The AI that deciphers ancient Greek graffiti

Host: Nick Petrić Howe

Welcome back to the Nature Podcast. This week, the AI that takes on ancient Greek inscriptions.

Host: Benjamin Thompson

And rewilding Argentina. I’m Benjamin Thompson.

Host: Nick Petrić Howe

And I’m Nick Petrić Howe.

[Jingle]

Host: Nick Petrić Howe

First up, reporter Alex Lathbridge has been finding out about how an AI is deciphering ancient Greek graffiti.

Interviewer: Alex Lathbridge

It’s 2022 and humanity has not yet invented time travel, so that means it’s somewhat difficult to work out what people several centuries ago were up to. But one thing that can help us understand the past is what ancient people wrote down. Yes, of course, people wrote in things like books and scrolls – we know that. But they also wrote in other places – maybe an artist’s tag on the side of a pot, or words graffitied on the side of a building, or an inscription on the wall of an ancient toilet. Who knows? The people who do know, and study these inscriptions, have a name for them: epigraphs.

Interviewee: Thea Sommerschield

So, an epigraph or an inscription is a text which has been written onto a durable surface, such as stone or metal or pottery. And ancient civilizations, everywhere around the world, used to write inscriptions to record everything like sacred calendars, laws, dedications, leases. Historians study them and use them today to reconstruct the history, language, religion, mentality and politics of the ancient world.

Interviewer: Alex Lathbridge

That’s Thea Sommerschield from Ca’ Foscari University of Venice and the Harvard Center for Hellenic Studies. She’s a type of specialist historian known an epigrapher, who... wait… you’ll never guess this one, trust me… studies epigraphs. Epigraphers break this process broadly down into three main tasks: restoring any missing text, identifying the inscription’s original location and establishing the date the inscription was created. Or in basic terms: what does it say, where is it from and when was it written? But remember, these are ancient items or, to be technical at this point, they are inscribed supports. These might have been damaged, worn away, or moved thousands of miles from home, so this can take a very long time, and epigraphers’ knowledge is constantly updated as new information is unearthed.

Interviewee: Thea Sommerschield

These are extremely time-consuming and complex tasks. The main difficulty is that we are often missing this key information, so the text could have been excessively damaged, or we might not know where it was written, or it might not mention events or names that are helping us to date it. And methods such as radiocarbon dating are unusable because of the inorganic nature of the inscribed supports – stone, for instance – and historians go about tackling these challenges by finding textual and contextual parallels in other similar inscriptions. But it's really difficult for a human to harness all possible relevant data and discovering underlying patterns every time among such large-scale data.

Interviewer: Alex Lathbridge

So, not an easy task, for humans at least. But what about AI? This is where Yannis Assael from DeepMind comes in. For the past five years, Thea and Yannis have been developing Ithaca – a machine-learning tool that, when given an inscription, aims to predict any missing text as well as date and place it. Here’s Yannis, to explain how Ithaca can shed light on Hellenic history.

Interviewee: Yannis Assael

Ithaca is a deep neural network. Ithaca is designed to capture the context and deal efficiently with missing parts of the text. One of the most important design decisions we did was instead of predicting single values, we wanted to produce a large number of visualisations to present all of it as hypotheses to the historians.

Interviewer: Alex Lathbridge

By producing suggestions in the form of visualisations, Ithaca shows where its attention gets focused. It might highlight one region of text that makes it think that an inscription is from 500 BC, or a few key words might point to it being from a certain part of Athens. This gives epigraphers the opportunity to understand Ithaca’s reasoning and potentially get a new perspective on the inscription that they’re working on. But how do you go about training an AI to work on these inscriptions when they are found on ancient stone and ancient pottery? Ithaca is bringing together a wealth of data on Greek epigraphs and combining it with advanced technology that’s been developed to understand the words that you and I use on a daily basis.

Interviewee: Yannis Assael

A big part of the process of training Ithaca was actually to create a data set that is formed from about 80,000 ancient Greek inscriptions. There were a lot of caveats in this because, for example, they're missing parts of text that are lost in millennia. So, we had to circumvent these caveats by artificially corrupting the text and hiding information for the model that we knew was there so we could use it as a label and a target to train our model.

Interviewer: Alex Lathbridge

So, Ithaca was trained on a large amount of data. Digitised inscriptions would have chunks removed by the team – like how a historian might be given a broken Greek tablet – and assessed to see if it was able to make valuable predictions. And it was. The historians involved in this study assessed themselves on a task and said that they were about 25% accurate. But with Ithaca helping out, the experts said that their accuracy went up to 72%. And just on its own, Ithaca achieves 62% accuracy. I mean, apologies to statisticians everywhere, but those are just statistics. What does it actually mean in terms of deciphering the mysteries of history?

Interviewee: Thea Sommerschield

We have actually found many cases of Ithaca matching known historical knowledge for these texts. So for example, we saw how Ithaca dated an inscription from Athens, which mentions the General Nikias, to 413 BC. And this matches exactly the date range which historians had proposed for this text.

Interviewer: Alex Lathbridge

So, that has shown that Ithaca can do what it set out to and is trained on a huge amount of data. But it is uncertain data, from a field where the same inscription can be studied and re-assessed for decades, with different experts having different opinions. Could this uncertainty and potential bias in the data that it’s trained on affect Ithaca’s attempts to make meaningful predictions?

Interviewee: Yannis Assael

That’s a very valid question. What you describe is the circularity existing. However, we have seen that such large-scale models – and this is visible in multiple fields of machine learning – work as denoisers and actually are able to discover underlying patterns and generalise using all these hypotheses.

Interviewer: Alex Lathbridge

So, what Yannis is saying is that by being trained on such a huge amount of data and offering results as suggestions, there’s a better chance that it can avoid the trap of making predictions that seem perfect but are actually false. Now, ancient Greeks were merely one of many civilisations from around the globe that inscribed text that we still have today, so which area of the past could be in Ithaca’s future? And is history the new home for machine learning?

Interviewee: Yannis Assael

Ithaca’s architecture is applicable to any language from Latin to Mayan, and any written medium, like papyri or manuscripts. So, we're actually really excited to see how this evolves as more data is available, and the new directions that Ithaca will take. And for this way we have open-sourced all of Ithaca’s source code and, at the same time, we have created a public interface that researchers can use for the current research.

Interviewee: Thea Sommerschield

Ithaca has changed the way I think about the relationship between AI and the humanities, and specifically about the ways in which we can use machine learning to predict the past and not just the future.

Host: Nick Petrić Howe

That was Thea Sommerschield and Yannis Assael talking to Alex Lathbridge. We’ll put a link to their paper and the Ithaca platform in the show notes. We’ve also made a video about it, so if you want to see the system in action, make sure you check that out. Again, there’ll be a link to it in the show notes.

Host: Benjamin Thompson

Coming up, the team working to reintroduce lost species and rebuild ecosystems in Argentina. Right now, though, it’s the Research Highlights with Dan Fox.

[Jingle]

Dan Fox

Bees get a buzz out of salty nectar, as even insect pollinators have been found to prefer the morish combination of salty and sweet snacks. Researchers applied sodium-laced artificial nectar to flowers of five plant species common in the meadows of New England in the US, and placed these flowers in a meadow. They found that, compared with flowers containing nectar without added sodium, blossoms spiked with salt had almost twice as many visits from pollinators and attracted nearly twice as many species of pollinators, with visits from bees, flies and butterflies. The findings suggest that sodium can draw pollinating insects to plants. With climate change expected to alter the water cycle and availability of sodium to plants, the authors say this might have negative effects on plant populations and visitors to their flowers. Read that research in full in Biology Letters.

[Jingle]

Dan Fox

Observations of the ‘megacomet’ Bernardinelli–Bernstein show that this massive ball of rock and ice is a colossal 137 kilometres in diameter, confirming its status as the largest known comet. Astronomers used the sensitive ALMA array in Chile to study microwaves emitted by the comet. Carefully distinguishing between radiation emitted by the comet’s nucleus and the fuzzy aura of dust that surrounds it allowed the team to make the first measurement of its size. The results show that, as suspected from the comet’s brightness, it is the biggest known of the most common type of comet, which hail from a region at the very edge of the Solar System called the Oort cloud. The authors also found that the comet’s reflectivity is typical of Oort cloud objects. The team plan to keep studying the object as it approaches and rounds the Sun in 2031. You can observe that research in Astronomy and Astrophysics.

[Jingle]

Interviewer: Benjamin Thompson

The world is facing a biodiversity crisis, with habitats being destroyed, and species disappearing at a rate not seen since the last mass extinction. This week, Nature has a Comment article from a group who are trying to reverse this trend in Argentina. They’re part of Fundación Rewilding Argentina, which is, as its name suggests, an organisation that aims to restore ecosystems to their wild state, driven in their case by reintroducing species that have been lost. While the reintroduction of species in order to rewild habitats can be controversial, it has been very popular around the world, from the reintroduction of wolves to Yellowstone National Park in the US, to beavers into parts of the UK. Fundación Rewilding Argentina want to do more than introduce one species. In their work, they’ve been reintroducing multiple species to a series of protected areas around the country, in the hopes that this might kick start a whole ecosystem restoration, even in heavily depleted areas. I called up Emiliano Donadio, one of the Comment’s authors, who told me about one of the regions that are working in – the Iberá wetlands in northeast Argentina.

[Jingle]

Interviewee: Emiliano Donadio

The Iberá is a 1.4-million-hectare landscape that is located in one province. This is a huge complex of ponds and grasslands and small patches of forest and, over more than 100 years, it has been destructed by humans. At the beginning, it was hunting, and then rice plantations and then pine plantations came, so that basically degraded most of the ecosystem, and what did happen is that every single last vertebrate, from deer to jaguars, disappeared. They were locally extinct.

Interviewer: Benjamin Thompson

And so, that was the situation there then and in the past few years, you and your colleagues have been working to reintroduce some of these species. Which ones are we talking about?

Interviewee: Emiliano Donadio

I would say that we are trying to reintroduce all of them. Pampas deer – we are reintroducing there. We have already two self-sustainable populations and they are thriving by themselves. Giant anteaters was the first species that we reintroduced. There are also two self-sustaining populations. And we are working on the reintroduction of other species, like the jaguar. We have released four animals so far – one male, three females – and two of those females were released each with two cubs.

Interviewer: Benjamin Thompson

And what are some of the key hurdles you’ve had to overcome in Argentina? In your article, you talk about opposition to buying private land with donations from outside the country, for instance.

Interviewee: Emiliano Donadio

Yes, we have to overcome many difficulties. The first one is land acquisition. We buy large chunks of land to have the territory where we can reintroduce species. And of course, because all the money comes from foreign donors, so the very first that was seen by the Argentinian society and politicians as a way to buy lands for foreigners. All that faded away once we started to donate this land to the national government so the national government would create national parks. The second big problem that we have is the lack of tradition in Argentina regarding the translocation of wild animals from one area where they are abundant to an area where they went extinct. We do it sometimes, but it’s very, very complex. The problem is that when species are extinct in Argentina at the national level, like giant otters and macaws, we could go to another country for giant otters, like Bolivia or Peru, but the loss are made for basically combatting wildlife trafficking, so it’s almost impossible to get a wild animal from a neighbouring country. So, we are basically sourcing animals from zoos.

Interviewer: Benjamin Thompson

And when you actually manage to get animals that you’re going to rewild into these areas, you talk about some of the hurdles that you have to overcome there as well, right, almost, I guess, at the end of the process, in many ways?

Interviewee: Emiliano Donadio

So, because we have to work mostly with animals of captive origin, these captive animals need to be trained. They need to learn the skills to survive in the wild. Basically, for predators, it’s just learning how to hunt. But for macaws, we need to teach them how to fly, how to find food and what is food in the wild, and we need to teach them what a predator is because macaws from captive origins, they are very accustomed to land on the ground to feed on fruits or seeds, and that’s where they get caught by predators. So, we train them, we show them a stuffed fox, and teach them that every time they go down, they can go through a very unpleasant situation. It’s been pretty good, actually. Most of the animals we have been able to release, they don’t get killed by predators.

Interviewer: Benjamin Thompson

And you introduce species not just to help restore their populations but because you hope their presence will have wider impacts on the whole ecosystem, from plants to insects, because ecosystems are so connected. Has that worked? What have you seen?

Interviewee: Emiliano Donadio

One of our main goals when we reintroduce these species is to bring back the interactions in which these species were involved because those interactions are the basis of a fully functional ecosystem. And for instance, with jaguars, we have released them within the last 12 months, but it’s amazing how right away they rebuilt this interaction with native prey species. So, over 60 or 70% of what jaguars are consuming by now is capybaras, which are the main prey of jaguars where capybaras are very abundant. It’s been reported for wild areas in Brazil. The other thing is that macaws are feeding on 40 different species of plants, which is amazing. Whether that ends on, for instance, macaws dispersing seeds and as a result creating new patches of forest, we don’t know. We are working with a researcher trying to measure that.

Interviewer: Benjamin Thompson

So, promising early results then, but your approach isn’t without its criticisms, right?

Interviewee: Emiliano Donadio

The most important one comes from other conservationists that think because rewilding involves bringing animals back to an area that could end badly because they will bring new diseases to the area. I mean, that can be handled. And actually, we don’t want translocated animals to be disease-free. I mean, parasites evolve with the species that we are managing. But we take care of that using a specific veterinary protocol, so we think that it shouldn’t be an argument to stop rewilding activities. The other criticism is associated with the size of the founding populations. Some people argue that they are too small and that would result in a genetically weak population. Those are also issues that can be easily handled by bringing an unrelated individual from somewhere else in the future. I mean, it’s been done so many times around the world. Populations have recovered from just a few animals and they are doing fine.

Interviewer: Benjamin Thompson

I mean, what is your ultimate aim here? Because obviously ecosystems are impossibly complex. By introducing some species, are you trying to get things back to the way they were, or is there a risk that you will change things even further? What does the end goal of your project look like?

Interviewee: Emiliano Donadio

So, one argument that our critics would use is, ‘You don’t have baseline, so you don’t know where you are going with your reintroductions.’ That’s true – we don’t have a baseline. Nobody was measuring what was going on 150 years ago. But ecological theory can give you an idea of where you should be going, so we use ecological theory and predictions to lead our work, still acknowledging that there is some uncertainty, right? We might not get there. But I mean, our goal goes, I would say, beyond that. We want a diverse biological world, and I don’t want to leave this world with less species than the species I found. I want to leave this world with more species in more places because it’s good for ecosystems, it’s good for people and because it’s the right thing to do.

Interviewer: Benjamin Thompson

In your Comment, you said there’s an important social component of your work as well. Why is that important to the rewilding project?

Interviewee: Emiliano Donadio

So, we work in marginal areas, right? So, when we come and we say, ‘Okay, livestock is bad for biodiversity and for rewilding. Pine plantations are bad for biodiversity and rewilding,’ we need to provide these people with an alternative. We need to incorporate people into our rewilding programs. We need to provide them with economic alternatives that are nature-friendly. So, we include them so they understand that the more jaguars, the more deer, the more alligators, the more tourists, the more jobs they get, the better they do, and that’s a positive feedback. And once we are gone, because we are not going to be there forever, they become key stakeholders of the area and they defend it. So, that’s what we need from local communities and that’s why social work and involvement and engagement of these people is critical. It cannot be done without them.

Interviewer: Benjamin Thompson

That was Emiliano Donadio from Fundación Rewilding Argentina. You can find a link to his Comment in the show notes.

Host: Nick Petrić Howe

Finally on the show, it’s time for the Briefing chat, where we discuss a couple of stories that have been featured in the Nature Briefing. Ben, what have you found for us to discuss this time?

Host: Benjamin Thompson

Well, I’ve got a story that was reported on in Science, and it’s based on a preprint paper that isn’t peer reviewed yet, but it did rather catch my eye. And I think it’s something that could quite literally catch my eye. This is the story of a newly described species of bacterium that can apparently grow to up to 2 centimetres long, and in the story they helpfully say that that’s about the same length as a peanut.

Host: Nick Petrić Howe

Well, when I think of bacteria, I do not think of things that size. Is this normal for bacteria or do other bacteria approach this sort of size?

Host: Benjamin Thompson

I mean, there are some pretty big ones, but if we think about E. coli, that’s about 2 micrometres in length. But in this study, this species can grow up to, say, 2 centimetres, which is 20,000 micrometres in length. And what’s interesting here is you can see it with the naked eye. It is a single cell. You can get things like algae which you can see filaments of, strands of, but that’s lots of cells kind of next to each other and this is, say, just one.

Host: Nick Petrić Howe

So, it’s a big cell then but, as I understand it, being small is actually kind of important when you’re a single cell.

Host: Benjamin Thompson

Well, I think for a long time, the general thoughts when it came to bacteria is that they were very small for a reason, and that reason is that most of the kind of chemicals and things that move about inside them happens because of passive diffusion, right? It just floats from one place to another and that’s how kind of the biochemistry happens. Now, passive diffusion doesn’t work at large sizes, but when it comes to larger bacteria like this one, there’s a few tricks they seem to have up their sleeve and, in this case, this bacterium and another one that’s been found previously have this kind of large, liquid-filled sack, and what that does is kind of push all the innards of the cell right up against kind of the cell wall. So, whatever needs to diffuse in and out only has to do so over a very short distance, so that allows it to grow to a much bigger size. And what else is neat about this new species, Nick, which is called Thiomargarita magnifica, is that its DNA also seems to be enclosed within a membrane, and this is kind of an interesting one because if we think back to sort of high school biology, if you think about the tree of life, you have your prokaryotes and your eukaryotes. We’re eukaryotes, and our cells have a nucleus, right? The DNA is within a nucleus. But prokaryotes, they don’t have a nucleus and the DNA is just sort of floating about. But this one is kind of different, if it’s shown to be correct, of course. But it’s got microbiologists, I think, quite excited.

Host: Nick Petrić Howe

And where might I be able to see these bacteria?

Host: Benjamin Thompson

Well, it’s a neat one because these were discovered ten years ago in the Caribbean by one of the researchers involved in this work, and these cells were growing on the surface of decaying mangrove leaves in a swamp. And he wasn’t really sure what they were, and it’s only, say, recently that the researchers have taken a look at it and analysed the DNA and tried to categorise what this actually is.

Host: Nick Petrić Howe

And other than being like just a weird thing that scientists are fascinated by, is there anything it can tell us more generally about bacteria and how they work?

Host: Benjamin Thompson

Well, one of the kind of big questions in science, I guess, revolves around the origins of complexity, right? If we think about eukaryotic cells being a lot more complex than prokaryotic cells, how did we get from one to the next? And it’s still an unanswered question, but I think this little oddity kind of blurs those lines a little bit. But whether this is just an oddity and it’s n=1 or whether it’s part of a larger thing is yet to be really understood, I think. But that’s my story for this week. Nick, what have you got?

Host: Nick Petrić Howe

Well, this week, I’ve been reading about the record-breaking and torrential rain that’s occurring at the moment in Australia.

Host: Benjamin Thompson

Yeah, absolutely. I’ve seen some of the pictures of this and it does look devastating for the people living in affected areas.

Host: Nick Petrić Howe

Yeah, so these were stories that I was reading in New Scientist and I also read an article in The Conversation, and basically, yes, it is quite devastating. Eight people have sadly lost their lives and thousands of homes have been damaged, and there’s more to come. The rains are still ongoing. And what it seems to be is just a result of many different things going wrong at a specific time and also climate change coming into the picture as well.

Host: Benjamin Thompson

So, when you say it’s many, many things going wrong at a specific time, what sort of things have happened then that have caused these floods?

Host: Nick Petrić Howe

Well, you might be familiar with certain oscillations in the Earth’s atmosphere and climate, such as El Niño and La Niña oscillations. These are temperature changes which occur across oceans, and El Niño is at the moment in the La Niña phase, which can lead to more rain in Australia. But this has coincided with some other oscillations in the Indian Ocean and in the Southern Ocean that have all come together in the worst possible combination to basically make a lot more rain.

Host: Benjamin Thompson

And what sort of levels of precipitation are we seeing then?

Host: Nick Petrić Howe

So, Brisbane has been one of the worst affected cities so far, and in a week they got 790 millimetres of rain, which might not mean a lot to you, but if I tell you that in London we get 690 millimetres in a whole year, it might give you an indication of just how much rain has fallen. And, as I say, there have been a lot of floods. Some rivers have risen over 14 metres above where they are typically, so it’s a heck of a lot of rain.

Host: Benjamin Thompson

And you mentioned climate change earlier there. What role is this playing in these floods, do we know?

Host: Nick Petrić Howe

Well, climate change, especially warming, has quite a literal role in the physical nature of the atmosphere. The warmer the atmosphere is, the more moisture it can hold. So, the Earth can only hold so much moisture at any given time, but the more the planet warms up, the more moisture that is and therefore the more rain that can actually fall. And additionally, we’ve seen many reports of how climate change can affect these different oscillations in the climate, these other systems, and cause extreme weather to be more common, and Australia is set to see many more floods like this, it seems. And this is particularly poignant for Australia because there have been reports, such as the Climate Change Performance Index, that have ranked Australia really quite low on how much it’s doing to tackle climate change.

Host: Benjamin Thompson

Well, let’s leave it there then, Nick, and our thoughts of course go out to anyone who is affected by those floods. Listeners, if you’d like to learn more about those stories, look out for links to them in the show notes, and you’ll also find a link on where to sign up for the Nature Briefing to get science stories like these delivered directly to your inbox.

Host: Nick Petrić Howe

That’s all for this week. If you want to reach out to us, then we’re on Twitter - @NaturePodcast. Or you can send us an email to podcast@nature.com. I’m Nick Petrić Howe.

Host: Benjamin Thompson

And I’m Benjamin Thompson. See you next time.