Ancient Herculaneum Scrolls Deciphered Using AI Technology

A groundbreaking advancement in the study of ancient texts has been achieved by a team of young scientists. Through the application of artificial intelligence, they have succeeded in interpreting content from the Herculaneum scrolls—a collection that has long been considered unreadable due to damage sustained during the Mount Vesuvius eruption nearly 2,000 years ago.

Victory in the Vesuvius Challenge

The trio, consisting of Youssef Nader, Luke Farritor, and Julian Schilliger, were awarded the grand prize for successfully reading four segments of the Herculaneum papyrus scrolls without physically unrolling them, thereby preserving their delicate structure. This achievement was possible thanks to their innovative use of machine learning algorithms—a subset of artificial intelligence. Their collective effort resulted in the discovery of texts potentially authored by Philodemus—a philosopher thought to have resided at the villa where these scrolls were originally found. These texts discuss the concept of leading a good life through the enjoyment of beauty, music, and gastronomy.

AI and the Unwrapping of History

The process to 'unwrap' the scrolls electronically began with high-resolution computed tomography (CT) scans performed at the UK's Diamond Light Source particle accelerator. This produced detailed 3D representations of the scrolls' structure. From here, the team used segmentation techniques to digitally unfurl the layers of papyrus. Following this, machine learning algorithms were employed to identify and reconstruct ink marks on the papyrus, allowing the ancient writings to once again be read and understood without the need for optical character recognition software traditionally used in text analysis.

Decades of Dedication Pay Off

The effort to decipher these scrolls is not a new endeavor. For decades, Brent Seales from the University of Kentucky has researched ways to read the damaged scrolls, with CT scans being one of the pioneering methods. The Vesuvius Challenge, launched in 2023, marked a significant acceleration in this research area. The competition saw rapid advancements, including the identification of distinctive 'crackle' patterns in the text by Casey Handmer, which led to the deciphering of the first full Greek word 'ΠΟΡΦΥΡΑϹ' (purple) through AI machine learning models. With further developments on the horizon, the next challenge is set to focus on reading an entire scroll by the end of 2024.

ancient, AI, discovery