How AI and Particle Accelerators Are Reading the Unreadable Herculaneum Scrolls

When Mount Vesuvius erupted in 79 AD, it did not just bury the citizens of Pompeii and Herculaneum; it entombed their knowledge. In the luxury seaside town of Herculaneum, a massive pyroclastic surge engulfed a sprawling estate believed to belong to Julius Caesar's father-in-law. Inside the estate was a vast library of papyrus scrolls. The intense, oxygen-deprived heat of the volcanic ash instantly carbonized the scrolls, turning them into brittle lumps of charcoal before they could burn away.[4][5]

For nearly 1,700 years, the library sat undisturbed under 60 feet of volcanic rock. When workers digging a well in 1752 finally stumbled upon the "Villa of the Papyri," they recovered more than 800 of these blackened, log-like artifacts. But discovery quickly turned to frustration. The scrolls were so fragile that any physical attempt to unroll them caused the carbonized papyrus to shatter into dust. For over two and a half centuries, the only surviving intact library from the classical world remained entirely unreadable.[3][5]

Today, that ancient silence has been broken. A convergence of particle physics, 3D imaging, and artificial intelligence has achieved what generations of archaeologists deemed impossible: reading the scrolls without ever opening them. The breakthrough, catalyzed by a global crowdsourcing competition called the Vesuvius Challenge, has rapidly accelerated from deciphering a single word to extracting entire columns of lost philosophical texts.[1][7]

The foundation for this revolution was laid by Dr. Brent Seales, a computer scientist at the University of Kentucky. Seales pioneered a technique called "virtual unwrapping." By placing the charred scrolls inside high-resolution X-ray micro-computed tomography (micro-CT) scanners—and later, particle accelerators like the UK's Diamond Light Source—his team could map the internal 3D structure of the rolled papyrus. Software could then trace the spiraling layers and digitally flatten them onto a computer screen.[1][4]

But the Herculaneum scrolls presented a unique, seemingly insurmountable physics problem. Unlike medieval manuscripts written with metal-based inks that glow brightly on X-rays, the Roman scribes at Herculaneum used a carbon-based ink made of soot and water. On a CT scan, carbon ink sitting on carbonized papyrus is practically invisible. The 3D scans successfully unrolled the pages, but the pages appeared completely blank.[5][6]

To solve the ink problem, Seales partnered with Silicon Valley entrepreneurs Nat Friedman and Daniel Gross in 2023 to launch the Vesuvius Challenge. They released thousands of 3D X-ray images of the scrolls to the public, alongside open-source virtual unwrapping tools, and offered over $1 million in prize money to anyone who could write software capable of detecting the invisible ink.[1][6]

The critical insight came from a contestant who noticed a faint "crackle pattern" on the surface of the digital papyrus—microscopic textural changes left behind where the wet ink had originally dried and altered the papyrus fibers. Luke Farritor, a 21-year-old college student, trained a machine learning model to hunt for this specific crackle pattern across the vast 3D datasets. In late 2023, his neural network highlighted a cluster of shapes that formed the Greek word "πορφύρα" (porphyras), meaning purple.[5][6]

That single word shattered the dam. Farritor teamed up with two other researchers, Youssef Nader and Julian Schilliger, to refine the AI models. By early 2024, their combined algorithms successfully revealed 15 columns of text containing more than 2,000 characters, claiming the $700,000 grand prize. Classical scholars immediately identified the text as a previously unknown work by the Epicurean philosopher Philodemus, discussing how music, food, and art influence human pleasure.[1][3][5]

The pace of discovery has only accelerated since. In 2025, researchers successfully identified the title of another still-sealed scroll—"On Vices"—and generated the first internal images of a scroll housed at the Bodleian Libraries in Oxford. Within those Oxford scans, scholars quickly spotted the ancient Greek word "διατροπή" (diatrope), meaning disgust, proving that the AI pipeline could be applied to scrolls scattered across different museums globally.[2][4]

The mechanism driving these discoveries relies entirely on pattern recognition, not language comprehension. The machine learning models are trained strictly to detect the physical presence of ink based on voxel (3D pixel) density and texture. The AI has no understanding of Greek or Latin; it operates like a blind copyist, meticulously highlighting microscopic surface anomalies. It is then up to human papyrologists to look at the AI's output, translate the fragmented letters, and reconstruct the historical context.[2][7]

This separation of labor is viewed as a massive advantage by historians. Because the AI does not "know" what it is reading, it cannot hallucinate or guess missing words based on modern linguistic biases. It simply maps the physical reality of the carbonized artifact, ensuring that the recovered texts remain authentic primary sources.[2][7]

The challenge now is scaling the technology. Digitally tracing the tightly wound, crushed layers of papyrus—a process called segmentation—has historically required hundreds of hours of manual human labor per square centimeter. Open-source developers are currently building next-generation algorithms that automate this segmentation, allowing the AI to "see" the boundaries of the burned sheets and smooth out the digital folds instantly.[1][4]

If this automated pipeline can be perfected, the implications for classical history are staggering. The 800 scrolls currently sitting in museum vaults represent only a fraction of what may still be buried. Archaeologists believe the unexcavated lower levels of the Villa of the Papyri could hold thousands more scrolls, potentially containing lost works by Aristotle, Sophocles, or early Roman historians.[3][5]

For centuries, the eruption of Mount Vesuvius was viewed as one of history's great cultural tragedies. Yet, in a profound irony, the volcanic ash that destroyed Herculaneum is precisely what preserved its library from the damp rot of the Italian climate. Now, armed with particle accelerators and neural networks, researchers are finally unlocking a time capsule that has waited two millennia to be read.[5][7]