AI Fully Deciphers a Carbonized Papyrus from Vesuvius

What happened?

The Vesuvius Challenge, an international competition rewarding the reading of Herculaneum papyri, announced that it has successfully deciphered continuously the preserved text of scroll PHerc. 1667, a Stoic philosophical treatise on ethics. The breakthrough is detailed in a scientific paper and relies on a new methodology combining phase-contrast X-ray microtomography performed at the European Synchrotron Radiation Facility (ESRF) in France, with a machine learning workflow that detects ink, virtually unrolls the papyrus, and systematically transcribes the text.

The team had previously deciphered isolated fragments in 2023 and won the grand prize of $700,000 in 2024 with the scroll PHerc.Paris.4. Now, for the first time, an entire scroll has been read from beginning to end of the preserved portion. The text reveals passages such as: “Having exerted ourselves to the utmost through research and learning, we will no longer be inferior to them in any respect, equally performing the things that befit them and possessing the same practical wisdom as they do.”

Why is it important?

The Herculaneum papyri were carbonized by the eruption of Vesuvius in 79 AD and remained buried until the 18th century. They are the only intact library from antiquity that survives, with hundreds of scrolls still unopened. Until now, attempts to physically unroll them destroyed outer layers, as happened with PHerc. 1667 itself, of which only an 8 cm core remains. The new non-invasive technique allows reading the interior without damaging the support.

The breakthrough represents a qualitative leap: from local and exceptional recovery of isolated words to systematic recovery at the full-scroll scale. This opens the door to reading hundreds of philosophical, literary, and historical texts thought lost forever.

Consequences and context

The success of the Vesuvius Challenge demonstrates the power of collaboration between archaeology, accelerator physics, and artificial intelligence. The developed methodology could be applied to other carbonized or damaged materials, such as the Oxyrhynchus papyri or even medieval manuscripts. For historians of philosophy, the discovery of new works by Epicureans and Stoics could change the understanding of ancient thought.

The team also identified the content of another scroll, PHerc. 139, as a copy of the eighth book of the treatise On the Gods by the Epicurean philosopher Philodemus. This confirms that the library of the Villa of the Papyri contained works by known authors and possibly lost texts.

What readers should know

• The deciphered scroll, PHerc. 1667, originally measured between 19 and 24 cm in height; only an 8 cm core remains due to previous manual opening attempts.
• The technique combines phase-contrast microtomography (more sensitive than conventional tomography) with an AI model trained to detect carbonized ink on carbonized papyrus.
• The workflow includes: scanning, ink detection, 3D modeling of the scroll surface, virtual “unrolling,” and automatic transcription with neural networks.
• The Vesuvius Challenge plans to apply the method to the hundreds of remaining scrolls, though each may require adjustments due to differences in preservation and ink.
• The full text of PHerc. 1667 is already available to the academic community, and new scrolls are expected to be published in the coming years.

Implications for the future

This milestone marks the beginning of a new era in reading unreadable ancient texts. The combination of synchrotron and AI could become standardized as an archaeological tool. Furthermore, it demonstrates that open innovation competitions (like the Vesuvius Challenge) can accelerate discoveries that would otherwise take decades.

“The key transition marked by this work is moving from exceptional local recovery to systematic recovery at the scroll scale,” the team wrote in their paper.

For tech companies, the case illustrates how machine learning applied to 3D imaging data can solve complex virtual restoration problems. AI and computer vision startups could find applications in heritage conservation, materials analysis, or even reading documents damaged by fires or disasters.