In October 2023, an email arrived on papyrologist Federica Nicolardi’s phone with an image that would transform her research forever, according to Nature. The image showed a fragment of a papyrus scroll that had been burnt in the eruption of Mount Vesuvius in 79 CE. The latest results revealed a strip of papyrus packed with neat Greek lettering, glowing bright against a darker background. “It was incredible. I thought, ‘So this is really happening’,” said Nicolardi, according to a report by Nature. She knew right then that papyrology would never be the same.
The breakthrough was part of the Vesuvius Challenge, a project that utilized artificial intelligence (AI) to decipher ancient texts that had remained unreadable for centuries. Brent Seales, a computer scientist at the University of Kentucky, and his colleagues tackled the task of reading the Herculaneum scrolls, which could not be seen at all, utilizing neural networks to detect ink patterns on the fragile scrolls, according to gadgets360.com. The Herculaneum scrolls were part of a cache of papyrus rolls buried and carbonized by the eruption of Mount Vesuvius, which wiped out the Roman cities of Pompeii and Herculaneum in 79 CE. The scrolls were discovered in the remains of a luxury Roman villa in Herculaneum in the eighteenth century, as reported by National Geographic.
In the spring of 2023, a college student named Luke Farritor, now 22, found himself riveted by a podcast as he drove to his SpaceX internship in Starbase, Texas, according to National Geographic. The podcast hosts were describing a competition with the goal of reading a 2,000-year-old scroll without physically unrolling it. “A lot of things about it were really compelling, the biggest one being that you're going to potentially discover a new library from the ancient world, and that's a big deal,” stated Farritor, according to a report by National Geographic. While focusing on space travel in his day job, Farritor, a computer science major, devoted his nights and weekends to the Vesuvius Challenge.
Contestants in the Vesuvius Challenge had to develop their own programs to interpret existing 3D scans of the coiled scroll, with the virtual-unwrapping step being the bottleneck currently limiting the data they had to work with, as reported by Nature. The main challenge in reading the Herculaneum scrolls was to virtually flatten the documents and distinguish the black ink from the charred remains to make the Greek and Latin writings legible, requiring the charting of subtle physical variations to detect writing on the charred material, according to Mundo Deportivo. The scribes of Herculaneum used carbon-based ink, invisible in scans because it had the same density as the papyrus, as noted by Nature.
In February 2024, a grand prize was awarded: computer science students Youssef Nader, Luke Farritor, and Julian Schilliger received $700,000 for producing 16 columns of clearly readable text, as reported by Nature. The winning team used a TimeSformer model, a variant of the transformer model usually used for videos. They relied on this neural network to reveal text attributed to an ancient Greek philosophical work. “We will have 400 columns of Greek text to read,” said Nicolardi, noting the challenge for papyrologists in managing the volume of text produced, as reported by Nature.
Scholars who study ancient societies oscillate between enthusiasm and doubt regarding the use of artificial intelligence to translate documents written thousands of years ago, according to Folha de Sao Paulo. There is no doubt that the technology can already do a lot and will advance enormously. The debate has focused on the technical potential of AI and whether it can substitute human work.
Researchers are using neural networks to tackle ancient languages for which only a small amount of text survives. Katerina Papavassileiou at the University of Patras and her colleagues used a recurrent neural network to restore missing text from 1,100 Mycenaean tablets from Knossos, Crete, as reported by Nature. Papavassileiou hopes to use models trained on Linear B to tackle Linear A, a script used by the Minoan civilization that shares many symbols with Linear B but has never been deciphered.
Seales noted that the success of AI in these applications underscores its role as a complement to human expertise. “What AI is doing is giving the papyrologists data to work on which they could not otherwise have,” stated Richard Ovenden, head of the Bodleian Libraries at the University of Oxford, as reported by Nature. He added, “It's making their work more important than it has ever been.”
As AI continues to evolve, its impact on science and technology promises to reshape our understanding of the world, with potential breakthroughs in decoding lost languages and exploring underground libraries, as reported by Nature. Researchers emphasize the need for interdisciplinary collaboration and open-source data to ensure transparency and replicability in AI applications for ancient texts, according to gadgets360.com.
In archaeology, AI models have accelerated the discovery of geoglyphs in Peru’s Nazca Desert, including symbols that had gone unnoticed by archaeologists.
The article was written with the assistance of a news analysis system.