AI and X-Rays Decode 2,000-Year-Old Herculaneum Scroll, Revealing Lost Philosophical Treatise

In August 79 AD, a superheated pyroclastic flow from Mount Vesuvius engulfed the Roman resort town of Herculaneum. Inside a luxurious seaside estate—believed to have belonged to Julius Caesar's father-in-law—an entire private library was instantly baked in an oxygen-starved environment. The intense heat carbonized the papyrus scrolls, transforming the greatest surviving collection of ancient literature into fragile lumps of charcoal. For centuries, these blackened artifacts sat in museum cases, holding the secrets of antiquity just out of reach.[1]

The physical reality of the Herculaneum scrolls makes them an archaeological nightmare. Any attempt to physically unroll the carbonized papyrus causes the material to shatter into dust, a tragic fate that befell many scrolls during early excavation efforts in the 18th and 19th centuries.

Even modern, non-invasive medical CT scans failed to reveal the text. The ancient scribes used a carbon-based ink made from soot and water, which absorbs X-rays in exactly the same way as the carbonized papyrus it sits on. To the scanner, the ink and the paper were indistinguishable.[2]

That impenetrable barrier began to crack with the launch of the Vesuvius Challenge in 2023. Founded by computer scientist Brent Seales and tech entrepreneurs Nat Friedman and Daniel Gross, the open-source competition offered over $1 million in prizes to anyone who could use artificial intelligence to decode the scrolls.

By crowdsourcing the problem and releasing massive 3D datasets to the public, the initiative sparked a rapid acceleration in heritage science, shifting the paradigm from wondering if a single letter could be read to planning the digital resurrection of the entire library.[4]

The most recent and profound breakthroughs have centered on a specific artifact known as PHerc. 172. Donated to the future King George IV in the early 19th century, the scroll is currently housed at the University of Oxford's Bodleian Libraries. Unlike some of its counterparts, PHerc. 172 proved to have a uniquely favorable chemical composition, making it an ideal candidate for the latest generation of high-energy scanning and machine learning analysis.[1][2]

In February 2025, the Bodleian Libraries and the Vesuvius Challenge announced a historic milestone: the first successful generation of images from deep inside PHerc. 172. The virtual unwrapping revealed a considerable portion of the papyrus, including multiple columns of text containing roughly 26 lines each. Papyrologists immediately began translating the newly visible Greek characters. One of the first words to emerge from the digital ether was "diatrope," an ancient Greek term meaning "disgust," which appeared twice within the visible columns.[1][2]

The momentum accelerated in May 2025, when researchers claimed the $60,000 First Title Prize by successfully decoding the scroll's colophon—the title page located at the very center of the rolled papyrus. Two independent teams utilized advanced segmentation to reveal that PHerc. 172 is a philosophical treatise titled "On Vices." The author was identified as Philodemus, a prominent Epicurean philosopher and poet who lived in the first century BC and whose teachings heavily influenced the intellectual elite of the late Roman Republic.[3]

This unprecedented access to the ancient world relies on a complex, multi-stage technological pipeline, beginning with particle physics. To see inside the dense, crumpled layers of charcoal, researchers transport the scrolls to massive synchrotron facilities, such as the Diamond Light Source in the UK and the European Synchrotron Radiation Facility (ESRF) in France. These particle accelerators propel electrons to near the speed of light, generating incredibly bright, high-resolution X-ray beams capable of mapping the internal structure of the scrolls at a microscopic scale of 9.2 micrometers.[1][4]

The synchrotron scans produce massive 3D datasets composed of billions of voxels, or three-dimensional pixels. The next step, known as segmentation, is the most labor-intensive part of the process. The papyrus inside the scroll is not neatly rolled; it is crushed, folded, and fused together by the volcanic heat. Researchers and specialized algorithms must painstakingly trace the continuous surface of the papyrus through the 3D volume, effectively mapping the topography of the crumpled sheet so it can be digitally flattened out on a computer screen.[4]

Once the papyrus is virtually unrolled, the artificial intelligence takes over. Machine learning models are deployed to hunt for the invisible ink. While the carbon ink lacks the density to show up clearly in standard X-rays, the AI is trained to detect microscopic textural anomalies on the papyrus surface—such as a faint "crackle pattern" left behind by the dried ink residue. Acting as a digital copyist, the neural network highlights these subtle topographical changes, painting the Greek letters back onto the digital papyrus for human scholars to read.

With the fundamental technology now proven, the Vesuvius Challenge published a comprehensive Master Plan in July 2025 to scale the operation. The central question is no longer whether the text can be recovered, but how quickly the process can be automated. The current pipeline still requires significant human intervention, particularly during the segmentation phase. The Master Plan outlines a roadmap for developing a "global optimal solution"—a fully automated AI system capable of isolating surfaces, unwrapping them, and detecting ink across multiple scrolls without constant manual correction.[4]

The financial and institutional backing required to execute this massive scale-up arrived in the summer of 2025. An international consortium led by Brent Seales secured a highly competitive $13.5 million Synergy Grant from the European Research Council. The initiative, dubbed the "UnLost" project, aims to optimize the scanning protocols and deploy the automated AI pipeline to X-ray and read the remaining 300 intact scrolls in the Herculaneum collection over the next two to three years, effectively building an open-access digital library of the ancient texts.[4]

The historical stakes of this endeavor are difficult to overstate. The vast majority of classical Greek and Roman literature was lost during the Middle Ages, with only a fraction of works surviving through centuries of manual copying by monks.

The Villa of the Papyri represents the only intact library to survive from antiquity. While the scrolls decoded so far have primarily yielded Epicurean philosophy, classicists harbor deep hopes that the remaining 300 scrolls could contain lost masterpieces—missing plays by Sophocles, poems by Sappho, or early historical accounts of the Roman Empire.[2]

Even if the library remains strictly philosophical, the recovery of texts like Philodemus's "On Vices" provides a priceless window into the minds of the ancient world. Epicureanism, which advocated for the pursuit of a tranquil life free from fear and pain, was a dominant intellectual force in the first century BC. These newly readable treatises offer fresh nuances on how the Romans debated ethics, pleasure, and virtue, restoring voices that were silenced by a volcano nearly two millennia ago.[3]

Despite the profound optimism, significant technical hurdles remain. The primary bottleneck is the sheer geometric complexity of the carbonized papyrus. The automated segmentation algorithms still struggle with areas where the scroll was severely crushed or where the layers have fused completely together. Until the machine learning models can reliably trace these chaotic internal structures without human guidance, the goal of reading the entire library within three years may face delays.[4]

Furthermore, the AI ink-detection models face a generalization problem. The models that successfully revealed the text in the first few scrolls relied on specific textural signatures and trace contaminants that may not be present in every document. Recent scans of other scrolls in the collection have shown that ink chemistry varies, and models trained on one scroll do not always perform well on another. Researchers are currently exploring even higher-resolution scanning techniques to capture new physical signatures that the AI can learn to recognize.[4]

Nevertheless, the trajectory of the research is irreversible. The collaboration between classical papyrologists, particle physicists, and machine learning engineers has achieved what was considered impossible just a decade ago. By transforming lumps of volcanic charcoal into readable digital texts, the Vesuvius Challenge is not merely recovering lost history; it is pioneering a new era of non-invasive heritage science. As the AI models grow more sophisticated, the world stands on the precipice of a digital Renaissance, ready to hear the voices of antiquity speak once more.[1]