AI helps Latin scholars decipher ancient Roman texts
Inscriptions in Latin were commonplace across the Roman world, from laying out the decrees of emperors to graffiti on the city streets. One mosaic outside a home in the ancient city of Pompeii even warns: "Beware of the dog".
These inscriptions are "so precious to historians because they offer first-hand evidence of ancient thought, language, society and history", said study co-author Yannis Assael, a researcher at Google's AI lab DeepMind.
"What makes them unique is that they are written by the ancient people themselves across all social classes on any subject. It's not just history written by the elite," Assael, who co-designed the AI model, told a press conference.
However these texts have often been damaged over the millennia.
"We usually don't know where and when they were written," Assael said.
So the researchers created a generative neural network, which is an AI tool that can be trained to identify complex relationships between types of data.
They named their model Aeneas, after the Trojan hero and son of the Greek goddess Aphrodite.
It was trained on data about the dates, locations and meanings of Latin transcriptions from an empire that spanned five million square kilometres over two millennia.
Thea Sommerschield, an epigrapher at the University of Nottingham who co-designed the AI model, said that "studying history through inscriptions is like solving a gigantic jigsaw puzzle".
"You can't solve the puzzle with a single isolated piece, even though you know information like its colour or its shape," she explained.
"To solve the puzzle, you need to use that information to find the pieces that connect to it."
Tested on Augustus
This can be a huge job.
Latin scholars have to compare inscriptions against "potentially hundreds of parallels", a task which "demands extraordinary erudition" and "laborious manual searches" through massive library and museum collections, the study in the journal Nature said.
The researchers trained their model on 176,861 inscriptions -- worth up to 16 million characters -- five percent of which contained images.
It can now estimate the location of an inscription among the 62 Roman provinces, offer a decade when it was produced and even guess what missing sections might have contained, they said.
To test their model, the team asked Aeneas to analyse a famous inscription called "Res Gestae Divi Augusti", in which Rome's first emperor Augustus detailed his accomplishments.
Debate still rages between historians about when exactly the text was written.
Though the text is riddled with exaggerations, irrelevant dates and erroneous geographical references, the researchers said that Aeneas was able to use subtle clues such as archaic spelling to land on two possible dates -- the two being debated between historians.
More than 20 historians who tried out the model found it provided a useful starting point in 90 percent of cases, according to DeepMind.
The best results came when historians used the AI model together with their skills as researchers, rather than relying solely on one or the other, the study said.
"Since their breakthrough, generative neural networks have seemed at odds with educational goals, with fears that relying on AI hinders critical thinking rather than enhances knowledge," said study co-author Robbe Wulgaert, a Belgian AI researcher.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
France 24
2 hours ago
- France 24
EU urged to act on forests' faltering absorption of carbon
Forests, which cover 40 percent of the European Union's territory, are expected to play a crucial role in efforts to meet targets for overall reductions of the bloc's emissions of planet-warming gases. But human and climate pressures, from logging to extreme weather and insect attacks, means their ability to absorb CO2 is "rapidly declining", according to an article, led by the European Commission's Joint Research Centre. "Reversing the decline in European forests' ability to store carbon is essential -- and still possible -- with bold, science-based action today," said co-author Giacomo Grassi, who is a member of the UN's IPCC expert task force on greenhouse gas inventories. Solutions include rapid reductions in carbon emissions, combined with efforts to improve management to make forests more resilient to climate impacts, and comprehensive monitoring. Many European countries still rely on periodic inventories, which cannot keep up with rapid changes to forest health. The authors emphasize the need to better understand forest dynamics. They call in particular for better measuring of carbon flows between the soil, vegetation, and atmosphere, as well as improving predictions of how extreme weather will affect carbon sinks in the future. 'Steeper decline' The research looked at official 2024 data showing that the amount carbon absorbed by Europe's forests, ecosystems and changes to land use plummeted by around a third in the 2020 to 2022 period, compared to 2010 to 2014. The authors said 2025 figures "suggest an even steeper decline". "This trend, combined with the declining climate resilience of European forests, indicates that the EU's climate targets, which rely on an increasing carbon sink, might be at risk," the authors said. Earlier this week another study in the Nature journal Communications Earth & Environment found that land accounts for a quarter of global emissions reductions in countries' climate plans and warned that a lack of funding and conservation focus was putting these in jeopardy. Piers Forster, Director of the Priestley Centre for Climate Futures at Leeds University, who was not involved in Wednesday's study said it underscores the urgent need to slash emissions across the board. "We can't bet our future on carbon removal -- either from planting more trees, from protecting forests, or from emerging technologies such as direct air capture and storage -- without understanding what is already happening to the land and natural systems," he said. Scientists have warned that it is still unclear how carbon sinks might behave as the planet warms in future, and exactly how much heat-trapping carbon dioxide they might soak up from the atmosphere. In April, research by Climate Analytics, a policy institute that independently assesses countries' climate plans, warned that major economies are overstating how much carbon their forests can absorb in a climate accounting fudge that could allow them to use even more fossil fuels.
France 24
4 hours ago
- France 24
Physicists still divided about quantum world, 100 years on
"Shut up and calculate!" is a famous quote in quantum physics that illustrates the frustration of scientists struggling to unravel one of the world's great paradoxes. For the last century, equations based on quantum mechanics have consistently and accurately described the behaviour of extremely small objects. However, no one knows what is happening in the physical reality behind the mathematics. The problem started at the turn of the 20th century, when scientists realised that the classical principles of physics did not apply to things on the level on atoms. Bafflingly, photons and electrons appear to behave like both particles and waves. They can also be in different positions simultaneously -- and have different speeds or levels of energy. In 1925, Austrian physicist Erwin Schroedinger and Germany's Werner Heisenberg developed a set of complex mathematical tools that describe quantum mechanics using probabilities. This "wave function" made it possible to predict the results of measurements of a particle. These equations led to the development of a huge amount of modern technology, including lasers, LED lights, MRI scanners and the transistors used in computers and phones. But the question remained: what exactly is happening in the world beyond the maths? A confusing cat To mark the 100th year of quantum mechanics, many of the world's leading physicists gathered last month on the German island of Heligoland, where Heisenberg wrote his famous equation. More than 1,100 of them responded to a survey conducted by the leading scientific journal Nature. The results showed there is a "striking lack of consensus among physicists about what quantum theory says about reality", Nature said in a statement. More than a third -- 36 percent -- of the respondents favoured the mostly widely accepted theory, known as the Copenhagen interpretation. In the classical world, everything has defined properties -- such as position or speed -- whether we observe them or not. But this is not the case in the quantum realm, according to the Copenhagen interpretation developed by Heisenberg and Danish physicist Niels Bohr in the 1920s. It is only when an observer measures a quantum object that it settles on a specific state from the possible options, goes the theory. This is described as its wave function "collapsing" into a single possibility. The most famous depiction of this idea is Schroedinger's cat, which remains simultaneously alive and dead in a box -- until someone peeks inside. The Copenhagen interpretation "is the simplest we have", Brazilian physics philosopher Decio Krause told Nature after responding to the survey. Despite the theory's problems -- such as not explaining why measurement has this effect -- the alternatives "present other problems which, to me, are worse," he said. Enter the multiverse But the majority of the physicists supported other ideas. Fifteen percent of the respondents opted for the "many worlds" interpretation, one of several theories in physics that propose we live in a multiverse. It asserts that the wave function does not collapse, but instead branches off into as many universes as there are possible outcomes. So when an observer measures a particle, they get the position for their world -- but it is in all other possible positions across many parallel universes. "It requires a dramatic readjustment of our intuitions about the world, but to me that's just what we should expect from a fundamental theory of reality," US theoretical physicist Sean Carroll said in the survey. The quantum experts were split on other big questions facing the field. Is there some kind of boundary between the quantum and classical worlds, where the laws of physics suddenly change? Forty-five percent of the physicists responded yes to this question -- and the exact same percentage responded no. Just 24 percent said they were confident the quantum interpretation they chose was correct. And three quarters believed that it will be replaced by a more comprehensive theory one day.
France 24
14 hours ago
- France 24
'Marathon at F1 speed': China bids to lap US in AI leadership
Assumptions that the US was far ahead in the fast-moving field were upended this year when Chinese start-up DeepSeek unveiled a chatbot that matched top American systems for an apparent fraction of the cost. With AI now at the forefront of the superpowers' tech race, the World AI Conference (WAIC) that ended Tuesday saw China set out its case to take charge on shaping its global governance too. China, the United States and other major economies are "engaged in a marathon at Formula One speed", said Steven Hai, assistant professor of tech innovation at Xi'an Jiaotong-Liverpool University. "Which country will attain the upper hand can only be assessed dynamically over the course of development." China and the United States dominate the AI sector -- only 10 to 15 percent of models developed in recent years were built without either's participation, according to Epoch AI, a non-profit research institute. While US companies like Google and OpenAI are still industry-leading, the institute labelled 78 percent of Chinese models "state-of-the-art" compared to 70 percent of models built with American participation. Beijing's stated aim is to become the world's leading AI "innovation centre" by 2030. "Now China is neck-and-neck with the United States in terms of core tech, that play (for global leadership) is more relevant than ever," said Tom Nunlist, associate director for tech and data policy at Trivium China. "With a solid AI offering and the US turning inward, the question is will Beijing's vision gain greater global traction?" In May, Microsoft's Brad Smith told the US Senate that "the number-one factor" in the tech race "is whose technology is most broadly adopted in the rest of the world". 'Sovereign AI' China's offer is technical and economical. "One of the biggest differences (with the US sector) is that most of the leading models in China... are open-weight and open-source," former Google CEO Eric Schmidt told an audience at WAIC. That means they can be adapted by other countries to fit their own needs, said George Chen, partner at Washington-based policy consultancy The Asia Group. "We already see some countries like Mongolia, Kazakhstan, even Pakistan are trying to adopt the DeepSeek model to build their own," he said. "China has a chance to win in the aspect of sovereign AI to export its model to those countries." The comparative low cost of Chinese technology -- software but also hardware, for example through firms like Huawei -- will be a big factor, especially for developing countries, Chen added. On Monday another Chinese start-up, Zhipu, announced its new AI model -- also open-source -- would cost less than DeepSeek to use. In June, OpenAI accused Zhipu of having close ties with Chinese authorities and noted it was working with governments and state-owned firms across Southeast Asia, the Middle East and Africa. "The goal is to lock Chinese systems and standards into emerging markets before US or European rivals can," it said. Washington has moved to protect its lead in AI, expanding efforts to curb exports of state-of-the-art chips to China in recent years. "While limiting China's share of the global AI hardware market, (these measures) have accelerated indigenous innovation and led Chinese firms to exploit regulatory loopholes," said assistant professor Hai, referring to "rife" smuggling and circumvention. Issues of trust? Other challenges to homegrown firms include the closed nature of the Chinese internet, and "general issues of trust when it comes to using Chinese tech", Trivium's Nunlist said. At WAIC, China sought to present itself as a responsible power. Premier Li Qiang emphasised the risks of AI and pledged to share technology with other nations, especially developing ones. His remarks contrasted sharply with US President Donald Trump's aggressive low-regulation "AI Action Plan" launched just days before and explicitly aimed at cementing US dominance in the field. China released its own action plan at WAIC, following a meeting attended by delegates from dozens of countries. Li also announced the establishment of a China-led organisation for international AI cooperation. However, China's foreign ministry did not respond to a request from AFP for details on the set-up of the organisation -- including any international participants -- and several foreign delegates said they had not been briefed on the announcement beforehand. Analyst Grace Shao wrote it was clear AI was still in its "infancy stage". "You can sense that vibrant energy but also the immaturity of the space," she wrote on Substack.
