In their TED talk, Youssef Nader and Julian Schilliger discuss how artificial intelligence is being used to decode ancient scrolls from the Villa dei Papiri, which were damaged during the eruption of Mount Vesuvius nearly 2,000 years ago. By leveraging CT scanning and advanced algorithms, their team successfully recovered over 2,000 letters and several columns of text, showcasing the potential of AI to unlock historical knowledge that has been inaccessible for centuries.
In the TED talk, Youssef Nader and Julian Schilliger discuss the innovative use of artificial intelligence (AI) to decode ancient scrolls that were buried during the eruption of Mount Vesuvius nearly 2,000 years ago. The scrolls, believed to contain the thoughts of the Greek philosopher Philodemus of Gadara, were discovered in the Villa dei Papiri but were severely damaged by the pyroclastic flow, making them difficult to read. Historically, attempts to unroll these scrolls physically resulted in their destruction, leaving only the most damaged ones intact.
The advent of modern technology, particularly CT scanning, provided a new avenue for researchers to explore these scrolls without causing further damage. Brent Seales, a professor at the University of Kentucky, pioneered the use of CT scans to create 3D models of the scrolls, allowing for the possibility of virtual unrolling. In 2022, the Vesuvius Challenge was launched to encourage teams worldwide to develop algorithms that could digitally unroll the scrolls and make the ink visible, which was a significant challenge due to the carbon-based ink used on carbon-based papyrus.
Julian Schilliger, a digital archaeologist, and Youssef Nader, an AI researcher, joined forces with other contestants to tackle the challenge. They focused on two main problems: unrolling the scrolls virtually and enhancing the visibility of the ink. Julian improved existing algorithms for virtual unrolling, achieving a remarkable 10,000-fold improvement in processing speed and accuracy. Meanwhile, Youssef worked on developing a machine-learning model to detect the faint ink on the scrolls, which was difficult to see against the papyrus background.
The team employed a novel approach to train their AI model by using intermediary goals and iterative learning. By continuously refining their predictions and creating new ground truths based on the model’s findings, they were able to enhance the model’s ability to identify letters and words within the scrolls. This iterative process led to the successful recovery of coherent text, with the team eventually decoding over 2,000 letters and several columns of text from the scrolls.
The talk concludes with a reflection on the potential of AI to unlock historical knowledge that has been inaccessible for centuries. The success of the Vesuvius Challenge demonstrates how technology can bridge the gap between the past and present, allowing researchers to access and understand ancient texts. Nader and Schilliger emphasize the importance of continued exploration and innovation in the field, as AI opens new windows into history and enhances our understanding of ancient civilizations.