AI Deciphers Herculaneum Papyri Without Unrolling: An Archaeological Milestone

What Happened?

An international team led by Brent Seales of the University of Kentucky has succeeded for the first time in reading a carbonized Herculaneum papyrus without needing to unroll it. Using high-resolution X-ray tomography and artificial intelligence algorithms trained to detect traces of ink, they managed to reconstruct the complete text of a scroll that is part of the Herculaneum Papyri library, buried by the eruption of Vesuvius in 79 AD. The study, published in the journal Nature Communications, describes how the AI was able to distinguish carbon-based ink from the charcoal of the papyrus, something that seemed impossible. The analyzed papyrus contains a philosophical treatise from the Epicurean school, likely by the philosopher Philodemus of Gadara.

This milestone is the result of decades of work. Seales, a pioneer in the field of virtual manuscript imaging, had already managed to partially read other scrolls in 2015, but without obtaining continuous text. The key advance was the use of convolutional neural networks trained on micro-CT scans of fragments where the ink was visible. The team generated a 3D model of the scroll and applied AI to segment the layers and detect the ink, successfully identifying Greek letters with unprecedented accuracy. The analyzed scroll, known as PHerc.Paris.4, measures about 10 cm in diameter and contains around 200 columns of text.

The project is part of the Vesuvius Challenge, a competition launched in 2023 by tech investor Nat Friedman and Seales himself, offering prizes for advances in reading papyri. The challenge has mobilized hundreds of independent teams, and the current success demonstrates the viability of open collaboration in digital humanities.

Why Is It Important?

The Herculaneum papyri are the only complete library from classical antiquity that has survived, but most are illegible because they were carbonized, and any attempt to unroll them destroys them. Until now, only loose fragments could be read, mainly from the Villa of the Papyri, a luxurious residence that may have belonged to Julius Caesar's father-in-law. This achievement shows that it is possible to read thousands of texts without damaging them, which could revolutionize the study of ancient philosophy, literature, and science.

Furthermore, the technique used—machine learning applied to X-ray images—sets a precedent for other damaged materials, such as scrolls from the Nag Hammadi library or carbonized Egyptian papyri. It could also be applied to reused medieval parchments (palimpsests) or even wax tablets. The potential impact is enormous: it is estimated that only in the Villa of the Papyri there are between 1,500 and 2,000 scrolls, of which only about 600 have been unrolled, and most of them fragmentarily. The rest, about 300 intact scrolls and hundreds more in fragments, await reading.

The recovered text, an Epicurean treatise on pleasure and friendship, has already revealed ideas that challenge previous interpretations. For example, it contains passages discussing the relationship between pleasure and virtue, a central theme in Epicurus' philosophy but of which only indirect quotations were preserved. This could change our understanding of a school that profoundly influenced the Renaissance and the Enlightenment.

What Consequences Will It Have?

In the short term, the method is expected to be applied to the hundreds of unread Herculaneum papyri, as well as similar scrolls from other collections. Seales' team has already announced plans to scan all the scrolls at the National Library of Naples, which houses the majority of the collection. The Vesuvius Challenge has established new prizes to motivate the reading of complete scrolls, with rewards up to $700,000.

In the medium term, it could drive the development of specialized AI tools for paleography and epigraphy, facilitating the reading of worn or damaged inscriptions. Tech companies like Google or Meta could collaborate in creating language models trained on ancient texts, allowing the reconstruction of gaps or even automatic translation. It also raises ethical challenges: who should have access to these texts, and how will intellectual property rights be managed for works two thousand years old? Although the texts are in the public domain, the high-resolution images and algorithms used could be subject to patents or restrictive licenses.

For the general public, the discovery brings closer the possibility of accessing lost works by authors such as Epicurus, Sappho, or Aristotle, which could change our understanding of ancient thought. For example, Philodemus is known to have written extensively on music, poetry, and rhetoric, topics barely known. Even an unknown work by Aristotle could appear, whose lost dialogues are believed to have been in the library.

However, the process will not be quick. Each scroll requires weeks of scanning and months of AI processing. Additionally, carbon-based ink is difficult to detect in heavily carbonized papyri, and some scrolls are so damaged that they might be illegible even with this technique.

What Readers Should Know

• The method is non-invasive and can be applied to other carbonized scrolls, such as those from the Nag Hammadi library (Egypt) or the Derveni papyri (Macedonia).
• The AI was trained on micro-CT scans of fragments where the ink was visible, then generalized to closed scrolls. The algorithm detects minimal differences in material density, as the ink contains lead and other heavy elements.
• The recovered text is a treatise on pleasure and friendship, typical of the Epicurean school. It is provisionally titled On the Nature of Pleasure.
• The project is part of the Vesuvius Challenge, a competition offering prizes for advances in reading papyri. To date, over $1.5 million in prizes have been awarded.
• The researchers have already announced plans to read the rest of the library, with the goal of publishing all texts in open access by 2030.

“These carbonized scrolls look like dead books, but they are not. They are beginning to speak again.” — Brent Seales, quoted by Gizmodo.