,000-Year-Old Secrets of the Herculaneum Scrolls

For more than two and a half centuries, the Herculaneum scrolls were considered one of archaeology’s most agonizing mysteries. Buried beneath a deluge of superheated volcanic ash, the ancient library was transformed into a collection of brittle, blackened lumps that resembled discarded charcoal. To the naked eye, they offered nothing. Yet, locked inside these carbonized cylinders were the unfiltered thoughts of the classical world, perfectly preserved but entirely inaccessible. Now, an unprecedented convergence of particle physics, crowdsourced citizen science, and artificial intelligence is finally coaxing these ancient voices out of the dark.[2][4]

The story of the scrolls begins in 79 AD, when Mount Vesuvius violently erupted, obliterating the Roman cities of Pompeii and Herculaneum. While Pompeii was buried in falling pumice, Herculaneum was engulfed by a massive pyroclastic surge—a wave of scorching gas and volcanic mud. This intense heat instantly carbonized organic material, flash-frying a luxury seaside estate believed to belong to Lucius Calpurnius Piso Caesoninus, the father-in-law of Julius Caesar. Inside this "Villa of the Papyri" lay the only large-scale library from classical antiquity to survive into the modern era.[2][4]

When farmworkers digging a well in 1752 accidentally struck the villa's marble pavement, excavators soon uncovered more than 1,800 of these carbonized scrolls. Early attempts to read them were catastrophic. Well-meaning researchers and monks tried to physically unroll the papyri using knives, rosewater, and rudimentary mechanical rigs. The fragile ash simply crumbled to dust, destroying hundreds of irreplaceable texts in the process. The remaining unopened scrolls were eventually locked away in storerooms in Naples and Paris, written off as an unsolvable puzzle.[3][5]

The fundamental roadblock to reading the Herculaneum papyri was a matter of chemistry. Unlike medieval manuscripts written with iron gall ink, ancient Roman and Greek scribes used a carbon-based ink made from soot and water. Because the volcanic eruption had turned the papyrus itself into carbon, the ink and the paper were chemically identical. To the human eye, and even to standard medical X-ray machines, the letters were completely indistinguishable from the charred pages they were written on.[2][6]

The paradigm began to shift through the pioneering work of Dr. Brent Seales, a computer scientist at the University of Kentucky. Seales spent two decades developing a non-invasive technique called "virtual unwrapping." Rather than physically opening a fragile artifact, his team used high-resolution X-ray tomography to peer inside, creating a three-dimensional digital model of the object's internal structure. Computer vision algorithms could then theoretically trace the crumpled layers of the material and flatten them into a readable two-dimensional sheet.[4][6]

Seales proved the viability of virtual unwrapping in 2015 when his team successfully read the En-Gedi scroll, a charred parchment from the Dead Sea region containing text from the Book of Leviticus. However, the En-Gedi scroll used a metal-based ink that glowed brightly on CT scans. When Seales applied the same technique to the Herculaneum scrolls, the carbon-on-carbon problem persisted. The layers could be digitally flattened, but the pages appeared entirely blank.[4]

Realizing that the sheer scale of the computational problem required a global effort, Seales partnered with Silicon Valley tech entrepreneurs Nat Friedman and Daniel Gross in March 2023. Together, they launched the Vesuvius Challenge, a crowdsourced machine learning competition offering over $1 million in prize money. They released thousands of 3D X-ray images of two intact scrolls to the public, challenging the global computer science community to build an AI capable of detecting the invisible ink.[2][3][6]

To give the AI a fighting chance, the scrolls first had to be scanned at an unprecedented resolution. Researchers took the fragile artifacts to particle accelerators, such as the Diamond Light Source in Oxfordshire, UK. These massive synchrotron facilities generate intensely bright X-rays capable of scanning the scrolls at a resolution of 2 micrometers—roughly one-fiftieth the width of a human hair. At this microscopic scale, the physical "texture" of the ink sitting on top of the papyrus fibers finally became visible.[1][4]

The challenge for the competitors was to train machine learning models to recognize these microscopic topographical changes. The AI had to learn the subtle "crackle" patterns and minute variations in thickness that occurred wherever a scribe's pen had deposited soot onto the ancient plant fibers. By feeding the algorithms thousands of examples of known ink patterns from small, already-opened fragments, the models slowly learned to predict where ink was present on the completely hidden layers of the intact scrolls.[5][6][7]

The first major breakthrough arrived in October 2023, courtesy of Luke Farritor, a 21-year-old computer science undergraduate at the University of Nebraska-Lincoln. Utilizing consumer-grade graphics processing units, Farritor's algorithm successfully detected a cluster of Greek letters hidden deep within a scroll. When papyrologists examined the output, they were stunned to read the word "πορφυρας" (porphyras), the ancient Greek term for purple dye. It was the first time in two millennia that a word had been read from an unopened Herculaneum scroll.[4][7]

The momentum rapidly accelerated. In February 2024, the Vesuvius Challenge awarded its $700,000 Grand Prize to a trio of researchers: Farritor, Youssef Nader, an Egyptian biorobotics graduate student in Berlin, and Julian Schilliger, a Swiss robotics engineer. By combining their approaches, the team developed a robust auto-segmentation tool that could trace massive sheets of papyrus through the 3D scans. Their combined AI model revealed 15 columns of continuous text, totaling more than 2,000 characters.[2][3][6]

When classical scholars finally translated the recovered text, they found themselves reading a previously unknown work of Epicurean philosophy. Experts believe the author is likely Philodemus of Gadara, a philosopher and poet who was the resident intellectual at the Villa of the Papyri. The text dives deeply into the nature of pleasure, which Epicureanism identifies as the highest good in human life, provided it is pursued with moderation and understanding.[2][4][5]

In the newly deciphered columns, the ancient author wrestles with the economic and psychological concept of scarcity. He questions whether the availability of goods, such as food or music, affects the pleasure they provide. Does a rare delicacy bring more joy simply because it is scarce? The author argues against this, writing: "As too in the case of food, we do not right away believe things that are scarce to be absolutely more pleasant than those which are abundant."[1][3]

The technology is now advancing at an exponential rate, moving beyond the initial prize-winning scroll. In early 2025, researchers applied generalized AI models to PHerc. 172, a Herculaneum scroll housed at the Bodleian Libraries in Oxford. The software successfully revealed multiple columns of text, including the repeated use of the ancient Greek word "διατροπή" (diatropē), meaning disgust. Because the AI models no longer need to be retrained for the unique internal geometry of every single scroll, the pace of discovery is rapidly scaling.[1]

The implications of these breakthroughs are sending shockwaves through the humanities. Recently, the technology was used to decode a scroll that pinpointed the exact burial spot of the philosopher Plato in Athens—a historical detail that had been lost to time. For classical papyrologists, the Herculaneum library represents the holy grail of archaeology. It is the only intact library from the classical world, offering an unedited window into the intellectual life of the Roman Empire right before its peak.[4][5]

The tantalizing prospect now facing scholars is what might be hidden in the remaining 800 unread scrolls, or in the thousands more that archaeologists believe are still buried in the unexcavated lower levels of the villa. The library could hold lost masterpieces of antiquity: missing plays by Sophocles or Aeschylus, complete scientific treatises by Archimedes, early Roman histories, or foundational texts of early Christianity that were otherwise destroyed during centuries of religious upheaval.[3][6]

Beyond Herculaneum, the success of the Vesuvius Challenge is poised to revolutionize the broader field of cultural preservation. Papyrologists and archivists are already preparing to deploy these AI-driven virtual unwrapping techniques on other severely damaged artifacts worldwide. From the carbonized scrolls of Petra in Jordan to water-logged manuscripts in European archives and fire-ravaged documents from modern conflicts, the technology offers a lifeline to history that was previously considered unsalvageable.[5]

What began as a tragedy in 79 AD has culminated in one of the most remarkable interdisciplinary triumphs of the 21st century. By bridging the gap between ancient scribes and modern machine learning, researchers have not just salvaged a few scraps of papyrus; they have built a time machine. The thoughts of our ancestors, locked in mud and ash for two millennia, are finally stepping out of the dark and back into the light.[1][3]