How old are the Dead Sea Scrolls? An AI model can help
Scientists are using it to estimate the age of ancient handwriting
EVER SINCE the Dead Sea Scrolls were discovered by Bedouin shepherds in the 1940s, debate has raged over their exact age. The scrolls, which contain the earliest surviving copies of books from the Hebrew Bible and other religious texts, mostly written in Aramaic and Hebrew, are thought to have been compiled sometime between 300BC and 200AD. Dating each of the 1,000-odd individual scrolls would help historians understand how literacy spread among ancient Jewish populations and the first Christians, and offer a valuable window into the genesis of the sacred texts. But scholars hoping to do so have had little but their own intuition to rely on.
Until now. In a paper published in PLOS ONE on June 4th, scientists report that a new artificial-intelligence (AI) model can date the ancient scrolls based on the style of handwriting they contain. This is possible because writing can change in distinctive ways even within a few generations (a much-mourned modern example is the decline in cursive). Scholars already look for such differences to estimate the age of ancient documents, but the degree of subjectivity involved means that different experts often reach conflicting conclusions.
The new AI model offers the promise of standardising the discipline. It draws its conclusions by accurately measuring small angles and curves within individual letters, as well as identifying patterns across larger chunks of text, in ways that humans cannot. It also allows its calculations to be examined, which the researchers hope will lead to more objective date estimates. Indeed, the model has already made several intriguing findings.
The model, called Enoch, was developed by a team led by Mladen Popovic, a scholar of religion from the University of Groningen in the Netherlands. To calibrate the model, Dr Popovic and his team extracted and carbon-dated tiny samples from 24 of the Dead Sea Scrolls. The team then fed Enoch the carbon-date estimates, as well as 62 scanned images of the dated scrolls. Their intention was to allow the model to find relationships between shapes and patterns in the scanned script and the physical age estimate given by the carbon dating. The team then validated the model by giving it extra, unseen scans from the carbon-dated scrolls as a test; it proved robust, providing age ranges that largely overlapped with the carbon-dating results. Enoch was then provided with images of 135 undated scrolls and asked to offer dates. The age ranges it gave were generally between 50 to 100 years older than human estimates.
The most striking of the new dates concerned two scrolls that contain fragments of the biblical books of Daniel and Ecclesiastes. Historians believe that the original text of the Book of Daniel was finished sometime around 160BC and the Book of Ecclesiastes in the third century BC. Enoch suggests the versions found in the Dead Sea Scrolls were written around those times, too. Dr Popovic says that though it is unlikely that the scrolls were written by the original authors of the Bible—an assessment he makes based on the quality of the script—they could have been contemporary copies, perhaps jotted down as scribes were listening to the originals being read out loud. The result is sure to spur further investigation.
An AI model that can help scholars date manuscripts “is a significant contribution", says Thea Sommerschield, a historian at the University of Nottingham who has made use of AI models to restore and explain ancient Greek inscriptions, and who was not involved with the work. Dr Popovic hopes that models such as Enoch will one day be able to help date ancient manuscript collections in any language.
Collecting enough data to train similar models for other scripts will take time. For now, Dr Popovic is happy to be reducing the outsize role that gut feelings play in palaeography. “Sometimes," he says, “our human mind is more of a black box than…the AI model that we have built."
Curious about the world? To enjoy our mind-expanding science coverage, sign up to Simply Science, our weekly subscriber-only newsletter.
topics
