AI is reading scrolls burned in Vesuvius.

The Guardian looks back at a carbonized papyrus scorched by the lava that consumed Pompeii. University of Kentucky computer scientists have been able to decipher letters printed on the paper with the help of (hopefully not-hallucinating) archeological artificial intelligence:

The discovery was announced on Thursday by Prof Brent Seales, a computer scientist at the University of Kentucky, and others who launched the Vesuvius challenge in March to accelerate the reading of the texts. Backed by Silicon Valley investors, the challenge offers cash prizes to researchers who extract legible words from the carbonised scrolls.

“This is the first recovered text from one of these rolled-up, intact scrolls,” said Stephen Parsons, a staff researcher on the digital restoration initiative at the university. Researchers have since uncovered more letters from the ancient scroll.

To launch the Vesuvius challenge, Seales and his team released thousands of 3D X-ray images of two rolled-up scrolls and three papyrus fragments. They also released an artificial intelligence program they had trained to read letters in the scrolls based on subtle changes that the ancient ink made to the structure of the papyrus.

The unopened scrolls belong to a collection held by the Institut de France in Paris and are among hundreds recovered from the library at the villa thought to be owned by a senior Roman statesman, possibly Lucius Calpurnius Piso Caesoninus, the father-in-law of Julius Caesar.

Two computer science students, Luke Farritor in Nebraska and Youssef Nader in Berlin, who took up the Vesuvius challenge, improved the search process and independently hit on the same ancient Greek word in one of the scrolls: “πορφύραc”, meaning “purple”. Farritor, who was first to find the word, wins $40,000 with Nader winning $10,000.

The race is now on to read the surrounding text. Dr Federica Nicolardi, a papyrologist at the University of Naples Federico II, said three lines of the scroll, containing up to 10 letters, were now readable with more expected to come. A recent section shows at least four columns of text.