“After all, human handwriting and all of its variations and idiosyncratic features is a deeply human thing,” paleographer Christopher Rollston cautioned, as researchers began asking whether a set of manuscripts already famous for their age might need to be dated all over again.
Nevertheless, this narrative continues with a tiny, near-incidental noise: a rock hitting pottery in one of the caves along the Dead Sea. Since that time, in 1947, the Dead Sea Scrolls, almost 1,000 manuscripts discovered in caves near Qumrān, became a sort of time capsule of the ancient Judaism and the origins of the Hebrew Bible. Their age has been lingering around in the scholarly imagination decades at around 2000 years, charted by two inaccurate instruments: radiocarbon dating of parchment and paleography, the trained eye of letterforms.
A more recent development has now made that map narrower by making the old tools cooperate. A team of researchers, with the University of Groningen archaeologist Mladen Popović as the primary investigator, used cleaned radiocarbon results together with the processed results of image-based handwriting analysis, and used them to train an artificial intelligence model, Enoch. In the underlying study, researchers performed radiocarbon dating on 30 manuscript samples dated using radiocarbon dating were analyzed in 30 different locations yielding 27 valid dates after pretreating the parchment to remove contaminants.
That step of cleaning, is more than meets the eye. Some scroll fragments were also coated with castor oil in the 1950s to make the faint ink easier to read, which is a conservation decision that may complicate subsequent measurements. The team tried to remove the fatty residues through chemistry before dating to make sure that the clock that was being read was part of the parchment itself rather than the preservation of the parchment in the middle of the century.
The images of ink traces of 24 radiocarbon-dated manuscripts were used to train Enoch, which was then tested on more images in the same set; the predictions made by the model were found to overlap the radiocarbon probability ranges in approximately 85 percent of cases. As the researchers provided it with the pictures of manuscripts that had not been dated, expert paleographers rated approximately 79 percent of the A.I. estimates as credible. Popovic put the ambition into very straightforward terms: What we have invented, Popovic put it is, is a very strong instrument that is empirically grounded–grounded on physics and on geometry.
The most interesting conclusions come to a point where scriptural arguments have been the hottest of all: the development of the scripts and the issue of the time when important biblical compositions were written down. According to the model, two styles of writing included over time the Hasmonean and Herodian one, and an attractive line of writing was not so well defined, the writing of one hand took over the writing of another. It also re-dated 4Q114, which includes lines in Daniel, to a span that extends to earlier than standard paleography could; the work shows an interval of 230160 BCE, instead of an unhesitant dating in the middle of the second century BCE. Effectually, a manuscript that had been considered as the duplicate of a subsequent reader starts to seem nearer in time to the reality that created the writing.
The revised ranges pull and tug at an older assumption concerning Qumrān itself. In the event that there are more manuscripts older than the primary occupation of the settlement, the caves could be the hiding place of a library, but not its workshop. The implication was captured well by Joan Taylor of King’s College London, who observed that the dates imply that most of the manuscripts discovered in the caves near to Qumran would not have been written in the location of Qumran, which was not settled until later.
To curators and conservators, the practical promise is less dramatic but greater: it is reduced motives to sample fragile texts. Destructive testing can never be comprehensive as Popović has had to argue since there are over 1,000 manuscripts in the larger collection. What is attractive about Enoch is not its substitution of human judgment, but the fact it suggests another prism, constructed out of pixels and probability, with which the scholar can choose which narratives regarding the scrolls have survived, and which are worth drawing a new date in pencil into the margin.
