To gain new, more accurate insights into the ancient texts of the legendary scrolls, researchers from the University of Groningen turned to artificial intelligence (AI) and radiocarbon dating. The team now claims that some of the Dead Sea Scrolls may be older than previously thought.
In the mid-20th century, the first Dead Sea Scrolls were discovered in the caves of Qumran, located in the Judean Desert, by Bedouin shepherds. The oldest manuscripts, which included legal documents and texts from the Hebrew Bible, had been dated for decades to the 3rd and 2nd centuries BCE.
Now, the university team, led by Professor Mladen Popović, has employed AI in their research to uncover new information about the dates of individual scrolls. The results have called into question long-held scientific beliefs about when, where, and by whom these legendary texts were created. The new analysis revealed that some scrolls date back to the 4th century BCE.
What Did the Scientists Discover?
“It’s like a time machine. We can shake hands with those people who lived 2000 years ago,” said Professor Popović. He noted that some scrolls had been dated using radiocarbon methods in the 1990s, but contamination from castor oil had hindered accurate results. This oil was used in the 1950s to help experts read the manuscripts.
In the journal PLOS One, the team explained how they conducted new radiocarbon dating on 30 samples from various manuscripts found at four sites. Initially, the scientists meticulously cleaned the artifacts of castor oil, as reported by The Guardian.
After successfully performing radiocarbon dating, the researchers found that while two samples turned out to be younger than previously believed, most were older. Among these was a fragment of a manuscript known as 4Q114, which contained verses from the Book of Daniel.
Next, the team utilized a machine learning model called Enoch. It was trained on 62 digital images of ink traces from 24 manuscripts that had been dated using radiocarbon analysis.
The model was then tested with 13 additional images from the same manuscripts. In 85 percent of cases, the model produced dates that matched the radiocarbon analysis results.
When the AI model was tasked with dating images from 135 undated manuscripts it had not seen before, it successfully dated 79 percent of them in accordance with the estimates of expert paleographers.
Meanwhile, Dr. Matthew Collins from the University of Chester (UK) cautioned that radiocarbon dating sheds light only on the age of the parchment, not on when the text was written on it. He also raised questions about how representative this small number of samples was for training the model.