Artificial intelligence has deciphered the charred papyrus scrolls that were burned during the eruption of Vesuvius.

For over 250 years, scientists have held a precious treasure - 1826 papyrus scrolls discovered in Herculaneum. This ancient Roman city, along with Pompeii, was buried under a layer of ash during the eruption of Vesuvius. The scrolls were found on the grounds of an opulent country villa, which historians believe belonged to Julius Caesar's father-in-law. It is considered the only completely preserved ancient library. It is thought to contain what were believed to be lost treatises by Aristotle, plays by Sophocles, Aeschylus, and Euripides, poems by Sappho, as well as unknown historical works by Titus Livius. The challenge is that the vast majority of the scrolls are so charred that they cannot be unrolled or read - they crumble in the hands of researchers. Over the years, numerous attempts have been made to unroll and read the manuscripts; however, only 150 texts have been deciphered.

Yet manuscripts, as is known, do not burn. With the advancement of computer technology, progress has been made. Professor Brent Seales from the University of Kentucky, an expert in machine learning, AI, and computer vision, has developed a method that allows the scrolls to be read without unrolling them. This is an incredibly complex procedure. First, the papyri are scanned using a synchrotron (a particle accelerator - Ed.), which accelerates electrons to nearly the speed of light. This is much more effective than conventional X-rays. Thanks to this, it is possible to discern objects at a scale of a few thousandths of a millimeter. Based on the obtained images, a 3D model of the scroll is reconstructed. Then, researchers need to separate one layer from another and unroll the digital image of the scroll. Finally, they must locate the traces of ink on the papyrus.

This method proved effective during the deciphering of a 1700-year-old parchment found in a burned synagogue in Ein Gedi, Israel (it was a fragment from the biblical book of Leviticus). In that case, the scribe used lead ink on parchment (calfskin) - a combination that is clearly visible in X-ray images.

However, most of the Herculaneum scrolls are written on papyrus (paper made from reeds) with ink made from charcoal or soot. Such ink is chemically indistinguishable from the papyrus, and X-ray scanning does not yield good results.

Thus, Seales and his colleagues decided to involve artificial intelligence specialists in solving the problem. With the help of sponsors from Silicon Valley, who raised about 1.5 million dollars, the Vesuvius Challenge project was launched. Brent Seales' team made thousands of high-resolution images taken with X-ray tomography publicly available. Anyone interested can try their hand at deciphering.

A breakthrough occurred last year when three students - Youssef Nader (Germany), Luke Farritor (USA), and Julian Schilliger (Switzerland) trained a neural network that identified over 2000 Greek letters from one of the scrolls.

It turned out to be notes from the Epicurean philosopher Philodemus. In the deciphered fragment, he reflected on the pleasures brought by music and food.

Although the students recognized only 5% of the text of the manuscript, the heroes were awarded $700,000. There is still much work and unresolved challenges ahead, but now researchers understand the direction in which to move.

- Today, we can confidently say for the first time that we are capable of reading the scrolls in their entirety, - said Stephen Parsons, the project manager of the Vesuvius Challenge.

Deciphering the remaining papyrus scrolls from Herculaneum will significantly enhance scholars' understanding of what occurred in antiquity. Researchers estimate that the volume of data held by this library will add approximately 20 percent to humanity's knowledge of the ancient period.