The Intersection of AI and Ancient History: Deciphering the Herculaneum Scrolls

A Historical Breakthrough via a Photo

In October 2023, a remarkable email landed on the phone of Federica Nicolardi, a renowned papyrologist at the University of Naples. The message included an image that would transform her research and open new avenues for the field of papyrology. The image depicted a fragment of a papyrus scroll, carbonized in the devastating eruption of Mount Vesuvius in A.D. 79.

This scroll was among hundreds unearthed from a luxury Roman villa in Herculaneum, near Pompeii, over two centuries ago. Previous attempts to unveil the scrolls, painstakingly peeling apart their brittle layers, had left much of the text forever unreadable. However, Nicolardi and a team of researchers have embarked on a groundbreaking project called the Vesuvius Challenge, utilizing cutting-edge artificial intelligence (AI) to unlock these ancient texts.

The Role of AI in Ancient Text Recovery

Technological advancements in AI have already revolutionized many sectors, including banking and medical research. Now, they are venturing into the realm of ancient texts. Deep learning, particularly through neural networks, is making it possible to read an entire segment of the Herculaneum scroll that had been inaccessible for nearly two millennia.

Nicolardi expressed her amazement: “In that moment, you really think ‘now I’m living something that will be a historical moment for my field.’” The implications of this discovery extend beyond the single text; it represents a new methodology for exploring ancient civilizations.

Current Innovations in Text Reconstruction

Traditionally, computers have assisted in categorizing and analyzing texts, but recent advancements are particularly exciting. Current research employs deep learning models, notably convolutional neural networks (CNNs) and recurrent neural networks (RNNs), to interpret ancient writings from different cultures.

For instance, Chinese researchers have used convolutional neural networks to reconstruct eroded Oracle Bone Script. Simultaneously, RNNs have been employed in Babylonian texts to suggest missing characters from historical records. These neural networks can analyze vast amounts of data more efficiently than human scholars, and they provide novel insights that were previously unattainable.

The Pythia and Ithaca Models: Bridging the Gap

In a collaborative effort at the University of Oxford, the Pythia model was developed to help decipher Greek inscriptions from Sicily. This project, initiated in 2017, demonstrated that AI could propose missing words effectively when trained on historical data.

Building on that success, researchers introduced the Ithaca model, which utilizes transformer technology – a more advanced AI framework. Ithaca not only restores gaps in texts but also provides estimates on their geographical origin. Its accuracy surpasses that of human experts when used jointly, demonstrating that collaboration between AI and specialists yields superior results.

Tackling Korea’s Vast Historical Archives

As AI helps revive ancient texts in the West, it is equally assisting researchers in South Korea facing the daunting task of translating extensive archives documenting the reigns of 27 Korean kings from the fourteenth to the early twentieth centuries. The project employs a unique blend of historical context and modern machine-learning techniques to translate Hanja, an ancient writing system, into contemporary Korean.

These advances showcase the global implications of AI in understanding and preserving ancient literary heritage. Scholars are increasingly recognizing AI’s potential to decode the past, despite the considerable volume and complexity of these historical records.

Decoding Lost Languages

On the other end of the spectrum, researchers are using neural networks to decipher ancient languages with minimal existing text. A recent study highlighted an effort to restore text from Mycenaean tablets using RNNs. While these texts contain few materials for training AI models, remarkable results have been achieved using innovative techniques, including ‘transfer learning’ that borrows insights from different but related datasets.

This confirms AI’s flexibility and adaptability in tackling both massive archives and sparse texts. By finding patterns across different languages, AI broadens the scope of possible inquiries into long-forgotten cultures.

The Vesuvius Challenge: Unmasking the Herculaneum Scrolls

One of the most compelling aspects of this AI revolution is encapsulated in the Vesuvius Challenge. Led by Brent Seales and his team at the University of Kentucky, participants aim to use advanced imaging and machine learning to read scrolls that are otherwise inaccessible.

The first time Seales successfully applied ‘virtual unwrapping’ technology, he uncovered text from a charred scroll dating back to the third century A.D. Now, with support from tech innovators, the challenge has gained momentum, attracting over 1,000 teams.

A landmark achievement occurred in February 2024, when a team won $700,000 for uncovering 16 columns of Greek text from previously unreadable scrolls. These texts are believed to have originated from a long-lost work of Epicurean philosophy, a revelation that could shift our understanding of ancient thought and culture.

Future Prospects of Ancient Text Discovery

The Vesuvius Challenge is just scratching the surface. According to Seales, there are hundreds of additional scrolls waiting to be analyzed in collections worldwide. The ongoing research shows promise for retrieving texts that hold potential insights into ancient Roman life, philosophy, and beyond. Seales envisions a future where AI-driven unwrapping techniques will significantly expedite the reading process, unlocking an “invisible library” of historical manuscripts hidden within other sealed artifacts, like Egyptian mummy wrappings and medieval book bindings.

Key Takeaways

The integration of AI in the study of ancient texts represents a monumental leap forward in the field of humanities. The advancements are not only enabling scholars to access texts thought lost forever but also transforming the types of questions we are able to ask about our past. As research progresses and machine learning methodologies continue to evolve, we stand on the brink of uncovering a vast wellspring of knowledge about ancient civilizations, promising to reshape our understanding of history in unexpected ways.