For hundreds of years, hundreds of burnt Herculaneum scrolls had been entombed in ash after the 2,000-year-old eruption of Mount Vesuvius in A.D. 79. Unearthed in the remains of a villa, it is said to have belonged to Lucius Calpurnius Piso Caesoninus, Julius Caesar’s father-in-law. Now, artificial intelligence is tearing down those ancient barriers and unveiling long-hidden secrets inscribed in these scrolls.
The 1,800 rolls of Herculaneum papyri are the only library from the classical world that has escaped intact in bulk, discovered in the 18th century by a farmworker. For many years, attempts to read them let the rolls crumble into pieces, often turning to dust. Indeed, the only way it has been possible for researchers to have any success is by piecing together the thousands of fragments—an agony of a mosaic-like task that could take hundreds of years. For example, the contest was staged last year for $1 million to whoever came up with a scheme that would allow for the reading of the 270-odd sealed scrolls. A team led by University of Kentucky’s Professor Brent Seales organized a contest called the Vesuvius Challenge, pinned related software, and thousands of 3D X-ray images to the web, hoping the world would send them solutions for crowdsourcing problems.
It wasn’t the first such contest: a team Seales had earlier succeeded in virtually unwrapping an ancient scroll from Israel using X-ray tomography and computer vision. But that technique wasn’t good enough either: the ink of the Herculaneum scrolls is still practically invisible to X-rays.
It was then, really, that artificial intelligence began to take over the course of things—with machine-learning algorithms trained on scans applied by three students—Luke Farritor from the U.S., Youssef Nader from Egypt, and Julian Schilliger from Switzerland—to the folds of 2,000 letters held inside one of the scrolls. They had created a 3D scan of the text using a CT scan of the scroll before its surface was turned into segments where a machine-learning model shined inked regions to show text.
Their effort took the first prize of $700,000.
One of the contest’s funders, Nat Friedman, posted on social media that they had managed to decipher new text from the ancient world that had never before been read from 15 columns found at the end of the first scroll. This very text would be authored by Philodemus of Gadara, a practitioner of the Epicurean philosophy. It expresses not only in music but in food and in general joy of life’s pleasures, for example, throwing shade at the rival philosophical bent: “nothing general or particular to say about pleasure.” In the early 2000s, the completion of the Vesuvius Challenge began as Seales and his team began working up non-invasive technologies to assist in uncovering text latent thousands of years deep within compressed, carbonized layers of papyri. Using a 2019 grant from the National Endowment for the Humanities, they worked out machine-learning methods that can finally make out letters and words picked out from the 3-D X-ray scans of the scrolls. In March 2023, Seales released both the software and images into the public domain, thus igniting an international competition to unroll the scrolls. “The funding we got from NEH was an absolutely crucial step without which the Vesuvius Challenge never would have been possible,” said Seales. In addition, software, examples, and experience gained during the funded research had served as a start point for the contestants to do the breakthrough work.
Curiously enough, none of the top contenders turned out to be experts in classical studies.
All of them, in contrast, found the problem irresistible due to a technical challenge and prize bait.
“The first couple of letters that started arriving on my phone, that was a really surreal development,” Luke Farritor said, an undergraduate at the University of Nebraska-Lincoln and a SpaceX intern. He had spent dozens of nights in the office, laboring over the scrolls, trying to tease some sense from them, and was attending a house party when the eureka moment arrived. To be honest, this AI-driven adventure is not a breeze-like one, as one may imagine it. Two main missions the participants were to hold a concentration for: segmentation and ink detection. Segmentation involved virtually unwrapping the scrolls to sketch clean surfaces of papyrus, while the ink detection involved complex algorithms to simply determine the patterns of ink from the scans. And it was only hours later that an Australian physicist thought to enlist the power of the human visual system after all. Hour after hour, he scoured page after page until he finally discovered ink, for which he won a prize. This was the breakthrough that headed the competitions, like Farritor, to build off of his work and finally crack whole sections of text. What Vesuvius Challenge has come up with will be quite literally changing papyrology and classics forever. Tobias Reinhardt, classicist at the University of Oxford, told Newsweek these advances could lead to a recovery of ancient texts at a volume not seen since the Renaissance. It is hoped that the methods developed will be applied to numerous other damaged texts around the world, giving a chance to unravel even more historical secrets.
But that’s not all for Nat Friedman. He wants to take it a step further than the system of the triumphant team and have 90 percent of the four scrolls that have already been scanned and read in 2024. This would literally be a game-changer for the understanding of two-millennium-old texts and would totally revolutionize knowledge of classical literature and philosophy.
For nearly 2,000 years, the scrolls from Herculaneum have kept their secrets. At the intersection of artificial intelligence and the united forces of both scientists and enthusiasts across the globe is when these ancient texts are ready to finally share their stories with the modern world.
