AI Deciphers 2,000-Year-Old Herculaneum Scroll Buried by Vesuvius

texts that have been sealed away for nearly 2,000 years
Researchers used AI and X-ray imaging to decipher a carbonized scroll from Herculaneum without physically unwrapping it.
Mark

So they're reading a scroll without opening it. How is that actually possible?

Mimi

X-ray tomography creates a 3D map of the scroll's interior. Then AI looks for ink—which has a different chemical signature even when carbonized—and traces out the letters.

Luke

But they've only read about 26 lines per column so far. The core of the scroll is still sealed. We don't know if the AI can handle the densest, most fragile part.

Mark

Why does the core matter so much?

Mimi

That's where the title might be. Right now they're reading fragments—they know the word for disgust appears twice, but they don't know what work this is or what it's about.

Luke

And that's the honest position. They've made real progress, but the most important part is still unknown.

Mark

How many scrolls are still waiting?

Mimi

Nearly 2,000 carbonized scrolls and fragments from Herculaneum alone. This is one of three at Oxford. Most are still unread.

Luke

The Vesuvius Challenge has been running for years and offering serious money—$700,000 for the grand prize. That's how competitive and difficult this work is.

Mark

What happens if they do read the core?

Mimi

We get a window into Roman life and thought from someone close to Caesar himself. These aren't copies—they're originals that survived by accident.

Luke

If the AI can handle it. That's still the open question.

  • Nearly 2,000 charred papyri from Herculaneum have sat unread for centuries, too fragile to open and too damaged to read by conventional means — a vast archive locked inside its own destruction.
  • The Vesuvius Challenge has injected urgency and competition into the effort, drawing international researchers with millions in prize money and producing a landmark moment when a 23-year-old decoded the first word from an unopened scroll in 2023.
  • Digital unwrapping — using X-ray tomography and neural networks trained to detect carbonized ink — allows scientists to read the scrolls' interiors without ever touching them, turning catastrophic fragility into a solvable engineering problem.
  • Oxford's team has now deciphered the final columns of PHerc. 172, with the scroll's innermost core — potentially holding its title and central arguments — still waiting to be unlocked by the same methods.

Buried beneath volcanic ash for nearly two millennia, the carbonized scrolls of Herculaneum have long stood as monuments to what history swallows whole — present but unreachable, preserved by the very catastrophe that sealed them. Now, through a convergence of artificial intelligence, X-ray tomography, and classical scholarship, researchers at Oxford are reading words from a scroll that no human hand has opened since Mount Vesuvius erupted in 79 CE. The first word to surface — an Ancient Greek term for disgust — is a small but profound signal that an entire lost library may yet speak again.

In July 2024, Oxford's Bodleian Libraries received a scan of a carbonized scroll from Herculaneum, the Roman town obliterated when Mount Vesuvius erupted in 79 CE. Six months later, artificial intelligence and advanced imaging had done what centuries of scholarship could not: rendered individual words legible from a document no one had opened in nearly two thousand years. Among the first words to emerge was the Ancient Greek term for disgust — appearing twice in the scroll's opening columns.

The work is part of the Vesuvius Challenge, a prize competition that has galvanized researchers worldwide. In 2023, a 23-year-old named Luke Farritor earned $40,000 for identifying the first word ever translated from an unopened carbonized scroll, then shared a $700,000 grand prize as part of a three-person team. The Musk Foundation has contributed over $2 million to the effort in total.

The technique making this possible was pioneered at the University of Kentucky: rather than physically unrolling the scrolls — a process that risks destroying them — researchers use X-ray tomography to build three-dimensional models of their interiors. Neural networks then scan these models for the chemical traces of ink, which remains subtly distinct from the surrounding papyrus even after carbonization.

The scrolls originated in the villa of Julius Caesar's father-in-law, sealed by the same pyroclastic flow that killed Herculaneum's residents without warning. PHerc. 172, one of three scrolls now held at Oxford, has yielded roughly the last 26 lines of each column of text. Its innermost core — the most fragile section, and possibly the one containing the work's title — remains unread. Richard Ovenden, director of the University Libraries, called the moment historic: a collaboration between librarians, computer scientists, and classical scholars that is making visible what was thought lost forever.

In July 2024, researchers at Oxford's Bodleian Libraries received a scan of a carbonized scroll from Herculaneum—a Roman town buried under ash and pumice when Mount Vesuvius erupted in 79 CE. Six months later, using artificial intelligence and advanced imaging software, they had deciphered enough of the text to read individual words. One of the first to emerge was the Ancient Greek term for disgust, appearing twice in the opening columns. The scroll, catalogued as PHerc. 172, is one of nearly 2,000 charred papyri discovered in Herculaneum in 1750, many of them still unread after nearly two millennia.

The breakthrough belongs to a larger effort called the Vesuvius Challenge, a competition that awards cash prizes to researchers who can translate the scrolls' contents. The competition has drawn international attention, particularly after a 23-year-old named Luke Farritor identified and translated the first word from an unopened carbonized scroll in 2023, earning $40,000 for that individual achievement. Farritor was also part of a three-person team that won the competition's grand prize of $700,000 that same year. The Musk Foundation, which sponsors the challenge, has contributed over $2 million to the project.

What makes this work possible is a technique pioneered at the University of Kentucky: digital unwrapping. Rather than physically opening the fragile, tightly wrapped documents—a process that risks destroying them—researchers use X-ray tomography to create detailed three-dimensional models of the scrolls' interiors. Computer vision algorithms then scan these models, looking for patterns that indicate the presence of ink. The ink, even when carbonized, retains chemical differences from the surrounding papyrus, making it detectable to neural networks trained to recognize these subtle distinctions.

The scrolls themselves come from the villa of Julius Caesar's father-in-law, where they were carbonized by the pyroclastic flow that swept through Herculaneum during Vesuvius's eruption. For centuries, they remained sealed and unreadable—physical artifacts of a lost world, but not accessible to scholarship. The eruption came without warning. Those residents who could not flee faced sudden death; the towns were buried so completely that they vanished from memory until their rediscovery around 1,700 years later.

PHerc. 172 is one of three carbonized scrolls now held at Oxford, a gift from Ferdinand IV of Naples and Sicily in the early 19th century. The team has so far managed to read approximately the last 26 lines of each column of text. The innermost core of the scroll—its most fragile and potentially most valuable section—remains undeciphered. Researchers are hopeful that this central portion may contain the work's title, which could provide crucial context for understanding what they are reading. Richard Ovenden, director of the University Libraries, described the moment as historic: librarians, computer scientists, and classical scholars working together to access texts that have been sealed away for nearly 2,000 years. The convergence of advanced imaging, artificial intelligence, and humanities expertise has made visible what was thought lost forever.

The astonishing strides forward made with imaging and AI are enabling us to look inside scrolls that have not been read for almost 2,000 years.
— Richard Ovenden, director of the University Libraries
Möchten Sie die ganze Geschichte? Das Original lesen bei Gizmodo ↗
Kontakt FAQ