AI Decodes Ancient Herculaneum Scrolls

For nearly 2,000 years, hundreds of papyrus scrolls lay buried and carbonized under volcanic mud, their secrets locked away by the eruption of Mount Vesuvius. Attempts to open them resulted in dust and destruction. Now, a team of students using artificial intelligence has achieved the impossible: reading the text inside these fragile artifacts without ever unrolling them.

The Vesuvius Challenge Breakdown

In 2023, tech executives Nat Friedman and Daniel Gross launched the “Vesuvius Challenge.” They partnered with Professor Brent Seales from the University of Kentucky to solve a specific historical puzzle. Seales had already developed a method called “virtual unwrapping,” but the ink on these specific scrolls presented a unique problem that required machine learning to solve.

The challenge offered a $700,000 Grand Prize to the first team that could decipher four passages of text from the high-resolution CT scans of the scrolls.

The Winning Trio

Three students combined their skills to claim the prize in early 2024. This was not a project by a large government agency or a tenured research lab, but rather a collaborative effort by young researchers:

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

Farritor was the first to identify a complete word in the scans: “porphyras,” which means “purple” in ancient Greek. This initial discovery proved the concept worked and paved the way for the group to decode significantly more text.

How the AI Technology Works

The science behind this discovery is a mix of high-energy physics and computer vision. The physical scrolls are lumps of hardened charcoal. You cannot simply X-ray them to see the text because the ink (charcoal and gum arabic) and the papyrus (plant matter) have the same density. To a standard X-ray, they look identical.

The process involves three distinct scientific steps:

  1. High-Resolution Scanning: The scrolls were scanned at a particle accelerator in Oxford, England. This machine, the Diamond Light Source, produced 3D CT scans with a resolution of roughly 8 micrometers. This is high enough to see the individual fibers of the papyrus.
  2. Segmentation and Flattening: Before reading the text, the computer must understand the geometry of the scroll. The scrolls are crushed and twisted like a stepped-on newspaper. The team used software to trace the layers of papyrus within the 3D block and digitally “flatten” them into 2D sheets.
  3. Ink Detection Models: This is where the AI shines. The machine learning models were trained to look for “crackle patterns.” When the ink was applied 2,000 years ago, it altered the way the papyrus fibers dried and cracked. The AI does not see the ink color; it sees the microscopic texture disruption caused by the ink.

What the Scrolls Reveal

The text deciphered by the students is a philosophical treatise. It is written in Ancient Greek and is likely the work of Philodemus, an Epicurean philosopher who lived in the villa where the scrolls were found.

The translated passages do not speak of wars or laws. Instead, they discuss the nature of pleasure and the senses. The author poses questions about whether goods are more desirable when they are scarce.

Key insights from the translated text:

  • Music and Food: The author compares the pleasure derived from music to the pleasure of food and drink.
  • The Capers Debate: In a specific example, the text argues against the idea that scarce things are inherently better. The author notes that even if “capers” were rare, they would not necessarily be more pleasing than other food.
  • Philosophy of Life: The text generally attacks the stoic view that luxury is bad. It leans heavily into Epicureanism, which posits that pleasure is the highest good, though it defines pleasure as a state of tranquility rather than excess.

The Villa of the Papyri

The context of this discovery is as important as the technology. The scrolls come from a specific location in Herculaneum known as the Villa of the Papyri. This massive estate is believed to have been owned by Lucius Calpurnius Piso Caesoninus, the father-in-law of Julius Caesar.

When Vesuvius erupted in 79 AD, Pompeii was buried in ash and pumice. Herculaneum, however, was hit by a pyroclastic surge. This superheated wave of gas and mud hit the villa at temperatures exceeding 320°C (600°F).

This instantly carbonized the library. If the scrolls had just been buried, they would have rotted away centuries ago. The carbonization process essentially turned them into lumps of coal, preserving the structure for two millennia.

Why Previous Attempts Failed

For centuries, archaeologists tried to open these lumps.

  • Mechanical Unrolling: In the 1700s, early explorers tried to peel them open. The scrolls shattered into dust.
  • The Piaggio Machine: A priest named Antonio Piaggio invented a machine to slowly unwind them using silk threads. This worked for some, but it took decades and often damaged the insides of the scrolls.
  • The “Split” Method: In the 19th century, some scrolls were simply cut in half to read the center, destroying the outer layers.

The AI method is the first non-destructive way to access the information.

The Future: Vesuvius Challenge Phase 2

The success of Nader, Farritor, and Schilliger is just the beginning. The Vesuvius Challenge organizers have announced the next phase for 2024 and beyond. The goal has shifted from reading a few paragraphs to reading entire scrolls.

The implications for history are staggering. The Villa of the Papyri is the only intact library surviving from the Greco-Roman world. Excavations were halted years ago, and archaeologists believe there are two lower levels of the villa that have never been dug up. These lower levels could contain the main library.

Potential discoveries could include:

  • Lost plays by Sophocles or Aeschylus.
  • Missing books of Livy’s history of Rome.
  • Early Christian texts or contemporary accounts of the first century.
  • Latin literature (as most scrolls found so far are Greek).

The code and the methodology used by the students are open-source. This means researchers globally can now apply these ink-detection algorithms to other damaged manuscripts, potentially unlocking secrets from Egyptian mummy masks or medieval book bindings.

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 scanning and analysis are done in cooperation with Italian heritage officials.

Can I participate in the decoding? Yes. The Vesuvius Challenge is an ongoing open competition. The scans are available for download, and there are smaller prizes available for improvements in segmentation tools and ink detection algorithms.

Why is the text in Greek and not Latin? Educated Romans were bilingual. Greek was the language of philosophy and higher learning. However, scholars believe there is a separate Latin section of the library that has either not been fully excavated or resides in the unread scrolls.

How accurate is the AI translation? The accuracy is verified by papyrologists (experts in ancient scrolls). The AI produces an image of the Greek letters, which the scholars then read. It is not the AI “inventing” text; it is the AI revealing the visual data that humans can then interpret.