AI unlocking Dead Sea Scrolls’ hidden past

Researchers are rewriting the history of the Dead Sea Scrolls in a groundbreaking blend of ancient mystery and cutting-edge technology.

Using artificial intelligence alongside advanced radiocarbon dating, researchers have uncovered evidence that some of these iconic manuscripts may be much older than people ever imagined, Caliber.Az reports via The Guardian.

The Dead Sea Scrolls, discovered in the mid-20th century in the Qumran caves of the Judean desert, include a wide range of documents from legal texts to parts of the Hebrew Bible. Traditionally, they have been dated between the third century BCE and the second century CE.

Researchers led by Prof Mladen Popović from the University of Groningen used AI alongside radiocarbon techniques to refine dating accuracy. “It’s like a time machine. So we can shake hands with these people from 2,000 years ago, and we can put them in time much better now,” Popović said.

A major advance came when the team addressed contamination issues from castor oil—a substance applied in the 1950s to aid manuscript reading but known to interfere with radiocarbon results. After cleaning samples to remove this, the researchers dated 27 manuscripts, discovering some to be older than previously believed. Notably, they found that two distinct writing styles—Hasmonean and Herodian scripts—existed simultaneously much longer than thought, and a manuscript containing verses from the Book of Daniel (4Q114) dates closer to the author’s time.

The team developed an AI model called Enoch, trained on digital images of ink from radiocarbon-dated manuscripts. When tested, Enoch’s age estimates matched radiocarbon results 85 per cent of the time and produced tighter dating ranges. It successfully dated 79 per cent of 135 previously undated manuscripts, providing new leads on texts such as Ecclesiastes.

While promising, experts urge caution. Dr. Matthew Collins of the University of Chester noted that radiocarbon dating shows parchment age, not writing date, and that small AI training datasets may limit accuracy. “It’s an important and welcome study,” he said, “but one we should use alongside other evidence.”

By Naila Huseynova