AI Just Read a 2,000-Year-Old Sealed Scroll

Volcanic fury struck in 79 AD, when Mount Vesuvius erupted, flash-carbonizing an entire Roman library in Herculaneum. This extraordinary, catastrophic event inadvertently preserved what remains the only intact library known to survive from antiquity, offering an unparalleled glimpse into the intellectual life of the classical world. Over 1,800 papyrus scrolls lay buried.

The eruption's intense heat, however, presented an intractable problem for scholars. It transformed the delicate papyrus scrolls into brittle, fragile charcoal logs, indistinguishable from the ash that entombed them. For nearly 2,000 years, any physical attempt to unroll these delicate artifacts caused them to crumble instantly into dust, rendering them utterly unreadable.

These carbonized cylinders represented a tantalizing, yet unreachable, treasure trove. Within them lay the potential for lost classical texts, including philosophical treatises and literary works by revered figures such as:

• Aristotle
• Livy
• Sappho

Scholars had long considered these ancient voices silenced forever, thought to be gone, locked away by time and tragedy. Deciphering them seemed impossible with conventional methods.

To peer inside these carbonized relics without destruction, scientists turned to cutting-edge particle accelerators. They performed high-resolution CT scans, generating incredibly detailed 3D digital maps of each scroll's complex, layered internal structure. This non-invasive method revealed the exact topography of the tightly wound papyrus, layer by delicate layer.

Despite this technological leap, a formidable obstacle remained. The ancient ink, derived from carbon, was virtually identical in composition to the charred papyrus itself. This made the text effectively invisible within the raw scan data, a problem that had stumped researchers for decades, rendering the detailed 3D maps frustratingly opaque.

While specialized software could then digitally flatten these intricate layers from the 3D map, creating a two-dimensional surface, this process alone was insufficient. Without a method to differentiate the carbon-based ink from its carbonized background, the virtual unrolling yielded only blank charcoal surfaces. The challenge had fundamentally shifted: no longer a physical problem of unrolling, it became a complex data science problem, demanding a novel approach to reveal the unseen words.

Particle accelerators provided the raw data, but true decipherment required artificial intelligence. Researchers developed AI models trained to discern the incredibly faint, subtle textural differences of the ancient ink. Unlike modern inks, the carbon-based Herculaneum ink did not absorb into the papyrus; instead, it sat slightly raised on the surface fibers, creating a microscopic topography. This minute elevation, invisible to the human eye, became detectable by specialized software.

This breakthrough accelerated dramatically with the launch of The Vesuvius Challenge (https://scrollprize.org/). Tech entrepreneurs gamified the monumental task, offering substantial cash prizes to coders and AI experts worldwide. This crowdsourcing initiative transformed a niche academic problem into a global race, fostering rapid innovation in machine learning algorithms.

Quickly, the challenge yielded results. A team of college students first identified individual characters, then the complete Greek word for 'purple'. This initial success soon escalated to the decipherment of extensive readable passages from Scroll 4 (PHerc. 1667), a Stoic philosophical treatise. This pivotal achievement confirmed that the entire Herculaneum library, once deemed lost to history, now lies within reach of virtual resurrection, opening an unparalleled window into ancient thought.

Researchers have finally achieved the impossible: reading an entire 2,000-year-old sealed scroll, PHerc. 1667, from end-to-end. This monumental effort unveiled a lost philosophical treatise on ethics, impulse, and human nature, possibly by Chrysippus or Philodemus. The success unequivocally validates this groundbreaking AI technology, demonstrating its capability to decipher ancient texts at scale.

With this proof of concept, the gates to a vast, lost library of knowledge now stand ajar. Over 1,800 Herculaneum papyri still await decipherment, representing the only intact library from antiquity. The Vesuvius Challenge actively targets these remaining scrolls, promising to rediscover an immense volume of classical literature previously thought forever lost.

This project is a landmark case, showcasing how AI can solve problems once considered physically impossible. It provides a powerful new toolkit for historians, archaeologists, and scientists, fundamentally altering how we engage with ancient artifacts. This is the ultimate example of code meeting history, bridging millennia with algorithms to illuminate our past.

What are the Herculaneum scrolls?

They are over 1,800 ancient Roman papyrus scrolls from a library in the town of Herculaneum. They were flash-heated and carbonized during the eruption of Mount Vesuvius in 79 AD, preserving them but making them too fragile to physically open.

How did AI read the burnt scrolls without opening them?

Researchers first used particle accelerator-powered CT scans to create a 3D digital map of the rolled-up scrolls. Then, they trained machine learning models to detect the subtle surface texture of the carbon-based ink, which is invisible to the naked eye, allowing them to virtually unroll the text in software.

What was written on the first fully deciphered scroll?

The first scroll read end-to-end, PHerc. 1667, contains a previously unknown Stoic philosophical treatise on ethics, impulse, and human nature. The text discusses pleasure and its role in morality, potentially attributable to Chrysippus or Philodemus.

What is the Vesuvius Challenge?

The Vesuvius Challenge is a global competition launched in 2023 with a prize pool to accelerate the deciphering of the Herculaneum scrolls. It crowdsourced the development of AI algorithms, leading to the first words being read in 2023 and the first full passages in 2024.