From an Idea to Rewriting Ancient History in Under a Year

🔷 Subscribe to get breakdowns of the most important developments in AI in your inbox every morning.

(This post was guest-written by @sporadicalia)

This week, former Github CEO, Nat Friedman announced “The scrolls have been read”, as the Vesuvius Challenge - an ML & computer vision contest he’s sponsored, solved a massive archaeological mystery.

So what really happened?

"What was really achieved? And why is it such a massive deal?" - you may be wondering.

It starts in 79 AD. Mount Vesuvius erupted, burying the nearby cities - most notably, Pompeii.

But another town, Herculaneum was also buried; and with it, a treasure trove of ancient papyri.

This now ancient library of papyrus scrolls was completely buried and lost to time -- that is, until 1709 when Herculaneum was accidentally rediscovered.

And in 1750, a tunnel reached the ruins of an enormous seaside mansion now known as the Villa of the Papyri.

The Villa was likely built by one Lucius Calpurnius Piso - the father-in-law of Julius Caesar. And it was found to contain "lumps of what appeared to be charcoal. It was gradually realized that these were papyrus scrolls, carbonized by the heat of the eruption."

You see, the scrolls are carbonized by the heat of the volcanic debris and ash, but in a stroke of historical luck, their contents are also preserved.

Whereas virtually every ancient text exposed to the air decays and disappears, the library of the Villa of the Papyri remains.

There's just one problem - they're nearly impossible to unroll and read. Early attempts resulted in a few unrolled scrolls but also unfortunately destroyed many of them.

So how do you read them? Let me introduce you to Dr. Brent Seales.

In 2015, using X-ray tomography and computer vision, a team led by Seales at the University of Kentucky pioneered the technique of "virtual unwrapping" to read one of the Dead Sea scrolls from Israel.

Inspired by the research, Nat Friedman reached out. Progress in unwrapping the Herculaneum Papyri was slow--Seales had revealed bits and pieces, but they were mostly unreadable.

But Friedman had an idea. In March of 2023, the Vesuvius Challenge was launched, offering $1 million in prizes to anyone who could deliver a breakthrough.

Just 10 months later, the Vesuvius Challenge has yielded results beyond what anyone would have expected so soon.

So let's quickly take a few tweets to explain what exactly is going on in "unraveling" the Herculaneum Papyri and why it had proved so difficult - until now.

It starts with a scan: Each scroll is placed within a particle accelerator and is peppered with high-energy X-rays, resulting in high-resolution CT scans that are carved into tiny slices and ultimately, combined into a 3D volume reconstruction.

Once scanned, the next step is identifying and capturing segments from the tightly-rolled layers, as well as flattening the 3D volume in and around the segment. These methods allow researchers to potentially capture ink that lays on top of or has seeped into the papyrus.

Enter the brilliant minds who took on the challenge--and eventually earned the Grand Prize of $700,000 by meeting the criteria of "4 passages of 140 characters each, with at least 85% of characters recoverable."

Congrats to Luke Farritor, Youssef Nader, and Julian Schilliger!

Luke Farritor is a 21-year-old college student and SpaceX intern from Nebraska. He received the first-place First Letters Prize in late 2023, as the "first person in history to read an entire word from the inside of a Herculaneum scroll."

Farritor's work was a massive win for the challenge, convincing everyone that the Grand Prize was "definitely achievable."

Then in October, Youssef Nader, an Egyptian PhD student in Berlin, was able to read a few columns of text, winning the second-place First Letters Prize.

His results were "particularly clear and readable, which made him the natural lead for the team that formed."

With the help of Julian Schilliger, a student at ETH Zurich who figured out ways to segment the text on the scrolls, winning three Segmentation Tooling subprizes and enabling the 3D-mapping of the papyrus areas - a super team was formed.

That brings us to this week. 275 years after the discovery of the Herculaneum Papyri, we can finally begin to read the scrolls!

The submission is frankly incredible - more than a dozen columns of text with entire paragraphs visible.

Scholars are already busy translating the ancient text - which comprises just 5% of the scroll it was pulled from - but it's the larger-than-life possibilities for what comes next that have historians and technologists alike incredibly hopeful.

Multiple scanned scrolls remain to be segmented and read. Meanwhile, hundreds, maybe thousands, of preserved scrolls remain buried in the Villa of the Papyri.

It's a treasure trove of ancient documents, the likes of which simply don't exist anywhere else. So what comes next?

The Vesuvius Challenge and Nat Friedman have already laid out "Stage Two" of the Master Plan:

• Use a particle accelerator to scan the rest of the scrolls

• $40k per scroll x 800 scrolls = $30 million cost

Scanning and segmentation at scale are still a major challenge. Even with proven techniques, the task of scaling up to scan, segment, and read the "800 extant scrolls" remains.

To do so, however, requires more money (as all endeavors do), of course!

And so, if you're a wealthy donor who wants to support and make possible some of the greatest historical and technical research out there today -- well now's your chance!

It's also important to recognize everyone who made the Vesuvius Challenge possible in the first place - from Nat Friedman himself, but also fellow donors like Alex Gerko, Daniel Gross, and Joseph Jacks.

Such groundbreaking research would not have been possible without them.

Become a subscriber for daily breakdowns of what’s happening in the AI world: