Welcome back, AI prodigies!

In today’s Sunday Special:

  • 📜 The Prelude

  • 🏺 The Ashes of a Roman City

  • 📠 How to Virtually Unwrap Scrolls

  • 🪨 Can AI Read Sealed Scrolls?

  • 🔑 Key Takeaway

Read Time: 7 minutes

🎓Key Terms

  • Deep Learning (DL): Mimics the structure and function of the human brain by processing data through multiple layers of artificial neurons.

  • Micro-Computed Tomography (Micro-CT): This 3D X-ray imaging technique safely reveals the internal structures of delicate artifacts.

🩺 PULSE CHECK

Login or Subscribe to participate

📜 THE PRELUDE

An Italian archaeologist kneels at the bottom of an excavation site, gently brushing away the packed layers of soil sheltering centuries of forgotten history. Suddenly, her headlamp flickers across something unusual among the fragments of pottery. She steadies her breathing, switching from a large paintbrush to a small pick, teasing away the stubborn dirt.

The headlamp steadies on this blackened cylinder as it emerges from burial, waiting to tell a story that’s slept beneath the rubble for generations. At first glance, it looks like a charcoal rock with a smooth, symmetrical formation carved by water, fire, and air. But it isn’t a charcoal rock. It’s a 2,000-year-old papyrus scroll, charred and carbonized by volcanic ash.

For millennia, archaeologists faced an inextricable dilemma: read or preserve? In other words, careful mechanical unrolling of the burnt and brittle 2,000-year-old papyrus scrolls always risked damaging their fragile form. Today, AI can read these 2,000-year-old papyrus scrolls before they’re even unrolled. So, how can computers see trapped ink that’s tightly packed on blackened papyrus paper?

🏺 THE ASHES OF A ROMAN CITY

⦿ 1️⃣ The Lost Library of Herculaneum

In 79 A.D., Mount Vesuvius erupted, sending superheated avalanches of gas, ash, and rock, known as pyroclastic flows, across the ancient Roman town of Herculaneum. Just below the volcano sat the Villa of the Papyri, a seaside retreat that housed thousands of books written on papyrus scrolls. The volcanic mud compressed, burned, and fused these papyrus scrolls into brittle, black cylinders.

In 1752, King Charles VII of Naples commissioned a team of excavators to explore the villa. At first, they found 21 papyrus scrolls scattered around the reception hall. Over the following decades, archaeologists recovered more than 1,700 papyrus scrolls from the villa, known today as Herculaneum Papyri.

⦿ 2️⃣ Early Unwrapping Attempts

One of the earliest systematic efforts came from Roman artist Camillo Paderni, employed by King Charles VII of Naples as an illustrator, excavator, and curator at the Museum Herculanense within the royal Palace of Portici. He sliced hundreds of the papyrus scrolls from end to end, copying any visible ancient text before the fragments disintegrated. This technique preserved bits of charred writing, but also destroyed most of the papyri.

In 1756, Italian priest and scholar Antonio Piaggio invented the Unrolling Machine (UM),” which leveraged a delicate framework of weights and pulleys to gradually separate the fused, charred layers at a rate of about 1 mm per day. For context, the first papyrus scroll required 40 years to fully unroll and revealed just snippets of ancient text with smudged letters resembling faint traces of Greek phrases.

By the 1950s, about 585 papyrus scrolls had been fully unrolled. Of those, just around 150 had been successfully deciphered. Most of the ancient texts belonged to the Epicurean poet and philosopher Philodemus of Gadara, who discussed the fear of death, the nature of the gods, and the cultural role of literature. He famously wrote about love’s reciprocity: “I know how to love him who loves me, and again I know right well how to bite him who bites me.”

📠 HOW TO VIRTUALLY UNWRAP SCROLLS

⦿ 3️⃣ Making a 3D Projection

In 2009, American computer scientist Brent Seales performed the first high-resolution Micro-CT scans of intact Herculaneum scrolls at the Institut de France in Paris. This method works by taking thousands of 2D X-ray sample images from multiple angles around the 2,000-year-old papyrus scrolls, capturing differences in ink density within the carbonized layers. These differences in density are reconstructed into 3D projections composed of millions of tiny cubes called Voxels, each as wide as a grain of sand.

It’s essentially a volumetric map of the papyrus scroll’s internal structure, but it’s still “rolled up.” In 2016, he developed a Volume Cartographer: a specialized computer software toolkit and set of cross-platform C++ libraries designed to virtually peel open these Voxels, exposing faint ink that’s full of cracks, voids, and warps.

🪨 CAN AI READ SEALED SCROLLS?

⦿ 4️⃣ How AI Detects Ink

In 2023, former GitHub CEO Nat Friedman launched the Vesuvius Challenge, awarding up to $1,500,000 in prizes to virtually unwrap the 2,000-year-old papyrus scrolls.

In 2024, 21-year-old American software engineer Luke Farritor won $700,000 for successfully deciphering more than 2,000 Greek letters, or just 5% of the papyrus scroll’s total text, which explored the dynamics between pleasure and abundance.

⦿ 5️⃣ Here’s How He Did It
  1. 🔴 Encoding Patches:

    • Once a papyrus scroll has been virtually unwrapped, it’s converted into a large grayscale image that’s divided into small pixel patches by a DL Model. For example, a large 512 × 512 grayscale image might be divided into 16 × 16 pixel tiles, producing 1,024 small pixel patches.

    • Each pixel patch is converted into a vector embedding: a list of numbers that enable computers to interpret microscopic thickness variations, crack textures, and fiber distortions to detect where ink once sat on the papyrus scroll. For example, Pixel Patch #1 is represented by numbers like: “{0.31, -0.84, 0.52, 0.07, -0.19....}.”

  2. 🟠 Analyzing Patches:

    • The DL Model analyzes two structural patterns across each pixel patch:

      1. 📍 Spatial Attention: Compares nearby pixel patches to see if they form curved strokes that resemble parts of Greek letters.

      2. 📌 Temporal Attention: Examines multiple layers of papyrus pages within the scrolls to find density changes that reveal compressed ink.

  3. 🟡 Mapping Patches:

    • After processing everything, the DL Model generates a probability score between 0 and 1 for the likelihood of ink within each pixel patch. For example, “{0.92}” means a “92% chance” of ink.

    • These probability scores generate a heat map. When clusters of high-probability scores align across neighboring pixel patches, they form curved strokes that outline Greek letters.

🔑 KEY TAKEAWAY

For centuries on end, these 2,000-year-old papyrus scrolls trapped the treasured voices of Roman poets, politicians, and philosophers. Trying to unroll them meant tearing through centuries of charred wisdom. With AI, even the tiniest ink particles are reborn as readable Greek letters to reveal the lost secrets of the Roman Empire without even touching the brittle carbonized papyrus.

📒FINAL NOTE

FEEDBACK

How would you rate today’s email?

It helps us improve the content for you!

Login or Subscribe to participate

❤️TAIP Review of The Week

“I gotta say, y’all actually make this AI stuff make sense. The examples help a lot.”

-Ethan (1️⃣ 👍Nailed it!)
REFER & EARN

🎉Your Friends Learn, You Earn!

{{rp_personalized_text}}

Share your unique referral link: {{rp_refer_url}}