How Did AI Read Ancient Scrolls That Were Sealed for 2,000 Years?

Artificial intelligence successfully read 2,000-year-old carbonized scrolls from Herculaneum by using machine learning to detect microscopic texture differences in CT scan data, revealing the first ancient text recovered without physically opening the scrolls. The breakthrough came through the Vesuvius Challenge, where three students won $700,000 for deciphering over 2,000 characters of previously unknown ancient Greek philosophy.

The Only Ancient Library to Survive Complete

When Mount Vesuvius erupted in 79 CE, it destroyed the cities of Pompeii and Herculaneum, but it also preserved something extraordinary: the only complete ancient library known to exist. The Villa of the Papyri in Herculaneum contained over 1,800 papyrus scrolls, representing the personal collection of a wealthy Roman patron who gathered the finest philosophical texts of his era.

Unlike other archaeological sites where manuscripts decomposed over millennia, Vesuvius’s superheated gases instantly carbonized these scrolls, transforming them into fragile tubes of compressed charcoal. This carbonization process preserved the scrolls’ physical structure while making them impossible to open without destroying them completely.

The 275-Year Problem That Stumped Scholars

When excavators rediscovered the scrolls in 1752 under orders from King Charles VII of Naples, scholars faced an impossible dilemma. These were genuine ancient texts—potentially containing lost works of philosophy, poetry, and history—but every attempt to physically unroll them resulted in the scrolls crumbling to dust.

For over two centuries, hundreds of carbonized scrolls sat in museums, sealed and unreadable. Scholars could only speculate about their contents based on external analysis, believing the collection focused heavily on Epicurean philosophy. The scrolls represented a tantalizing glimpse of lost knowledge from the ancient world, forever out of reach.

Virtual Unwrapping: Reading Without Opening

The breakthrough began with Brent Seales, a computer science professor at the University of Kentucky, who spent decades developing a technique called virtual unwrapping. Using high-resolution CT scans, researchers could map the interior structure of rolled scrolls in three-dimensional detail, then computationally flatten them as if they had been physically unrolled.

The CT scans used in the Vesuvius Challenge were conducted at facilities including the Diamond Light Source synchrotron in the United Kingdom, generating X-rays ten billion times brighter than the sun. These scans achieved a resolution of approximately eight micrometers per voxel—twenty times finer than the width of a human hair.

The Invisible Ink Challenge

Despite the extraordinary resolution of the CT scans, researchers faced a critical problem: the ancient ink used in Herculaneum was carbon-based, meaning it had nearly identical density to the carbonized papyrus. To the X-ray scanners, the ink was essentially invisible against the charcoal background.

This seemingly insurmountable obstacle led to the creation of the Vesuvius Challenge in 2023—an open competition offering $700,000 to anyone who could use artificial intelligence to detect the subtle physical differences between inked and non-inked areas of the papyrus.

The AI Breakthrough and Historic Discovery

Hundreds of researchers, students, and programmers from around the world began analyzing the publicly released CT scan data. The winning approach used machine learning to identify microscopic texture differences in how the papyrus crinkled where ink had been pressed into it thousands of years ago.

In October 2023, 21-year-old Luke Farritor became the first person to identify a word within a sealed Herculaneum scroll. The word was “porphyras”—Greek for “purple.” This single discovery proved that AI could read what human eyes could not see, earning Farritor a $40,000 milestone prize.

The Winning Team and Their Revolutionary Results

Three competitors ultimately formed a winning team: Luke Farritor from Nebraska, Youssef Nader from Berlin, and Julian Schilliger from Zurich. On February 5, 2024, their combined submission won the $700,000 grand prize by revealing over 2,000 characters of ancient Greek text—approximately five percent of a single scroll’s content.

The recovered text appears to be a previously unknown work by Philodemus of Gadara, an Epicurean philosopher from the first century BCE. The passage discusses the nature of pleasure and whether physical pleasure or more refined forms represent the greatest good in human life—a philosophical debate that had been frozen in time for nearly two millennia.

What This Means for Ancient Knowledge

The implications of this breakthrough extend far beyond a single scroll. Approximately 800 intact scrolls from Herculaneum remain sealed, potentially containing lost works of philosophy, history, and literature from the ancient world. The Herculaneum library was curated by someone with extraordinary resources and taste, suggesting it may hold unknown masterpieces from antiquity.

The success of the Vesuvius Challenge has sparked continued research and funding for AI development to read more scrolls. What began as one researcher’s decades-long pursuit has become a global effort to recover knowledge that humanity had accepted as lost forever, demonstrating how modern technology can unlock secrets from our ancient past.