AI unlocks secrets of Vesuvius-scorched ancient scrolls after 2,000 years

What was thought lost for two thousand years is being found again
Researchers use AI to decipher carbonized Herculaneum scrolls buried by Mount Vesuvius in 79 AD.

Nearly two thousand years after Vesuvius silenced the library of Herculaneum, artificial intelligence has begun to give those carbonized scrolls their voice back. Scientists have trained machine learning algorithms to detect microscopic traces of ink on charred papyrus, recovering philosophical texts that no human eye could read unaided. The disaster that seemed to erase these words may, in a strange reversal of fate, have been the very condition of their survival — waiting for a technology that did not yet exist to come and listen.

  • For centuries, thousands of Herculaneum scrolls sat in collections — visible, tangible, and utterly silent, their ink fused into blackness by the 79 AD eruption.
  • The core tension: irreplaceable ancient knowledge locked inside fragile, carbonized documents that crumble under the very attempts to read them.
  • AI trained on high-resolution imaging now detects where ink altered papyrus at a microscopic level, reconstructing letters without physically unrolling or damaging the scrolls.
  • The first deciphered fragments have already surfaced a genuine 2,000-year-old philosophical treatise — not a later copy, but the actual words of a Roman thinker.
  • Thousands of scrolls in the Herculaneum archive remain unread, and researchers are now racing to scale and refine the technique before fragmentation claims more of what survives.

In 79 AD, the eruption of Mount Vesuvius buried the Roman town of Herculaneum along with thousands of papyrus scrolls — documents that carried the intellectual life of an ancient civilization. The heat carbonized them so completely that ink and surface became visually inseparable, leaving scholars able to see the scrolls but unable to read them. For nearly two millennia, those texts waited in silence.

Artificial intelligence has now broken that silence. Machine learning algorithms, trained to detect subtle variations in surface texture and composition, can analyze high-resolution imaging data and identify where ink once altered the papyrus at a microscopic level — reconstructing letters without further unrolling or damaging the fragile documents. It is, in effect, teaching machines to perceive what human vision cannot.

The results have already proven remarkable. Researchers have deciphered fragments of an authentic philosophical treatise — not a later transcription, but the direct words of someone who lived two thousand years ago, engaging with questions of ethics, knowledge, and existence. The eruption that appeared to destroy these texts may instead have preserved them in a form only modern technology could recover.

The Herculaneum collection holds thousands of documents, most still unread. Scientists are now working to refine and scale the technique, though challenges remain: fragmentation, uneven carbonization, and scrolls fused together. Yet the foundational assumption — that these texts were simply beyond reach — has been overturned. What seemed lost for two thousand years is being recovered, one letter at a time.

In 79 AD, Mount Vesuvius erupted with such violence that it buried the Roman towns of Pompeii and Herculaneum under ash and pumice, freezing them in time. Among the casualties of that catastrophe were thousands of papyrus scrolls—documents that would have offered direct windows into how educated Romans thought, what they read, what they argued about. For nearly two thousand years, those scrolls remained locked away, their surfaces charred black and brittle, their contents seemingly lost to the heat that had carbonized them into illegibility.

Archaeologists have long known the scrolls existed. They could see them, handle them carefully, even unroll some of them. But reading what was written on them proved nearly impossible. The heat had darkened the papyrus so thoroughly that ink and surface became visually indistinguishable. Scholars could make out fragments here and there, but whole passages remained opaque—literally and figuratively. The philosophical treatises, letters, and administrative records trapped inside those blackened rolls seemed destined to remain silent.

Then artificial intelligence changed the equation. Using machine learning algorithms trained to detect subtle variations in surface texture and composition, scientists have now begun to successfully read text from the Herculaneum scrolls without unrolling them further or risking additional damage. The technology works by analyzing high-resolution imaging data, identifying where ink has altered the papyrus at a microscopic level, and reconstructing the letters beneath the char. It is, in essence, teaching computers to see what human eyes cannot.

The breakthrough has already yielded results. Researchers have deciphered fragments of a philosophical text—a treatise from the ancient world that offers rare, direct evidence of how Romans engaged with ideas about ethics, knowledge, and the nature of existence. These are not secondhand accounts written centuries later by other authors. These are the actual words of someone who lived two millennia ago, preserved by the very disaster that seemed to have destroyed them.

What makes this achievement significant extends beyond the immediate content of the scrolls themselves. The Herculaneum collection contains thousands of documents. Most remain unread. If this AI-assisted decipherment technique proves reliable and scalable, it could unlock an entire archive of lost voices—administrative records that illuminate how Roman society functioned, personal correspondence that reveals how people actually lived, philosophical arguments that show what educated Romans debated in their studies. The eruption of Vesuvius, which seemed to erase these texts from history, may instead have preserved them in a form that only modern technology could recover.

Scientists are now working to refine the process, testing it on additional scrolls and developing methods to read larger sections of text more efficiently. The challenge remains substantial: many scrolls are fragmented, some are stuck together, and the carbonization is not uniform across all documents. But the fundamental barrier—the assumption that these texts were simply unreadable—has been broken. What was thought lost for two thousand years is being found again, one letter at a time, by machines trained to see through the darkness.

Finally able to read them
— Scientists quoted in reporting on the breakthrough
A Conversa do Hearth Outra perspectiva sobre a história
Inventor

Why does it matter that we can read these particular scrolls now? We already know a lot about ancient Rome from other sources.

Model

We know what later writers told us about Rome. These scrolls are the actual thinking of people who lived there—unfiltered, not rewritten by someone else's hand centuries later. That's a different kind of knowledge entirely.

Inventor

But they're just fragments, right? Pieces of text?

Model

Yes, but even fragments can be revelatory. A single sentence from a philosophical argument, or a name in a ledger, can reshape how we understand what mattered to people then.

Inventor

How does the AI actually read something that's been burned black?

Model

It detects the microscopic ways ink changes the surface of papyrus—density, texture, chemical composition. The human eye can't see those differences. A machine trained on thousands of examples can.

Inventor

So it's not magic. It's pattern recognition.

Model

Exactly. Which means it will get better. The more scrolls we feed it, the more accurate it becomes.

Inventor

What happens if the AI misreads something?

Model

That's the real work ahead—verification. But even if it's imperfect, it's infinitely better than the silence we had before.

Quer a matéria completa? Leia o original em Google News ↗
Fale Conosco FAQ