Artificial Intelligence Estimates Ages of Dead Sea Scrolls Using Carbon-14 Data
A new AI system called Enoch has analyzed the handwriting of 135 ancient manuscripts to estimate when they were produced.
A collection of 135 Dead Sea Scrolls has received new age estimates following a study that utilized an artificial intelligence system named Enoch. Published in 2025, the research focuses on how ancient writing preserves historical clues, while also emphasizing why these chronological estimates require careful interpretation by scholars.
Discovered in caves starting in 1947, many fragments of the Dead Sea Scrolls reached researchers without reliable production dates. The overall collection contains texts associated with periods ranging from the third century BC to the second century AD. Establishing the specific era of each copy helps researchers better understand the circulation of writings and the evolution of intellectual traditions in ancient Judea.
For decades, specialists relied primarily on paleography, the study of ancient scripts, to estimate the age of documents lacking recorded chronological indications. However, researchers faced a shortage of examples with known dates among the investigated centuries, which weakened certain comparisons and increased doubts regarding historical writing styles. To address this challenge, a team led by Professor Mladen Popović from the University of Groningen combined material evidence with graphic patterns.
The Enoch artificial intelligence system analyzes visual differences in how ancient scribes drew letters without needing to understand the meaning of the words. A neural network called BiNet first identifies ink traces in digitized images, separating these marks from the material's texture. The program then measures characteristics such as curves, slants, and shapes, generating numerical descriptions of the writing for each examined fragment.
The system performs comparisons across two scales, considering both minor stroke curvatures and the overall appearance of the forms used by ancient scribes. This method allows the model to find regularities among manuscripts from close time periods without letting the subject matter of the text influence the graphic indicators of its production. According to the study published in PLOS One, this approach prevents similar words from artificially swaying estimates across different periods.
Training for the Enoch model began with 24 manuscript samples whose materials had undergone carbon-14 dating, a method that provides probable age ranges for organic material. Researchers then linked these intervals to the visual characteristics of the writing, teaching the system to recognize relationships between image shapes and chronology. Because physical measurements do not offer a single exact year, the system works with probability distributions to present chronological estimates accompanied by margins of uncertainty.
During tests, the average absolute error ranged from 27.9 to 30.7 years, depending on how small variations in predictions were handled. The system subsequently estimated dates for 135 manuscripts lacking direct dating, and specialists judged 107 of those results to be plausible, representing about 79 percent of the evaluated document set. Nevertheless, this paleographic evaluation does not independently confirm each date through individual physical measurement, nor does it guarantee equivalent performance on documents from other historical periods.
The new analyses suggest that certain writing styles may have circulated earlier than previously proposed dates for specific Dead Sea Scrolls. For instance, the findings indicate that Hasmonean and Herodian script styles may have coexisted since the end of the second century BC in that region. Consequently, the chronology of these texts has gained new possibilities for discussing when certain religious works began circulating in handwritten copies during that era.
Sources
- A formação de fósseis que leva o nome desta cidade cearense tem uma camada de 2 metros tão cobiçada que ganhou uma impressão digital química - Brasil 247 - Tendencias brasil247.com
- Uma inteligência artificial analisou a escrita de 135 manuscritos antigos para estimar quando eles foram produzidos - Olhar Digital olhardigital.com.br