Artificial intelligence is enabling researchers to read carbonized scrolls buried by Mount Vesuvius nearly 2,000 years ago, date biblical manuscripts more precisely, map hundreds of new Nazca geoglyphs in Peru, and reconstruct rules of a forgotten Roman board game.
Vesuvius Scrolls Yield New Philosophical Texts
In June 2026, the Vesuvius Challenge team announced the first near-complete virtual unrolling of a Herculaneum scroll (PHerc. 1667), revealing approximately 1.5 meters of continuous Greek text across 20 columns. The content discusses ethics, impulse (horme), practical wisdom (phronesis), and human nature in terms consistent with Stoic philosophy.
A second scroll (PHerc. 139) confirmed the existence of “Philodemus, On Gods, Book 8,” extending a previously known single-book work by the Epicurean philosopher into at least an eight-book series.
Earlier, in 2024, AI-assisted decipherment of another Herculaneum scroll provided details on Plato’s burial site in a private garden near the Academy’s shrine to the Muses in Athens, along with accounts of his final night listening to a flute player.
The technique relies on CT scans, X-ray imaging, and machine-learning models trained to detect ink traces within charred papyrus layers without physically unrolling the papyrus. Hundreds of additional scrolls from the Villa of the Papyri remain for future analysis.
Dead Sea Scrolls Redated Closer to Biblical Composition
An AI model named Enoch, trained on radiocarbon-dated fragments and handwriting geometry, has produced date estimates for 135 Dead Sea Scroll manuscripts. In 79% of cases evaluated by paleographers, the predictions aligned with realistic ranges, often 50–100 years earlier than prior scholarly estimates.
For manuscript 4Q114 containing portions of the Book of Daniel, Enoch, and new radiocarbon data placed the text between 230–160 BCE, aligning it more closely with the period of its presumed composition around the Maccabean revolt. Similar shifts occurred for other biblical texts.
The model combines angular and allographic writing-style features with Bayesian ridge regression, achieving mean absolute errors of 27.9–30.7 years relative to radiocarbon benchmarks, arXiv reported.
Nazca Desert Geoglyphs Nearly Double in Known Count
Between September 2022 and February 2023, researchers from Yamagata University’s Nasca Institute, using an AI object-detection model on high-resolution aerial imagery, identified 303 new figurative geoglyphs in Peru’s Nazca region, per a PNAS paper. This nearly doubled the previously cataloged total of roughly 430 after a century of human surveys.
The new figures, mostly smaller relief-type glyphs depicting humanoids, domesticated animals (especially llamas), decapitated heads, and ceremonial motifs, cluster near ancient foot trails—averaging 43 meters away, suggesting they were viewed and created at the individual or small-group level. Larger line-type geoglyphs, by contrast, more often show wild animals and align with ceremonial pathways.
Subsequent AI-supported surveys have continued to add to the catalog.
Roman Board Game Rules Reconstructed from Wear Patterns
In 2026, researchers used the Ludii AI game simulation system to analyze a limestone slab (Object 04433) from the Roman settlement of Coriovallum (modern Heerlen, Netherlands). By testing hundreds of rule variants against microscopic wear patterns on the stone, the AI identified the most likely gameplay as a blocking game.
One player controls four “dogs” and the opponent two “hares”; the goal is to immobilize the opponent by restricting movement rather than capturing pieces. The reconstructed game, dubbed Ludus Coriovalli, matches wear patterns produced in thousands of simulated matches and aligns with blocking-game traditions previously documented only from later medieval Europe.
Broader AI Applications and Future Outlook
Beyond archaeology, AI systems are accelerating scientific discovery. Stanford’s Biomni agent performs multi-step biomedical workflows at expert level, including rare-disease diagnosis and molecular cloning protocol design.
Models such as DT-Transformer predict disease onset from electronic health records, while frameworks like SLIViT analyze 3D medical scans with specialist-level accuracy in far less time.
Virtual cell models, protein design tools, and multi-agent research systems are shortening timelines for hypothesis generation, data analysis, and experiment planning. Researchers anticipate continued gains in reading remaining Herculaneum scrolls, mapping additional cultural sites, and integrating AI into laboratory and clinical pipelines.
These tools do not replace human expertise but extend it, allowing verification of outputs and interpretation of recovered data at previously unattainable scale.
Provided by Dallas Express
