AI Discovers Author of 2,000-Year-Old Charred Scroll from Vesuvius

For the first time, researchers have successfully identified both the author and title of a document that has been encased within a charred scroll for nearly 2,000 years, achieving this remarkable feat without having to peel back a single layer of the scroll itself.

This extraordinary scroll, known as PHerc. 172, was unearthed from the ashes of Herculaneum, an ancient Roman town that was tragically buried under the volcanic ash and debris resulting from the catastrophic eruption of Mount Vesuvius in 79 CE. It is one of only three surviving Herculaneum scrolls that are currently housed at Oxford’s Bodleian Libraries.

Utilizing advanced high-resolution scans and innovative machine learning techniques, scholars have been able to virtually “unwrap” the papyrus and discern the name within it: On Vices, authored by the Epicurean philosopher Philodemus.

The complete title of this treatise is On Vices and Their Opposite Virtues and In Whom They Are and About What, as reported by Fine Books Magazine. This ancient work can be likened to a form of self-help literature, as it delves into the principles of living a virtuous life through the avoidance of vice. Composed in the first century BCE, this text is now being read for the first time since it was entombed in the devastating volcanic eruption nearly two millennia ago.

This remarkable discovery, validated by multiple research teams, has earned the project’s collaborators the prestigious $60,000 First Title Prize from the Vesuvius Challenge, an open-science competition that aims to make ancient texts accessible through the power of AI.

In recent years, artificial intelligence has played a pivotal role in deciphering the ancient, carbonized scrolls originating from Herculaneum, the Roman town that was engulfed by the eruption of Mount Vesuvius in 79 CE. These invaluable scrolls, first discovered in the 18th century within what is now referred to as the Villa of the Papyri, represent one of the few surviving libraries from the classical world.

Due to their extremely fragile and charred state, conventional manual methods of unrolling these scrolls often resulted in their destruction. However, researchers are now harnessing the power of advanced imaging techniques and machine learning to read these ancient texts without the need to physically open them, preserving their integrity.

A significant breakthrough occurred in 2015, when scientists successfully employed X-ray tomography to read a different ancient scroll from En-Gedi, producing a 3D scan that could be virtually “unwrapped.” Building on this foundational work, researchers at the University of Kentucky created the Volume Cartographer, a sophisticated program that utilizes micro-CT imaging to detect faint traces of carbon-based ink on the scrolls.

Given that the ink used does not contain metal—a common component of many ancient writing materials— a neural network had to be meticulously trained to identify subtle patterns that indicate the presence of ink on the carbonized papyrus. In 2019, researchers successfully showcased this innovative technique, paving the way for broader applications in the field.

These groundbreaking advancements culminated in the Vesuvius Challenge, which was launched in 2023 with the aim of crowdsourcing the decoding of unopened scrolls. Participants engage with AI tools—especially convolutional neural networks and transformer models—to identify and reconstruct the text embedded within the scrolls. In October 2023, the first word (“purple”) was successfully read from an unopened scroll, earning a $40,000 prize. The challenge is ongoing, providing further opportunities to decipher additional texts and enhance the underlying technology.

Brent Seales, a computer scientist affiliated with the University of Kentucky and a co-founder of the Vesuvius Challenge, remarked to The Guardian that the current bottleneck for the team lies in the processes of cleaning, organizing, and enhancing the scanning data to enable researchers to effectively interpret the carbonized ink as coherent text.

Crucially, the process of digitally unwrapping these ancient texts is driven by human expertise. While AI identifies probable areas of ink on these historical documents, it is the scholars who analyze the patterns to ascertain whether they constitute meaningful words or phrases. The overarching objective is not only to recover these lost philosophical texts, many of which could be attributed to Epicurus or his disciples, but also to create a scalable system for digitizing and decoding ancient literature—ultimately revolutionizing our comprehension of the classical world.