AI Just Deciphered an Ancient Herculaneum Scroll Without Unrolling It

By Decrypt AI, Edited by Stacy Elliot

A 21-year-old University of Nebraska-Lincoln student has used AI to decipher ancient Greek letters inside a sealed scroll from Herculaneum, Italy, according to a report in Nature. The breakthrough offers hope for reading hundreds more scrolls buried during the 79 AD eruption of Mount Vesuvius.

Luke Farritor developed a machine learning algorithm that detected subtle texture differences between inked and blank areas of the rolled papyrus. His neural network highlighted over 10 Greek characters, including "porphyras" meaning purple, winning him the $40,000 "first letters" prize in the Vesuvius Challenge contest.

The contest was organized by researchers seeking to read the fragile, unopened scrolls found in Herculaneum, the only intact library surviving from ancient Greco-Roman times. Attempts to physically open the scrolls have yielded only fragments. Most of the 600+ scrolls, held in the National Library of Naples, remain unreadable.

In 2016, Brent Seales of the University of Kentucky used X-ray scanning to virtually unwrap and read a burnt scroll from Israel. But the ink on the older Herculaneum scrolls lacks metal and is invisible in scans. Seales realized texture differences caught by CT scans could indicate inked areas. He trained an AI on opened fragments to prove this, then scanned intact scrolls at high resolution.

The public Vesuvius Challenge launched in 2019, releasing scan data to engage AI experts globally. Farritor taught himself Latin as a child and was drawn to the contest. In June, the hint of visible "crackle" texture prompted him to refine his algorithm. One night in August it detected five clear letters, allowing rapid refinement to identify 10+ characters for the prize.

The word "purple" has not been found in opened scroll fragments before. As the first intact text read inside a still-rolled scroll, it offers hope for deciphering titles, authors, and full contents. The advance could reveal unknown ancient works, unlike surviving classical texts copied over centuries. Most Herculaneum scrolls relate to Epicurean philosophy, with parts of Epicurus' On Nature.

AI is increasingly aiding study of ancient texts, from Greek inscriptions to Egyptian cartonnage. Seales envisions machine learning unlocking an "invisible library" in layered palimpsests and discarded parchment. For now, Vesuvius Challenge leaders remain optimistic about reading full scrolls before the December grand prize deadline. Farritor's algorithm already sees more characters emerging.

Recommended News