NEWS23 July 2025The best results come when a human and the model work together.ByRachel FieldhouseRachel FieldhouseView author publicationsSearch author on: PubMed Google ScholarFragment of a military diploma from Sardinia. AI model Aeneas can predict text lost from a damaged inscription (grey text) without needing to know the length of the section that is missing.Credit: Yannis Assael et al./NatureAn artificial intelligence (AI) model can predict where ancient Latin texts come from, estimate how old they are and restore missing parts. The model1, called Aeneas and described in Nature today, was developed by some of the members of the team who created a previous AI tool that could decipher ancient Greek inscriptions.Studying ancient inscriptions, known as epigraphy, is challenging because some texts are missing letters, words or sections, and languages change over time. Historians analyse texts by comparing them with other inscriptions containing similar words or phrases. But finding these other inscriptions is incredibly time consuming, says co-author Thea Sommerschield, an epigrapher at the University of Nottingham, UK.Another challenge is that new inscriptions continue to be discovered, so there is too much information for any single person to know, says Anne Rogerson, who studies Latin texts at the University of Sydney, Australia.How AI is unlocking ancient texts — and could rewrite historyTo make it easier to restore, translate and analyse inscriptions, a team including researchers from universities in the United Kingdom and Greece, and from Google’s AI company DeepMind in London, developed a generative AI model trained on inscriptions from the three of the world’s largest databases of Latin epigraphy. The combined data set contained text from 176,861 inscriptions — plus images of 5% of them — with dates ranging from the seventh century bc to the eighth century ad. The model comprises three neural networks, each designed for different tasks: restoring missing text; predicting where the text comes from; and estimating how old it is. Along with the results, Aeneas also provides a list of similar inscriptions from the data set to support its answer, ranked by how relevant they are to the original inscription.“Aeneas can retrieve relevant parallels from across our entire data set instantly” because each text has a unique identifier in the database, says co-author Yannis Assael, a research scientist at Google DeepMind.The team tested the accuracy and usefulness of the model by asking 23 epigraphers to restore text that had been removed from inscriptions. The specialists were also asked to date and identify the origins of inscriptions, both alone and with the help of the model. On their own, the experts dated inscriptions to within around 31 years of the correct answer. Dates predicted by Aeneas were correct to within 13 years.When it came to identifying the geographical origin of the inscriptions and restoring parts of a text, the specialists who had access to the model’s list of similar inscriptions and its predictions were more accurate than specialists working alone or the model alone. The specialists also dated the inscriptions to within around 14 years of the right answer when they had the model’s list and predictions.Historian helperdoi: https://doi.org/10.1038/d41586-025-02335-xRead the related News & Views: ‘AI for Latin inscriptions supplies missing text and predicts date and location’ReferencesAssael, Y. et al. Nature https://doi.org/10.1038/s41586-025-09292-5 (2025).Article Google Scholar Download references AI reads text from ancient Herculaneum scroll for the first time How AI is unlocking ancient texts — and could rewrite history The AI historian: A new tool to decipher ancient textsSubjectsMachine learningDatabasesHistoryLanguageLatest on:Jobs Associate or Senior Editor, BMC PsychiatryJob Title: Associate or Senior Editor, BMC Psychiatry Locations: New York or Shanghai – Hybrid working model Application Deadline: August 3rd, 20...New York City, New York (US)Springer Nature LtdAssociate or Senior Editor, Nature Biological, clinical, and social scienceTitle: Associate or Senior Editor, Nature Biological, clinical, and social science Organization: Nature Portfolio Locations: New York or Jersey Cit...New York City, New York (US)Springer Nature Ltd