For nearly two thousand years, hundreds of ancient scrolls lay buried under volcanic mud, resembling nothing more than lumps of charcoal. Historians and archaeologists knew these artifacts held lost secrets of antiquity, but opening them was physically impossible without destroying them. Now, a breakthrough collaboration between computer scientists, student researchers, and artificial intelligence has achieved the impossible. They have successfully read the text inside the carbonized Herculaneum scrolls without ever unrolling them.
In 79 AD, the eruption of Mount Vesuvius destroyed Pompeii and Herculaneum. While Pompeii was buried in ash, Herculaneum was hit by a pyroclastic flow—a superheated wave of gas and mud. This surge carbonized the library of a luxury estate known as the Villa of the Papyri, thought to be owned by the father-in-law of Julius Caesar.
When excavators discovered the library in the 18th century, they found over 800 scrolls. Early attempts to open them were disastrous. The papyrus had turned into fragile blocks of carbon. Pulling them apart caused them to crumble into dust. For centuries, these scrolls remained a tantalizing locked box. Scholars suspected they contained lost works of Greek philosophy and literature, but the data remained inaccessible.
The turning point came in 2023 with the launch of the Vesuvius Challenge. This was a machine learning and computer vision competition launched by tech entrepreneurs Nat Friedman and Daniel Gross, alongside computer scientist Dr. Brent Seales from the University of Kentucky.
Dr. Seales had spent years developing “virtual unwrapping” techniques. He used high-resolution CT scans to map the internal structure of the scrolls. However, there was a significant hurdle. Unlike medieval manuscripts that used metal-based inks which show up clearly on X-rays, the Herculaneum scrolls used carbon-based ink (charcoal and water). On a CT scan, carbon ink on carbonized papyrus is invisible to the human eye. They possess the exact same density.
The Vesuvius Challenge released the high-resolution 3D X-ray scans of the scrolls to the public. They offered over $1 million in prizes to anyone who could build an AI model capable of detecting the subtle texture changes left by the ink on the papyrus fibers.
The breakthrough did not come from a massive government laboratory or a tech giant. It came from a global team of students and independent researchers who collaborated online.
The Grand Prize of $700,000 was awarded to a team of three:
The process began when Luke Farritor successfully trained a machine learning model to recognize a “crackle” pattern. The ink, when it dried 2,000 years ago, altered the surface of the papyrus slightly. The AI detected these microscopic topographic differences. Farritor became the first person in history to see a full word from the inside of an unopened scroll: porphyras, the ancient Greek word for “purple.”
Following this initial success, the team combined their efforts. They improved their code to auto-segment the layers of the scroll and detect ink with higher accuracy. By the submission deadline, they had deciphered four passages of text containing 140 characters each, with at least 85% of characters recoverable.
The deciphered text provides a fascinating glimpse into the mind of an ancient philosopher. The text is written in ancient Greek and is attributed to Philodemus of Gadara, an Epicurean philosopher and poet who lived in the Villa.
The translated passages do not reveal state secrets or historical dates. Instead, they are a meditation on pleasure. Philodemus writes about the senses, specifically music and food. In one section, he questions whether things that are scarce offer more pleasure than things that are abundant.
He challenges the views of other philosophical schools, likely the Stoics, regarding how to define the “good life.” The text essentially serves as a blog post from antiquity, offering a sophisticated critique of how humans experience joy. The final sentences uncovered so far suggest a playful tone, where Philodemus puts down his ideological opponents.
The success of the Vesuvius Challenge proves that the concept works. We can read the invisible history inside these burnt artifacts. This victory has profound implications for the study of the ancient world.
The Villa of the Papyri is the only library to survive from the Greco-Roman world. The 800 scrolls excavated so far are likely just a fraction of the collection. Archaeologists believe thousands more may still be buried in the unexcavated lower levels of the Villa.
This technology opens the door to rediscovering “lost” history. Scholars hope to find:
The organizers have announced the next phase of the challenge. The goal is no longer just reading snippets. The new objective is to read 90% of a single scroll. As AI models become more efficient at “segmentation” (digitally peeling the layers of the papyrus), we may soon have access to a digital library of literature that hasn’t been read in two millennia.
How does the AI see the ink if it is invisible to X-rays? The AI does not “see” the ink in terms of color or density contrast. Instead, it analyzes the texture. The dried ink created a microscopic crinkle or fracture pattern on the papyrus fibers. The machine learning models were trained to recognize this specific texture as evidence of a letter.
Are the physical scrolls damaged during this process? No. The scrolls remain inside their protective cases. The entire process uses non-invasive CT scans (Computed Tomography) generated by a particle accelerator called the Diamond Light Source in the UK. The “unwrapping” happens entirely on a computer screen using 3D data.
Who owns the physical scrolls? The physical scrolls are the property of the Italian government and are housed at the National Library of Naples. The Vesuvius Challenge works in partnership with the library and conservators to ensure the safety of the artifacts.
Can I try to read the scrolls? Yes. The Vesuvius Challenge is an open-source project. The 3D scans and the winning code from the previous competition are available online. Anyone with programming skills and a powerful computer can download the data and attempt to improve the ink-detection algorithms.
What is the difference between this and standard OCR? Standard OCR (Optical Character Recognition) reads visible text from an image (like scanning a PDF). This project requires “Volume Cartography.” The software must first define the 3D shape of the twisted, crushed papyrus sheet within the scroll, digitally flatten it, and then detect the texture of the ink. It is significantly more complex than standard text recognition.