“I am can’t believe It worked!” says Nat Friedman, co-founder of the Vesuvius Challenge, which offers a $1 million prize to anyone who can use artificial intelligence (A.I) to refer to papyrus scrolls carbonized by the eruption of Mount Vesuvius in 79 AD.advertisement. But it works. On February 5 Mr. Friedman announced that a three-person team had been awarded $700,000 for successfully extracting four paragraphs of text from a scroll known as the Banana Boy, each at least 140 characters long and at least 85% legible. The three winners, Luke Faritor, Youssef Nader and Julian Schilliger, are all computer-science students.
The scroll is one of hundreds found in the library of a Roman villa in Herculaneum, believed to have belonged to Julius Caesar’s father-in-law. Along with hundreds of other scrolls in the villa’s library, it was damaged by the burning gases that engulfed the city during the same eruption that also buried the nearby city of Pompeii.
The text from the scrolls is difficult to read because the heat has reduced them to brittle charcoal logs; All attempts to unroll them physically result in them falling apart. So the focus shifted to finding ways to practically unravel them through computer analysis 3D Scroll scan created using X– ray This makes scrolling a software problem—but a very complex one
Virtual unrolling is a two-stage process pioneered by W. Brent Sills, a computer scientist at the University of Kentucky. The first stage, called segmentation, involves marking the edges of the rolled-up papyrus sheet inside. 3D Scan, then extract 2D Image of the surface of the scroll. The second stage, ink detection, analyzes the resulting images to tease out the ink of the scroll’s text from the papyrus background. This is especially difficult for the Herculaneum Scrolls, which are written in carbon-based ink, so there is little contrast with the background of the carbonized papyrus.
Dr. Sills, along with Mr. Friedman and Daniel Gross, two technology entrepreneurs, pondered A.I Techniques can be fruitfully brought to bear on both of these problems, and it has launched a prize challenge to find out. A community of thousands of enthusiasts has since developed a variety of tools and techniques to speed up the tedious process of segmenting and identifying individual letter inks and then whole words. In October 2023 Mr. Faritor and Mr. Nader were awarded the Small Prize for independently extracting the first legible word (“porphyrus”, meaning “purple”) from the Banana Boy Scroll (so named because of its size and shape).
The two students then joined Mr. Schilliger, further developing the machine-learning technique involved in ink detection. By manually labeling regions known as ink, they can train a neural network to find more of them; These were then fed back into the model to improve its detection capabilities. Mr. Nader modified the neural network into a novel architecture called TimesFormer, which produced sharp results. Mr. Schilliger, meanwhile, developed a tool to further automate the segmentation process (much of which must still be done manually).
The deadline for submitting results for the grand prize was late December, and the trio were awarded after the entries were evaluated by a team of papyrologists. (The three runners-up will each receive a smaller prize of $50,000.) The winning entry featured 15 columns of text written in Greek. Reading it was “mind blowing”, said Federica Nicolardi, a papyrologist at the University of Naples Federico II, who was one of the judges. The text is believed to be a previously unknown work on the pleasures of Philodemus, an Epicurean philosopher who lived in Herculaneum.
Mr. Friedman now wants to expand the entire process. With ink detection solutions, he says, “the bottleneck is now segmentation.” Mr Schilliger’s auto-segmentation tool is a big step forward, and he has agreed to open source it and collaborate with others to improve it. More prizes are being offered as incentives. Mr. Friedman, meanwhile, aims to scan more scrolls and standardize the scanning process using a particle accelerator from Britain’s Diamond Light Source.
That will cost money. After awarding $1.2 million, some of it from his own pocket, Mr. Friedman is looking for other backers to help the project. He hopes that deciphering the ancient scrolls will lead to the rediscovery of lost works from antiquity—”each scroll is a mystery box,” he says—and, ultimately, will reignite interest in further excavating the Villa of Herculaneum, which may contain thousands more. ■