AI meets the Dead Sea Scrolls

It’s been a big year for artificial intelligence. From President Donald Trump’s $500 billion Stargate AI initiative to the recent deal lending Disney characters to OpenAI, headlines have touted the promise of AI and its potential benefits—or possible harms—for modern life.

When an academic study earlier this year applied artificial intelligence to the science of Biblical manuscript dating, mainstream media reported it as a curious development. But what many missed at the time was how the new AI technique might lend unexpected credence to Biblical apologetics.

Tucked within caves overlooking the Dead Sea in Israel’s West Bank two millennia ago were the data centers of an ancient scribal community committed to preserving the sacred writings of Judaism. Used as early as 150 B.C. and rediscovered in 1947 by a Bedouin shepherd, the caves at Qumran contained manuscripts of Old Testament books like the Psalms, Isaiah, and Daniel, among other Hebrew writings. These manuscripts, known as the Dead Sea Scrolls, predated the oldest previously known copy of the Old Testament by more than 1,000 years.

Yet precisely dating the individual Dead Sea Scrolls hasn’t been simple. Few Hebrew or Aramaic manuscripts from the time period have dates directly written on them. The age of the Dead Sea Scrolls is significant because the older they are, the closer in time they would have been to the original manuscripts from which they were copied. That’s especially important for the Book of Daniel, whose date of authorship has been challenged by some modern scholars.

The traditional view is that the prophet Daniel completed the book around 536 B.C., during the reign of the Persian King Cyrus (Daniel 10:1). Yet liberal scholars doubt Daniel could have predicted future events. Daniel 11 describes the “abomination of desolation,” whose initial fulfillment was widely linked to the Seleucid King Antiochus IV Epiphanes sacrificing a pig to Zeus on the altar of the Jewish Temple around 168 B.C. Skeptical scholars say the book was likely composed afterward, around 160 B.C.

But the new AI dating of several Dead Sea Scroll manuscripts, including a section of Daniel, generally dated the documents as older than previous estimates. Christian apologists say that should prompt skeptics of Biblical history to question their presuppositions.

To date ancient scrolls, scholars largely rely on paleography, where they examine scripts and handwriting styles to estimate the date a document was written or copied. They also rely on carbon 14 dating, which estimates the age of organic materials. But some critics argue that scholars can endlessly manipulate carbon dating to suit their preconceptions.

With over 1,000 Dead Sea Scroll fragments lacking verifiable time stamps, researchers at the Netherlands’ University of Groningen set out to address the problem. They used a computerized neural network called BiNet to detect ink-trace patterns on 24 Dead Sea Scroll fragments and analyze letter curves and angles at the microlevel—a sort of computer-precision paleography. They implemented BiNet in their AI date-prediction model, which they named Enoch, and trained it on dated Qumran manuscripts—one a divorce certificate with a historical date written on it, and the rest with carbon 14 dates.

The researchers used Enoch to compare handwriting analysis with the team’s carbon dating of the fragments. In the study, published in PLOS One in June, the team’s carbon 14 results provided older date ranges for most of the tested manuscripts, finding only two are likely younger than previously thought.

“These results elicited some surprise and raised eyebrows in the direct scholarly community,” Lambert Schomaker, a co-author of the study and scientific director of the Artificial Intelligence and Cognitive Engineering department at Groningen University, told me in an email. According to the study, Enoch dated some parchments as B.C. that human paleographers had analyzed as being copied in A.D. The AI model also predicted radiocarbon dates within a 30-year margin—a precise range when dating documents over 2,000 years old.

Other researchers have begun to cite the study, with one academic paper at a September AI research conference calling it “paradigm-shifting.” In October, Associates for Biblical Research also highlighted the study.

Christian apologists say the results are particularly exciting for Biblical researchers.

“The AI is not using a human paleographer’s assumptions about the date of a manuscript to come to these conclusions. It’s only using hard data,” said Titus Kennedy, director of the Biblical archaeology program at Shepherds Theological Seminary and a research fellow at Discovery Institute. He said Enoch is “putting it all together,” showing AI’s capability to understand the correlation between carbon 14 dates and script style.

John Bloom, a professor at Biola University who teaches in its master’s program on science and religion, said the researchers’ carbon dating makes the study stand out. He referred to one of the Enoch-dated manuscripts, 4Q114, which contains Daniel 8-11.

“The 4Q114 is one example of several where the paleographic opinion was way off from the carbon dates,” said Bloom, who holds doctorates in physics and ancient Near Eastern studies. “The carbon 14 date for this manuscript really calls into question the skeptics arguing that [Daniel] chapter 11 had to be written in 164 B.C. or so.”

The study gave a radiocarbon date range for the Daniel fragment as 230-160 B.C. Bloom suggested the middle of that range, 195 B.C., was the best carbon date—and he added that a date of roughly 300 B.C., noted in a study appendix, might also be valid.

Robert Newman, a professor emeritus of New Testament at Missio Seminary, says in his article “Who Wrote the Book of Daniel—and When?” that the first attack on the authorship of Daniel came from a philosopher named Porphyry (c. A.D. 234-305). Porphyry believed it was impossible for anyone to predict the future, concluding that someone must have written Daniel between 165 and 160 B.C., long after the Babylonian exile. According to Newman, this thinking surfaced again in the 1800s with the rise of critical Biblical scholarship and can be found everywhere from seminaries to Sunday schools today.

Bloom noted that the study of apologetics takes into consideration the thoughts and attitudes of the people who were contemporaries of the Biblical authors and who considered the Biblical texts sacred. “They were much closer to the history and to copies of manuscripts than we are,” Bloom explained. “So what they accepted, I think is reasonable for us to accept.”

Since the year 160 B.C. is still within the range given by the Enoch study—just barely—Bloom doesn’t emphasize the study like he would if the results came in around, say, 400 B.C. But he said he does use it to show his students, “Here’s a wonderful case where the bias against miracles prevents people from following where the data seems to be going.”

For skeptics of the book of Daniel, Kennedy asks a simple question:  If 4Q114 had been copied just a couple of years after its original authorship, “How in the world, so quickly, did it get recognized as being part of the Biblical cannon?”

The results have certainly met with skepticism from scholars who believe the Book of Daniel cannot be older than the 160s B.C. Schomaker, whose role on the research team focused on evaluating patterns of handwritten shapes “from the ‘AI’ point of view,” defended the validity of the radiocarbon dated material Enoch was trained on.

“In my opinion, the physicists did an excellent job in making sure that the resulting date estimates are not biased,” he told me. He admitted that, since Enoch’s date ranges are learned from the carbon 14 labeled fragments, skeptics of the carbon 14 dates will likely doubt the resulting AI handwriting analysis.

Bloom also noted that the researchers took pains to ensure the carbon 14 samples weren’t contaminated by other organic materials. In recent decades, researchers have been figuring out how to remove the castor oil applied in the past to make manuscript letters legible. Groningen University developed a chemical process to remove the oil. “They’re really dating the parchment,” Bloom explained, “not any bacteria or oils or other junk that might have gotten absorbed into the parchment.”

Schomaker emphasized this study is just the beginning, adding that the researchers can boost Enoch’s accuracy by feeding it more manuscripts known to be from that time period.

As AI technology like Enoch gains recognition from worldwide scholars, it may prompt scholars to reevaluate other ancient manuscripts. “We have not applied Enoch yet to other scripts, though there is interest in that, but it would require some additional work,” Mladen Popović, the study’s lead author and professor of Hebrew Bible and ancient Judaism at Groningen, said in an email. “That is for scholars in those other fields to take further.”

Popović noted Enoch’s source code is freely available for other research groups to use, meaning more scrolls can be tested right away because others don’t have to start coding from scratch. According to Popović, scientists are currently conducting further research on manuscripts in Norway.

Bloom hopes the research will continue to advance, though he noted that few people have the skills and time to do the tedious work of studying ancient documents. “I think that computer tools that help scholars piece together manuscript fragments, read ancient texts, and identify which scribe wrote what text, is a great advance to the field of archaeology and paleography,” he said.