Researchers Launch AI Project to Trace Dead Sea Scrolls

The European Research Council has awarded Professor Mladen Popovic of the University of Groningen an Advanced Grant of EUR 2.5 million, confirmed on the university's own announcement page, to lead a five-year project titled Tracing Scribes and Scrolls. The project, reported by The Jerusalem Post and JNS, brings together the University of Groningen, the Israel Antiquities Authority (IAA), and several other European laboratories and institutions, including the University of Pisa, the University of Naples Federico II, the University of Southern Denmark, KU Leuven, and the Egyptian Museums in Berlin and Turin. Researchers will analyze roughly 250 samples of parchment, papyrus, and ink drawn from the IAA's Dead Sea Scrolls collection, and for the first time will compare their chemical signatures against ancient Egyptian papyri. The project builds directly on Popovic's prior ERC-funded work, The Hands That Wrote the Bible, which pioneered AI-based identification of individual scribes; the new phase aims to locate where those scribes worked geographically, not just distinguish them from one another.

The pipeline described by Popovic and IAA researcher Dr. Ilit Cohen-Ofri layers three distinct evidence types: chemical "fingerprints" from lab analysis of parchment, papyrus, and ink; AI pattern-recognition applied to that chemical data to surface signatures not easily detected by conventional methods; and paleographic and codicological analysis of handwriting and physical manuscript construction. Cohen-Ofri said the work will produce an database of chemical composition data for the scrolls that the IAA has not previously compiled at this scale. Combining lab-grade chemical assays with visual and structural features from a small, irreplaceable sample set is a recognizable pattern for practitioners working on few-shot, multimodal inference in domains where data collection cannot be repeated or scaled.

The project is a useful reference case for cross-lab data harmonization: multiple European labs will need to calibrate chemical measurements against each other before the AI layer can treat their outputs as comparable inputs. It also raises the standard challenges of explainable modeling in a domain where provenance claims must trace back to specific, auditable measurements rather than black-box outputs, and of building reliable inference from roughly 250 samples spanning multiple material types.

Neither the university announcement nor the press coverage has published technical protocols or model details yet. Practitioners interested in the methodology should watch for peer-reviewed publications or preprints from the Groningen team, and for any public release of the chemical composition database Cohen-Ofri described, which would let outside researchers evaluate the approach directly.