In the year 79 CE, Mount Vesuvius in  Italy erupted, destroying and burying the Roman towns of Pompeii and Herculaneum in  a violent inferno of fire, ash and rock.

This was obviously bad news for the  thousands of people who died that day, but strangely enough, it may have  helped to preserve a library of scrolls containing ancient knowledge  otherwise lost to time. See, although those scrolls were burned to  a crisp in the conflagration, for the first time, scientists have used X-ray scans and  machine learning to begin to read them. [♪ INTRO] Almost everything we know about  the ancient Greek and Roman world comes from only a tiny fraction  of their written literature.

Books and scrolls are fragile,  and stuff just gets lost. But in 1750 CE, a farmer found what  turned out to be a buried library with thousands of scrolls,  in a villa near Herculaneum. The villa belonged to an important Roman  statesman, possibly even a relation of Julius Caesar, who liked to show off their  wealth and taste with big collections.

The more things change. The scrolls may contain works  no modern person has ever seen, which would hugely expand our  knowledge of the classical world. But they were all, y know, destroyed by a volcano.

So, end of story, right? Well, it turns out that the eruption  may have helped to preserve the scrolls, in an odd, roundabout, haphazard sort of way. See, rather than being reduced to ash, these scrolls were buried,  blackened, and only mostly destroyed.

And mostly destroyed is still slightly intact! They were just waiting for someone  to work out how to read them! Then, in 2015, scientists trying to decipher  a similarly-charred ancient document came up with a way to scan burned  parchments without actually touching them.

I mean, we cannot stress enough  how brittle these things are. People have tried to unroll  the Herculaneum scrolls and it went exactly as you'd think it would. The idea was to use X-rays, since the  beams pass through most materials, and different materials show up  differently in the resulting images.

In particular, it was a kind  of scan called a CT scan. That's the same medical imaging  trick that lets doctors build up a detailed, 3D, layer-by-layer  picture of something inside your body. Likewise, the researchers made complete 3D scans of the scrunched-up parchment.

From that, they could use clever  computer algorithms to work out how the pages were crumpled up in the 3D bulk, allowing them to digitally unfold the  parchment and read the original text, which turned out to contain  passages from the Hebrew Bible. The same trick should have worked  with the Herculaneum scrolls, but here, there was a problem: both  their ink and their writing surface are made of carbon, which  all look the same to X-rays. So in 2023, a group, which included  one of the Bible parchment scientists, launched a cash prize called the Vesuvius  Challenge, to get volunteer scientists to help decipher scans they  created of the Herculaneum scrolls.

The competition organizers carefully  designed the rules to ensure success, like by splitting the contest into multiple  smaller goals, each with smaller prizes. Contestants could only claim prizes  by sharing their methods openly, so others could then take their ideas  and leapfrog past them to bigger prizes. Part cooperation, part competition!

Then, the organizers actually hired  some of the most dedicated contributors to work on parts of the project in parallel  with the contestants, to spur people on. See, the problem of ink detection  here is fiendishly complex. At first glance, the unwrapped CT-scans  of the scrolls look totally text-less, but the writing is actually hiding in plain sight.

The smoking gun is the ‘crackle’ of certain parts of the scroll -  tiny cracks caused by ink residue. But this can be hard to tell  apart from other kinds of cracks and marks due to damage. It's a bit like making a rubbing of  a notepad to see what was written on the page above, but, like, with  the difficulty set to extreme.

The big breakthrough was using artificial  intelligence to scan images of the scrolls. Most of the attempts to unroll the  scrolls in the last few centuries have served to destroy them, but in some  instances ink can be read off of them, including via infrared scans. This meant that there was  a database of scroll images that had been successfully read.

Just what you need to train  a machine learning model to identify more text from images. So the contestants each designed  algorithms to scan images in tiny chunks, compare details to known ink sources,  and then spot ink in uncategorized images from scrolls that have been  scanned and digitally unwrapped. Of course, AI on its own can be unreliable,  so the organizers were careful to proof-read the code and verify the work  using human eyes whenever possible.

Also, the algorithm could only recognize  millimeter-sized bits of scroll textures, not whole letters, so it couldn’t  ‘hallucinate' fake letters like some image-processing  AIs are infamous for doing. The hard work eventually paid off. In late 2023, several people  won the first major prize for successfully deciphering one  of the words from the scroll.

Which was the Greek word for  “purple,” in case you were wondering. Soon after, three people who'd won prizes  earlier in the competition teamed up, and combined their expertise to win  the first Grand Prize of $700,000, for deciphering a whole section of text. The really cool thing is that they  weren't archaeologists by training.

They were computer scientists,  many of them students, handing archaeology a huge gift. In fact, the competition is still ongoing:  the winners shared their methods online and there's still chances to win further prizes. The passage that’s been deciphered so far is actually Greek philosophical  musings on the nature of pleasure.

That might sound like first  century People magazine, but it's actually Epicurean philosophy  - an intellectual movement that was taken seriously in its time and that  historians would love to recover more of. And look, we could talk all day about  the amazing historical implications of just this initial discovery,  but this is all just a warm-up. The hardest physics and computer  science parts are hopefully over, but if the scrolls can be scanned at scale, historians could have a mountain  of new discoveries to make.

In fact, some archaeologists think we  haven't even recovered all the scrolls yet. There might be more that remain buried  in the southern Italian countryside. And after thousands of years,  with the power of modern science, we may finally be ready to unlock their secrets and change our knowledge of  the classical world forever.

Meanwhile, our patrons are changing  our knowledge of the right-now world by helping us make these videos for everybody. You guys are super cool, so thanks. [♪ OUTRO]