AI Unlocks the Carbonized Scrolls of Herculaneum

This article explores how artificial intelligence, combined with high-resolution X-ray imaging, enables paleographers to read ancient scrolls buried by the eruption of Mount Vesuvius in 79 AD. By utilizing 3D mapping, virtual unwrapping, and machine learning algorithms, researchers can now detect carbon-based ink on carbonized papyrus without physically destroying the fragile artifacts. This technological breakthrough is finally revealing lost texts from antiquity that have remained unreadable for nearly two millennia.

The Challenge of the Herculaneum Papyrus

When Mount Vesuvius erupted, a wave of superheated gas carbonized hundreds of papyrus scrolls stored in a library at Herculaneum. While this process preserved the scrolls from decaying over time, it turned them into brittle blocks of charcoal. Early attempts to physically unroll them destroyed many fragments. Furthermore, the ancient ink used on these scrolls was made from soot and water, meaning it is also carbon-based. To the naked eye and standard X-ray machines, the black carbon ink is completely invisible against the black carbonized papyrus.

High-Resolution Scanning and 3D Mapping

The process of reading these scrolls begins with non-invasive imaging. Researchers place the intact scrolls inside a particle accelerator, known as a synchrotron, to perform high-resolution micro-computed tomography (micro-CT) scans. These scans capture thousands of X-ray cross-sections of the scroll at a microscopic level, creating a massive, detailed 3D digital model of the tightly wound internal layers.

Virtual Unwrapping Investigation

Before artificial intelligence can search for letters, the 3D digital model must be flattened. Because the scrolls were crushed and warped under volcanic mud, the internal layers of papyrus are folded and fused together in complex ways.

Researchers use specialized software to perform "segmentation." This involves tracing the individual, winding layers of the papyrus within the 3D scan. Once a single continuous layer is mapped out in three-dimensional space, the software digitally flattens it into a two-dimensional image. This process is known as virtual unwrapping.

Machine Learning Detects Invisible Ink

Once a section is virtually unwrapped, machine learning models are deployed to find the ink. Because the carbon ink and the carbonized papyrus have virtually the same density, the AI cannot simply look for color contrast. Instead, the AI is trained to detect microscopic structural differences.

When ink was applied to the papyrus thousands of years ago, it sat on top of the plant fibers, creating a tiny, almost imperceptible thickening on the surface. Deep learning algorithms are trained on small fragments of broken scrolls where letters are faintly visible under infrared light. The AI learns the exact micro-textures and sub-millimeter thickness variations that indicate the presence of dried ink.

Once trained, the AI scans the virtually unwrapped 2D images of the intact scrolls. It highlights areas where it detects the specific texture of the ink, translating those microscopic bumps into clear, white pixels against a dark background.

Assisting Paleographers in Translation

Artificial intelligence does not translate the text itself; rather, it acts as a digital magnifying glass for human experts. The AI generates flattened images featuring glowing clusters of Greek letters.

Paleographers—experts in ancient writing—take these AI-generated images to transcribe and translate the text. Because the AI reveals complete lines and columns of text in their original layout, paleographers can identify the handwriting style, understand the context of damaged words, and piece together philosophical works that have not been read since the Roman Empire. Through initiatives like the Vesuvius Challenge, this collaboration between computer scientists and historians is rapidly accelerating, turning an impossible archaeological problem into a scalable digital recovery effort.