AI Reads the Herculaneum Scrolls: A Breakthrough in Archaeology

For nearly 2,000 years, hundreds of ancient scrolls lay buried in mud and volcanic ash, inaccessible to historians. The eruption of Mount Vesuvius in AD 79 carbonized these documents, turning them into fragile lumps of charcoal that crumble if touched. Now, a combination of particle physics and artificial intelligence has achieved the impossible. Without physically opening the scrolls, researchers have successfully read entire paragraphs of text hidden inside.

The Problem: Carbon on Carbon

The “Villa of the Papyri” in Herculaneum contained the only surviving library from the classical world. When Vesuvius erupted, the pyroclastic surge toasted the papyrus scrolls instantly.

For centuries, archaeologists tried to open them. Early attempts involved mechanical contraptions that often destroyed the artifacts. In recent decades, scientists turned to X-ray Computed Tomography (CT) scans. This works well for many things, but it failed here for a specific chemical reason:

• The Paper: Made of carbon-based plant matter.
• The Ink: Made of charcoal and gum (soot), which is also carbon.

Because the ink and the paper have the same density, they look identical to a standard X-ray. The writing was invisible.

The Solution: The Vesuvius Challenge

In 2023, tech executives Nat Friedman (former CEO of GitHub) and Daniel Gross launched the Vesuvius Challenge. They partnered with Professor Brent Seales from the University of Kentucky, who had spent decades developing “virtual unwrapping” software.

They released high-resolution 3D scans of the scrolls to the public and offered over $1 million in prizes to anyone who could build an AI capable of detecting the ink.

The Breakthrough Team

The grand prize was claimed in early 2024 by a team of three young students who met online:

• Youssef Nader: An Egyptian PhD student in Berlin.
• Luke Farritor: A 21-year-old college student from Nebraska and SpaceX intern.
• Julian Schilliger: A robotics student from Switzerland.

Luke Farritor was the first human to see a complete word inside the sealed scroll. He trained a machine-learning model to look for subtle texture changes—specifically a “crackle” pattern—where the ink sat on the papyrus. The first word revealed was “porphyras,” the ancient Greek word for “purple.”

Building on this, the team combined their efforts to win the $700,000 Grand Prize. They successfully deciphered four distinct passages totaling over 2,000 characters.

What Do the Scrolls Say?

The text deciphered so far does not contain lost secrets of the Bible or government conspiracies. Instead, it offers a fascinating glimpse into the mind of an Epicurean philosopher.

Scholars believe the author is Philodemus of Gadara, the resident philosopher of the Villa of the Papyri. The text is a treatise on pleasure, music, and food.

In one translated section, the author discusses the senses and the source of pleasure. He argues that the availability of goods does not affect the pleasure they provide. He specifically asks if things that are scarce offer more pleasure than those that are abundant.

The text includes a humorous, relatable comparison:

“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.”

The author closes the thought by criticizing those who “talk about nothing else” but the distinct taste of capers. It is a specific, human moment frozen in time from 2,000 years ago.

How the AI "Reads" the Invisible

The process relies on a technique called “Ink Detection.” Here is the technical breakdown of how the students achieved this:

1. 3D Scanning: The scrolls were scanned at the Diamond Light Source in the United Kingdom. This particle accelerator produces X-rays 10 billion times brighter than the sun, allowing for incredible resolution (approx. 8 micrometers per voxel).
2. Segmentation: Before looking for ink, the computer must understand the geometry of the scroll. The scroll is wrapped in a spiral, but it is also crushed and twisted. The software virtually “peels” the 3D volume into flat 2D sheets.
3. Training the Model: The AI does not “read” Greek. It analyzes surface texture. The students trained a ResNet (residual neural network) on small fragments of broken scrolls where ink was visible to the naked eye.
4. Prediction: The AI applies what it learned to the closed scrolls. It looks at the texture of the papyrus fibers and predicts, pixel by pixel, where ink is likely present.

Why This Changes History

The success of the Vesuvius Challenge is arguably one of the most significant archaeological breakthroughs of the 21st century.

The Volume of Data There are approximately 800 scrolls stored in the National Library of Naples that have never been read. Furthermore, archaeologists believe the Villa of the Papyri has only been partially excavated. The scrolls found so far are mostly in Greek. A second library containing Latin texts may still be buried underground.

Lost Literature Most classical literature has been lost to time. We are missing huge portions of history:

• Livy: Only 35 of his 142 books on the history of Rome survive.
• Plays: Hundreds of plays by Sophocles, Aeschylus, and Euripides are missing.
• Philosophy: Early works of Stoicism and Epicureanism were previously known only through quotes in other books.

The AI models developed by Nader, Farritor, and Schilliger prove that these works can be recovered. The technology is non-invasive, meaning the artifacts remain safe while the data is extracted.

What Comes Next?

The Vesuvius Challenge has moved into its next phase. The goal for 2024 and beyond is to read 90% of the four scanned scrolls.

The cost of scanning is also a hurdle. The original scans required a massive particle accelerator. Researchers are now working on table-top scanning solutions that can provide similar resolution at a fraction of the cost. This would allow museums around the world to scan their unreadable papyri or even Egyptian cartonnage (mummy masks made of recycled papyrus) to find hidden texts.

Frequently Asked Questions

Who owns the physical scrolls? The 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 the Institut de France to access the data.

Is the AI guessing the words? No. The AI is not a large language model like ChatGPT that predicts the next word based on context. It is a computer vision model. It detects the physical presence of ink material. Papyrologists (experts in ancient texts) verify the letters visually to ensure they form coherent Greek sentences.

Can this technology read other damaged books? Yes. This technology has potential applications for other damaged historical documents, such as medieval manuscripts used in book bindings or documents damaged by fire or water. However, the current model is specifically trained on carbonized papyrus and carbon ink.

How long will it take to read all the scrolls? At the current pace, it could take years or decades. However, the Vesuvius Challenge aims to automate the “segmentation” process (the virtual unwrapping). Once that step is automated, reading the entire collection could happen very rapidly.