How Artificial Intelligence Is Making 2,000-Year-Old Scrolls Readable Again

The innovative “Vesuvius Challenge” unlocked a mystery that had confounded archeologists for centuries

Emily Lankiewicz / Vesuvius Challenge

When Mount Vesuvius erupted in 79 C.E., it covered the ancient cities of Pompeii and Herculaneum under tons of ash. Millennia later, in the mid-18th century, archeologists began to unearth the city, including its famed libraries, but the scrolls they found were too fragile to be unrolled and read; their contents were thought to be lost forever.

Only now, thanks to the advent of artificial intelligence and machine learning, scholars of the ancient world have partnered with computer programmers to unlock the contents of these priceless documents. In this episode of “There’s More to That,” science journalist and Smithsonian contributor Jo Marchant tells us about the yearslong campaign to read these scrolls. And Youssef Naderone of the three winners of last year’s “Vesuvius Challenge” to make these clumps of vulcanized ash readabletells us how he and his teammates achieved their historic breakthrough.

A transcript is below. To subscribe to “There’s More to That,” and to listen to past episodes on the complex legacy of Sojourner Truth, how Joan Baez opened the door for Taylor Swift, the problem of old forest roads and more, find us on Apple Podcasts, Spotify or wherever you get your podcasts.

Youssef Nader: My name is Youssef Nader, I am a PhD student at the Free University of Berlin, and today I’m speaking to you from Alexandria.

Chris Klimek: Youssef spends most of his time in Berlin, but we caught him while he was visiting family in Alexandria, Egypt—which is a very busy traffic city. He said he was five stories up, and it still sounded like he was on the street.

Nader: We arrived in Alexandria somewhere around 2 a.m. in the morning, so I got some sleep and I woke up to have the interview, basically.

Klimek: Youssef grew up in Cairo, so from a young age he was surrounded by ancient history.

Nader: Papyrus was invented by ancient Egyptians almost 5,000 years ago, so learning about papyrus making, and how the ancient Egyptians went around documenting their history, is something you learn about very early on, and something that sticks with you. It’s very common to have souvenirs from Egypt, which is like papyrus with some hieroglyphs and some writings, and it’s a very common souvenir or gift that we bring people from here, and I brought my friends a couple of times. So yeah, it’s sort of a cultural heritage.

Klimek: Today, Youssef is a PhD student who works with machine learning and A.I.

Nader: I do work with image data, but I usually work with 2D images, like photos you take of your dog and stuff like that.

Klimek: One day, Youssef heard about something called the Vesuvius Challenge. It involved some unreadable ancient scrolls and the hope that some A.I. expert might be able to help, with a reward of $700,000.

Nader: It had all of the interesting elements: Papyrus, which rings a bell for an Egyptian, of course; playing around with historical data of 2,000 years ago just on my laptop is not something you come by very often; very interesting technical problem; a big monetary prize. It was just all of the right elements that make it worthwhile.

Klimek: It was a big challenge, but Youssef decided he was up to the task.

Klimek (to Nader): Have you ever seen one of the scrolls in person?

Nader: I have. I recently visited and I got to see scrolls up close. And it’s crazy. I could not believe that this is the same thing I’m working on on my computer, because it doesn’t look like there is hope. When you look at the scroll up close, it really looks like a piece of charcoal, and the sheets look like they merged together, it’s just one, and they’re very, very small. One of the scrolls was just my finger tall, so it was really crazy to think that this is what we’re working on and we were reading. It’s a little bit of science fiction.

Klimek: From Smithsonian magazine and PRX Productions, this is “There’s More to That,” the show where we may not welcome our robot overlords, but we are willing to let them help us read historically significant ancient papyrus scrolls. In this episode, we learn more about the Vesuvius Challenge, what happened and what A.I. means for the future of archaeology. I’m Chris Klimek.

Klimek: What are the Herculaneum scrolls, and why are they important?

Jo Marchant: They’re a collection of carbonized papyrus scrolls from around 2,000 years ago, ancient Roman times, that were buried by the eruption of the Vesuvius volcano. The same one that buried Pompeii.

Klimek: Jo Marchant is a Smithsonian contributor who’s covered this story for several years now.

Marchant: Often, the scrolls are described as the only intact library we have that survives from the ancient world. Because they were buried by the volcano, you’ve got these carbonized scrolls that were kept underground for all that time, so they have survived. But the only problem is you can’t unwrap them to read them without destroying them. So, they’ve been this big archaeological mystery since they were discovered in the 18th century.

Klimek: What do they look like now?

Marchant: Some of them have been pulled apart, and are basically crumbled into dust and they’re in hundreds of pieces, but there are a few hundred—the worst, most charred cases, if you like— that were left intact as a lost cause. They’ve been described as saggy, brown burritos, is one of the least rude descriptions that I’ve heard. They’re kind of crumpled, crushed, wrinkled. They look like nothing. They were thought, supposedly, to have been pieces of coal by the workmen who first uncovered them in the 18th century, so they just really look like very sorry objects indeed. You would not think that you were going to get a lot of information out of them.

Klimek: Do we know how they were recognized when they were found as carbonized scrolls? It sounds like they could have easily been mistaken for something else.

Marchant: Yeah, a lot of them were supposedly just thrown away, or burned, even, for heat by the workmen, these 18th-century workmen who had first uncovered them.

Klimek: What these workmen had discovered was an ancient library buried underground since the Vesuvius eruption in 79 A.D.

Marchant: The library itself was situated in this luxury Roman villa on the shore of the Bay of Naples. It possibly belonged to Julius Caesar’s father-in-law at one point, this beautiful villa with walkways, columns, statues, works of art, courtyards, this luxury residence. The workmen are digging tunnels, essentially, through the site, uncovering it, find these lumps, initially just think that they’re coal, burn them, throw them away. To be honest with you, I don’t know exactly how it was first realized that that was not what these things were, that they were actually incredibly precious. But once that was realized, then there was incredible interest, then, in trying to read them. This was a really unique, spectacular find. We just don’t have literary written sources from the classical world. Most of the works of literature or philosophy or whatever it is that we have have been copied and therefore selected through the centuries. But to actually have these original pieces from the time is just really, really incredible. So, there was all sorts of efforts to try and open these scrolls, most of which ended up being very, very destructive.

Klimek: What else has hindered efforts to read the scrolls, aside from the fact that they fall apart if you try to physically unroll them?

Marchant: Yeah, so technically, this is an incredibly difficult challenge. There have been attempts to open them, and essentially you end up with hundreds of pieces or strips, because it’s incredibly thin, this papyrus, you might have hundreds of rolls. So, imagine it’s tearing off in strips, but then you’ve got different layers then stick together. So, each of your strips might consist of a different number of layers, and then you’ve got to try and piece those together as a jigsaw. So, there has been a lot of work going on among papyrologists to try and decipher, translate, interpret those pieces, sticking the bits back together. But then they were kind of put aside as a lost course. I think a lot of people thought that those were never going to be read, they were just going to sit there in the library archive.

Klimek: As Jo mentioned, the scrolls were incredibly fragile, but that’s really just the beginning of why researchers were so stumped. First, how could they separate all the layers of paper?

Marchant: You’ve got to find a way of looking inside them, working out where the surfaces of the papyrus are, and then reading the ink. They’re so crumpled, and you’ve got all of these layers, some of them are stuck together, rolled very tightly. How do you even image and find the surfaces?

Klimek: Yeah. Then there was the ink itself.

Marchant: A lot of ink from ancient papyri has got iron in it, so if you X-ray, that ink will glow very brightly. But the problem with this ink is it’s just carbon and water. It has exactly the same density in X-ray scans as the papyrus. So, you can do your X-rays, you can do beautiful 3D scanning, whatever you’re going to do. But it’s like doing an X-ray of a body: You’re looking for the bones, but the bones are completely transparent; the ink doesn’t show up.

Klimek: Enter Brent Seales, a professor at the University of Kentucky.

Marchant: He’s a computer scientist, so he’s not a classicist, quite an unusual person to be spearheading this attempt to read these ancient scrolls. But he was originally interested in computer vision and then got interested in, how could you use algorithms to flatten out images? One of the first things that Brent Seales worked on was a very old copy of Beowulf in Old English that was kept in the British Library. Part of the problem when you take photographs of very old manuscripts like that is it’s all kind of warped, and sort of folded and cockled. The surface isn’t flat, so if you just take a photograph of it, you’re not going to be able to see all of the writing. So, the idea was to develop software where you could scan the not flat three-dimensional surface, and then flatten it out, so that you would have a nice, flat surface, you could read all the writing.

So, then moving from there to actually virtually unwrapping something that was rolled up. And a few years ago, the team did that on an old scroll from Ein Gedi, on the shores of the Dead Sea, that was burned by fire in the sixth century A.D. And they took the CT scans of that and were able to then virtually unwrap that surface, and see that, written inside, was actually some text from the Book of Leviticus. So, that was an incredible advance.

Klimek: Then in 2005, a colleague showed Brent Seales the Herculaneum scrolls.

Marchant: And he told me that that just blew his mind, just the scale of that challenge, and the potential for the information that you could find. But he’s quite interesting, in that he isn’t so interested specifically in some of these ancient Greek and Roman sources that most papyrologists would be interested in, he’s actually a devout Christian, and he is really interested in the origins of Christianity. The volcano erupted in 79 A.D., these scrolls were buried, so this was the time when Christianity was just beginning, and the philosophers in ancient Greece and Rome, in that world, would’ve been very aware of what was happening, probably interested in this new religion that was starting up. But he told me that what he was really dreaming of, really interested in, is finding out more information about that. Can we find information from early Christian sources?

There’s the huge technical challenges, but one of the biggest problems he’s actually had is getting access to the scrolls to even study them, and to try to develop these techniques, because they’re incredibly precious and incredibly fragile. So, curators who are in charge of these collections, the last thing they want to do is give them to some computer scientist who wants to carry them off to a particle accelerator somewhere and send beams of X-rays through them. This is something that’s taken nearly 20 years to really come together.

Klimek: I love this. “Can I borrow this irreplaceable treasure of yours? I’ll bring it right back, I just need to run it through my particle accelerator first.”

Marchant: Exactly, exactly.

Klimek: “It’ll be fine.”

Marchant: And I’ve spoken to curators, and they’d say you breathe on these things, they will fall apart. They are so fragile. So, it really is a kind of perfect storm of difficulties.

Klimek: Remarkably, the scrolls were eventually taken to a particle accelerator in the U.K. for 3D scanning.

Marchant: You’re making a 3D reconstruction of that volume, and then you have to go through, really painstakingly, slice by slice, and kind of mark where all the surfaces are. If you think about looking at one of these scrolls in a cross section, you’ll see a spiral of where the papyrus is all wound together, and you have to mark where all those surfaces are, and then what Brent Seales and his team did was work on software for algorithms that could take that data and then unwrap that spiral into flat surfaces. So, you get a kind of flat image of what that surface looks like in the CT scan, that you can then work on and try and look for the ink.

But as I mentioned, the ink in those images is transparent; you can’t see it. So, that then was the next challenge. How are you actually going to make that ink visible? They had one tiny fragment which had one letter on it, sigma, and they were able to carry that to the Diamond Light Source in Oxfordshire, and the idea was that just using that one letter, they were trying to come up with imaging techniques, and that’s where, a few years ago, they had the idea of using machine learning, these artificial intelligence techniques, to try to do that.

If you take some of the papyri that has been opened, some of these fragments, and you train your machine learning algorithm, you show it, “This is what ink looks like, and this is what not ink, just the blank papyrus, looks like.” You can teach it to be able to tell the difference, so then you can run that same algorithm on your CT scans from inside the wrapped-up scroll. That was the approach, but they realized that this was going to be an incredibly labor intensive … a lot of work to do this. And I think that’s the point at which Nat Friedman, the Silicon Valley entrepreneur, he had heard about the Herculaneum scrolls and contacted Brent Seales to say, “Right, what’s happening with this? Is there anything that I can do to help?” And that was the origins of this Vesuvius Challenge competition.

Klimek: Nat Friedman is the former CEO of GitHub, an online platform where computer programmers collaborate.

Marchant: And this whole project, actually, I find fascinating, because of the different worlds that come together. You’ve got the computer scientists. You’ve got these classicists and papyrologists who have their own culture and world. You’ve got the curators—they’re just really wanting to keep everything safe, they’re conservators. So, very different motives, very different cultures that these people are coming from. If you think of papyrologists, often it will take them years, decades to do a translation and edit an edition of a particular source. They’re so painstaking, they’re working character by character, just trying to work everything out. And then you’ve got the Silicon Valley entrepreneurs coming in, going, “Speed is everything! We are going to solve this now!” And you throw those two worlds together, I find it completely fascinating how, actually, in this case, that’s actually worked really well. It’s really triggered a lot of progress and creativity.

Klimek: So, how does all of this bring us, in 2023, to the Vesuvius Challenge?

Marchant: Nat Friedman told me that during the pandemic, during lockdown, he’s looking for things to do, like we all were, looking for distractions. Starts reading about ancient Rome, getting very interested in that whole world, finds out about the Herculaneum scrolls through just Googling, Wikipedia, all of this. Eventually comes across an online talk by Brent Seales talking about all of the work, and this problem with not being able to see the ink, and how he thinks that machine learning, artificial intelligence, might be the answer to that. And Nat said, from this talk, it sounded like Brent was pretty much there. He was going to solve it pretty soon, so he just thinks, “Oh, I look forward to finding out what happens with that.” Then, a couple of years later, it was like, “Oh, they don’t seem to have read the scrolls yet.”

So, he got in touch with Brent Seales to invite him to a retreat where a lot of tech figures, funders, that sort of whole community get together. Seales initially just ignored the email, just didn’t really believe who it was from. So, it took a bit of chasing, but he eventually realized that yes, this was Nat Friedman who was trying to get in touch with him. He went along to this retreat. It’s a camp-out in the woods in Northern California, where they all sit around fires and discuss projects, and, I don’t know, important decisions in the tech world get made. But nobody was actually interested in funding this project.

So, Nat Friedman, afterward, is thinking, “I don’t want this guy to go home with nothing, after I promised him that we’d be able to do something to help his project.” Basically, he said, “Why don’t we do it as a competition?” He and his longtime funding partner, Daniel Gross, put forward initial funding for the competition, and the idea was that you make all of your data open source to public, just put it out there, and then you set goals for people who can make different advances toward reading the scrolls. So, things like first person to detect ink, first person to detect a word, first person to read a whole passage. You set all of these different minds onto the challenge at once.

And the actual design of the competition is really interesting and really clever, I think, because rather than just having one prize and everyone’s working alone, because you’ve got these progress prizes, every time somebody wins a progress prize, all of their work, all of their data, all of their algorithms get made public. So, the way that Brent put it to me is you level everybody up, then, so everybody has the advantage of that, and then they all start working on the next challenge.

I asked Brent Seales, actually, was that difficult? If you’ve worked on a project for nearly 20 years, and your dream was you were going to be the person to read the scrolls, is that a hard decision to make, then, to say, “Actually, it’s not going to be me. I’m going to do this prize. I’m going to make everything I’ve done so far, everything I’ve worked for, all of our software, all of our data, let’s just make it public, put it out there, and then someone else can come and do that last step, and they will be the person to read it.” Can you imagine? How hard. And he said yeah, it was really difficult. The whole team had to talk about that together, and make sure that they were all OK with that.

Seales also said something else to me: He said often with archeology, and I’ve come across this with other stories I’ve written, actually, that somebody decides that they’re going to be the one to solve a mystery or whatever it is, make a discovery, and it’s almost like the ego takes over, it’s theirs, and they’re going to be the one to have all the glory. And he said this was almost a way to prove to himself that he wasn’t that person. That he’s doing it so that the scrolls can be read.

They put everything out there, made it public, launched the award toward the beginning of 2023, and it all went from there. I think they had more than a thousand teams, in the end, from all over the world, like China, Ukraine, Australia, U.S., Egypt, and they were all on this Discord, this chat platform for gamers, discussing latest advances and questions, because they were just releasing little flat images of the surfaces inside these scrolls, a little piece at a time. And then what the entrants for the Vesuvius Challenge were doing was then they would take those segments, those flat segments, and use those to then train their machine learning models to try and recognize that ink.

Klimek: Were there any unsuccessful avenues that were part of this that were included in your reporting? Any attempts that didn’t pan out?

Marchant: I think there were lots of teams trying different things, trying to train their algorithms in different ways. So, one thing that Seales thought they might be able to do was to train the algorithm on the letters from the parts of the scrolls that have been read, but that ended up really not working very well. It seems that you have to train your algorithm on the same scroll of the scans that you’re trying to read, which is obviously very difficult, because you can’t see the ink. How are you going to do that?

One of the first real key breakthroughs, there was an ex-physicist called Casey Handmer. He was actually looking at the images that were coming out from inside this scroll visually, and just spending hours and hours poring over them. He was convinced that if a machine learning algorithm could see a difference, a lot of those are trained based on the human visual system. So, he was thinking, “If a machine can see it, it must be possible for a human to see it, if we just look carefully enough.” So, he’s pouring over these images and eventually notices this very strange, very subtle difference in texture.

So, normally in the CT scans, you can sort of see the woven strands of the papyrus, and then in some places there was this … It’s described as being like cracked mud on a riverbed, those geometric kind of cracks you get. So, they called it crackle. He was trying to look at this, trying to work out where it was, and then realized in one place, it seemed as if it was forming the shape of a letter. So, he was like, “Oh my goodness, this is the ink.” This is not showing up as a different color, it’s not glowing bright or anything, but there’s just this very, very subtle difference in the texture of the surface where the ink is sitting on the papyrus. And he was awarded the First Ink Prize for doing that. So, then other competitors were able to use that to train their algorithms. Now they’ve got a foothold, they’ve got something to start training their algorithms on the difference between ink and not ink.

Klimek: After that, the race was on. Who would find the first word to read from the Herculaneum scrolls?

Klimek (to Nader): Can you give us a simple definition of what machine learning is?

Nader: Machine learning is about how to teach a statistical model to map your input data to some output result that you want. For the Vesuvius Challenge especially, we wanted to teach the A.I. model what ink looks like.

Klimek: Nader again.

Nader: So, you give the A.I. model some small images, some patches of the image, because the segments are really huge, it’s like hundreds of thousands of pixels by hundreds of thousands of pixels. It’s crazy resolution. So, you take a small piece, you show it to the A.I. model, and the A.I. model needs to say, “I see ink in this small piece” or not. And to train this, you need some examples to show it to begin with. So, we tell it, “OK, this is what ink looks like. This is what ink doesn’t look like.” And you show it these examples, and then it’s able to learn, “OK, how do I differentiate between the two?” And then it notices, “OK, there’s this pattern on top of the papyrus that looks quite like cracks, that maybe this I can use to detect the signal.”

And of course there were very interesting problems, because to begin with, we can’t see the ink ourselves, so it didn’t have the data that we can show to the A.I. model to say, “OK, this is what ink looks like.” And it took a lot of experiments and a lot of ways to find a first footing of ink from small pieces that fell off the scrolls: first two letters. How do we go from two letters to 2,000 letters? You train an A.I. model to learn these two letters that you found, and it has a slightly better idea of what letters look like, so it finds another ten letters. You take those 12 letters now, and you train a new one with the 12 letters. The new A.I. is better, so it finds maybe 20-something letters.

And the beginning was incremental. I would usually just take the predictions from an A.I. model, like, “OK, these are letters.” I would paint over them in Photoshop to make some examples of what ink is, so just like a black and white image, and I would give it to the next A.I. model. Of course, my drawing is not very accurate, and it was a question of how do you allow the A.I. model to disagree when you have some mislabeled stuff? How do you guarantee that the A.I. model is not hallucinating, not making up letters? And we had to operate on a very, very small scale, such that the letter is never seen by the A.I. model. It only predicts pixel level: ink, no ink, ink, no ink. And then we, as humans, when we look at the big picture, we see, “OK, yeah, this is actually Greek, this is what it means.”

Klimek: This is how you can have confidence in one set of findings before you move on to the next set. You’re verifying the machine learning conclusions with human eyes before you feed those discoveries back into the A.I. to try to solve the next set.

Nader: Yeah. So, in the training phase, I was verifying this by my own eyes, which, I’m no expert in Greek, I actually don’t know any Greek. So, I was just looking at what makes sense as a writing, like any kind of written language. You have some ink deposits, and you draw a letter in some shape. It makes sense that the letters are all on a single row, it doesn’t make sense that there’s scrambled rows; fixed-size columns, stuff like that. I go to sleep thinking about the Vesuvius Challenge. I wake up, check some stuff, continue working, eat, sleep, then repeat. I wasn’t even getting proper sleep because I’m going to bed and thinking, “OK, did I actually try that thing? Maybe I have a different idea, maybe I should do this.” And I run something overnight, and I check in the morning if it worked or not. So yeah, we were grateful that the first words that we found was not something like, “and,” “the,” for example. That would’ve been underwhelming. It had some meaning, it had some kind of zest to it, and I think that was really cool.

Klimek: Youssef was one of two people to find that first word. It was—drum roll, please—“purple.”

Marchant: So, that was the first word, “purple.” Which is lovely, I love that it was just such a rich, evocative word.

Klimek: Marchant.

Marchant: So, immediately that said to the papyrologists, “We think this is a new work we’ve never seen before.” Because “purple” is quite a rare word. Purple, porphyras, is the name of a dye. It was made from sea snails, so very expensive, difficult to make, so used to dye the emperor’s robes. This was a sign of wealth, luxury, rank. It’s just this lovely sort of … Yeah, just evocative word. So, that was the First Letters Prize, awarded in October to Luke Farritor, who got the first place for that prize.

Klimek: Luke Farritor was a 21-year-old computer science student at the University of Nebraska. Youssef won second place. The two reached out to one another after the announcement, eventually deciding to team up. They were joined by a third student named Julian Schilliger. Together, the three set their sights on the next phase of the competition.

Marchant: When the whole challenge was set up in March 2023, they had this big $700,000 Grand Prize for reading the first passages from the scroll. And a deadline was set for that prize, which was the 31st December 2023, so the end of that year. Nat Friedman said it was getting nearer and nearer to the end of the year, and they’re not getting any entries for this Grand Prize. They were getting pretty worried. They were starting to send out messages going, “So, how’s everyone getting on? Let us know your progress!”

Klimek: Entrants to the Vesuvius Challenge worked right down to the wire. Youssef and his teammates were no exception.

Klimek (to Nader): What were your last few days like, prior to the deadline?

Nader: They were quite sleepless. I was trying to make sure that I’m not submitting on the last day, which I usually do in every other thing. I knew that a lot of people would be submitting at the very last day or the very last minute. I was also not sure about … There was a time factor. If you get to the threshold of winning first, you win. I was not sure: Where are we on that? Do we have the best models? Where are we? You don’t know about other teams. And so you also want to guarantee that you’re first, in case there’s a tie. So, there was the time factor and the quality factor, and you’re trying to, “OK, do I submit now? Do I try to make it better over the next week? Is it getting better? It’s not getting better.” And I made one submission 22nd December, and one 30th December, so, one day before the end of the competition.

I was just planning to go back to Egypt to visit my family after the long haul of the Vesuvius Challenge. It was the date after I arrived in Egypt. They sent us an email, saying, “Hey, the evaluation process is still ongoing, we’d like to meet with you guys.” Of course, we’re in different time zones, and they wanted to make sure we’re all in one meeting when they tell us the news. So, we didn’t know that we were getting the announcement, and we were suspicious. “OK, why do you need all three of us in a meeting?” We were like, “We can answer the questions over email.” Julian was saying, “Yeah, it doesn’t make sense.”

We went to the meeting, and then they were asking us normal questions, and we were like, “OK, yeah, maybe it’s still ongoing.” And then Nat was like, “How would you guys feel if we told you that you won the Vesuvius Grand Prize?” And it was like, “What?” And I think it took us a couple of days for it to sink in, actually, that we actually won. And we were in disbelief, but we were ecstatic, and it just felt amazing.

Marchant: The three of them working together, they’d actually read, I think it was more than 2,000 characters from this scroll, more than 5 percent of the entire scrolls. And these are really big, long, long scrolls. And it was discovered that it was a work of philosophy by an ancient Greek philosopher called Philodemus. And that in itself was not a huge surprise, because of the scrolls that had been attempts made to open them and partially read, a lot of those scrolls were written in Greek and were philosophy works by Philodemus. He was a follower of Epicurus who founded the Epicurean school of Greek philosophy. They thought everything in nature was made of atoms that swerve and collide. And there’s so many works, actually, of Epicurean philosophy that they think that that part of the library was probably the working library of this philosopher, Philodemus.

And it seems to be a work on pleasure, and the senses, and on what gives us pleasure, possibly relating to music. It’s mentioning the color purple, it’s mentioning the taste of capers. There’s a character called Xenophantus who is mentioned, who is possibly, there’s a Xenophantus known who was a very famous flute player, who apparently his playing was so evocative and stirred the heart so much that his playing always caused Alexander the Great to immediately reach for his weapons. So, you get a sense of all these lovely sensory sources of pleasure that are being mentioned in this piece. So yeah, papyrologists are really, really excited about that. But then also what this means for what else we could be reading from now.

Klimek: I asked Youssef what other archaeological problems he’d like to see machine learning tackle.

Nader: I think there are very interesting projects of machine learning in archaeology, even outside of reading a scroll. I think there has been discussions of using similar techniques to read writings on wrappings of mummies. I know of one other project in our university that has to do with using 3D reconstruction and imaging for archaeological sites, using drones to scan the sites, and figure out structures and stuff. There are some interesting problems that are either really hard to solve, or require a lot of man effort, and A.I. could really help us speed things up.

Klimek: Do you think most people who don’t have your specialized background and education, do people understand generally what artificial intelligence is?

Nader: Artificial intelligence has been getting a lot of bad reputation recently, also because of how it has been used. I think sometimes people think it’s a lot smarter than it actually is, and some people think it’s a lot dumber than it may be. I believe it’s a very interesting tool, depends really how you use it. A lot of the fear and concern from A.I. comes from not treating it as a tool, but as an entity of its own that wants to do either good or bad. But the good or bad is basically coming from the human operating the tool. I think there’s a lot of debates coming from the world-leading experts in A.I. about what actually are the risks, and how to interpret what we are doing. So, it’s still kind of an ongoing process, but there is some awareness of, OK, there is this new technology that is shaping the world.

And I’m glad that the Vesuvius Challenge came at this time, because it also shows, yeah, you can do harm with A.I., but you can also do so much good, and so much benefit to mankind. So, some people are starting to think, “Yeah, maybe this is not really as bad as we thought.” Or, “We could really use this for our own good.”

Klimek: Thank you, Youssef, this has been fascinating.

Nader: Yeah, thank you, Chris.

Klimek: To read more of Smithsonian magazine’s coverage of the Vesuvius Challenge, check out the links in our show notes. And as always, we’d like to send you off with a dinner party fact. This time, we bring you a brief anecdote about another fragile thing that lives buried, not under ash, but under ice.

Megan Gambino: Hi, I’m Megan Gambino, and I’m a senior web editor at Smithsonian magazine. I recently edited a story about ice worms. I had no idea what these things were until this story, and they’re tiny, about inch-long, worms that live in glacial ice. They’re actually the only macroscopic animals that live in glaciers. But what I found interesting about them is that they’re both hardy and fragile at the same time. And what I mean by this is they can live for years without food, and they live at freezing temperatures, and yet they can only survive at this tiny temperature range, hovering right around 32 degrees Fahrenheit. Any colder, they get hypothermia; any warmer, they get room temperature, their membranes melt. So, I found that they were this interesting critter that was both tough and delicate at the same time.

Klimek: “There’s More to That” is a production of Smithsonian magazine and PRX Productions. From the magazine, our team is me, Debra Rosenberg and Brian Wolly. From PRX, our team is Jessica Miller, Genevieve Sponsler, Adriana Rozas Rivera, Ry Dorsey and Edwin Ochoa. The executive producer of PRX Productions is Jocelyn Gonzales. Our episode artwork is by Emily Lankiewicz. Fact-checking by Stephanie Abramson. Our music is from APM Music.

I’m Chris Klimek. Thanks for listening.

Get the latest History stories in your inbox?

Click to visit our Privacy Statement.