A darkened ancient papyrus scroll from Herculaneum, its carbonized surface illuminated to reveal faint Greek letters — recovered not by human hands but by machine learning and light.

The Virtual Unrolling

The Virtual Unrolling

When Silicon Meets the Sealed Scroll — There is a profound symmetry happening right now in the world of archaeology and ancient history. At the very same moment our culture is arguing about whether AI is a tool or a threat, a group of researchers is using it to do something that can only be described as a miracle of recovery. For nearly two thousand years, the library of the Villa of the Papyri in Herculaneum has sat in silence. When Mount Vesuvius erupted in AD 79, it did not just bury the city — it turned an entire collection of scrolls into what scholars have called “charred lumps of coal.” To physically unroll them would be to destroy them. They have been readable in theory and unreadable in practice, sitting in museums for centuries as objects of frustration. Until now. —

What the Vesuvius Challenge Is

In 2023, computer scientist Brent Seales — who had been working on this problem for over twenty years — partnered with Silicon Valley entrepreneur Nat Friedman to launch an open-source competition called the Vesuvius Challenge. The goal was simple and enormous: use AI and high-resolution CT scanning to read what no human eye could see. The technical problem was devilish. The ink used on these scrolls was carbon-based, chemically almost identical to the carbonized papyrus it was written on. There was no contrast for X-rays to detect. You were looking for writing that had become, in a sense, invisible — hidden inside layers of fused, collapsed scroll that could not be opened. The solution was to train machine-learning models to detect something subtler: the microscopic texture of ink embedded in the papyrus. Not the chemistry. The physical texture. The AI learned to see the difference between ink and burned fiber at a resolution far beyond what human vision can process.

“The process required me to trace the writing to create black and white images of the ink. I came to learn how this specific scribe used to write his letters — how he would draw a letter from a certain point. You can see, with high resolution scans, ink deposits where he starts drawing the letter.” — Youssef Nader, 2024 Grand Prize winner

In early 2024, a team of three student researchers — none of whom could even read ancient Greek — won the Grand Prize by successfully deciphering over 2,000 characters from a sealed scroll. The text that emerged, written by the Epicurean philosopher Philodemus of Gadara, contemplates how abundance or scarcity affects sources of pleasure — music, food, the good life. It had been waiting, perfectly preserved and completely inaccessible, for 1,945 years. —

What Has Come Out of the Scrolls

The 2024 breakthrough was only the beginning. A scroll now held at the Bodleian Libraries at Oxford — scroll PHerc. 172 — was scanned at the Diamond Light Source synchrotron in Harwell, Oxfordshire. This facility uses a particle accelerator to generate X-ray beams billions of times brighter than the sun, producing detailed 3D models of the scrolls’ internal structure without a single layer being touched. The AI then worked through those images and, for the first time in nearly two millennia, the inside of that scroll is now visible. Among the first words recovered: *ἀδιάληπτος* — “foolish.” *Διατροπή* — “disgust.” *Φοβ* — “fear.” *Βίου* — “life.” Scholars believe the scroll is another work of Philodemus, one of the leading Epicurean thinkers of the first century BC. But perhaps the most astonishing single discovery came from a different scroll entirely. A team of Italian researchers, using AI alongside optical coherence tomography and infrared hyperspectral imaging, deciphered over a thousand words from a scroll that had been sitting in a museum since 1750. What they found was a text called the *History of the Academy* — written by Philodemus about Plato’s school in Athens. Buried inside that text was something scholars had searched for since antiquity. The exact burial location of Plato. After 2,400 years, a carbonized scroll — recovered from the ruins of a villa destroyed by a volcano — told us that Plato was buried in a private garden within the Academy, near a shrine to the Muses, in a space reserved for him alone. The Academy was later destroyed by the Roman general Sulla. The neighborhood where it stood in Athens is still called *Akadimia Platonos* — Plato’s Academy — today. —

The Irony That Should Not Be Lost

I wrote about this at length in my post on AI — how Socrates himself argued *against* writing. He feared that if his students wrote things down, they would stop genuinely remembering. They would carry the appearance of wisdom without the substance of it. The irony: we know almost nothing about Socrates because he refused to write. Everything we have comes through Plato, his student, who chose to use the tool. Now here is the second layer of that irony. Plato’s own location — the burial place of the man who preserved Socrates — was itself lost for two millennia. And it was recovered, in 2024, by artificial intelligence reading a scroll no human hand could safely open. The tool Socrates feared gave us Plato. And another kind of tool — one Plato could not have imagined — gave us back Plato’s grave.

This technology does not create the content. It removes the barriers of decay so that we can engage with the wisdom of the past. It is not the author. It is the assistant. The GPS for a territory we thought was gone forever.

The Vesuvius Challenge has now awarded over $1.7 million in prizes. Its current goal is to fully read ninety percent of the four completely scanned scrolls. Scholars estimate there are still hundreds of scrolls buried at Herculaneum that have never even been excavated — potentially including lost works of Stoic philosophy, Epicurean thought, and literature from the first century BC that exists nowhere else on earth. —
Technical Brief — What Is Actually Being Used
01 — The Vesuvius Challenge

An open-source competition launched in 2023. Teams worldwide build machine-learning models to detect the texture of ancient ink inside 3D CT scans of carbonized papyrus. Over 1,000 teams have entered. The Grand Prize was claimed in early 2024. Phase two is now underway.

02 — Diamond Light Source Synchrotron (Oxfordshire, UK)

A particle accelerator that generates X-ray beams billions of times brighter than the sun. Used to scan scroll PHerc. 172 from Oxford’s Bodleian Libraries in July 2024. Produces detailed 3D structural models without any physical contact with the fragile papyrus layers.

03 — Optical Coherence Tomography + Infrared Hyperspectral Imaging

The imaging combination used by the Italian team that uncovered Plato’s burial location. Works by building up layers of imaging data — visible light, infrared, thermal — that together reveal text invisible in any single spectrum. Then AI processes the combined dataset.

04 — The Invisible Library

Scholars estimate there are hundreds of thousands of text-based heritage materials in libraries and museums worldwide — scrolls, codices, tablets — that are too damaged or fragile to read. Brent Seales calls this the “invisible library.” The methods developed through the Vesuvius Challenge are now being applied to this broader problem.

Why This Matters Beyond the Headlines

The library at Herculaneum was not a general collection. It was a private Epicurean library — the personal books of a philosophical school. Most of what is still sealed inside these scrolls will be Epicurean thought: ethics, mathematics, music, the history of philosophy. These are not peripheral texts. Epicurean philosophy shaped how the ancient world thought about pleasure, pain, mortality, and the good life. Almost all of it was lost. The texts that shaped Roman intellectual culture in the first century BC — the very conversations that Cicero, Virgil, and Horace were reacting to — are largely gone. Or were largely gone. Some of those conversations may be in these scrolls, waiting for a machine-learning model to find the ink inside the carbon. I think about that every time someone tells me AI is only about replacing human workers or generating fake images. Sometimes the tool shows up exactly when the work demands it. The ancient world is not gone. It is sealed. And we are, right now, in the extraordinary position of being the generation that gets to open it — not by destroying what remains, but by reading it where it lies. That is not a small thing. —

Discover more from JAYMS.COM

Subscribe now to keep reading and get access to the full archive.

Continue reading