Google DeepMind has introduced Aeneas, a pioneering artificial intelligence (AI) tool that is designed to transform the study of ancient texts, particularly Roman inscriptions.
This generative AI model, building upon Google’s Ithaca helps historians by interpreting, restoring, and dating fragmentary Latin inscriptions more accurately and efficiently than ever before, ultimately helping uncover hidden patterns in these historical records.
Ancient Roman inscriptions, found on monuments, everyday objects, and documents, offer invaluable insights into the lives and societies of the past. However, many inscriptions survive only as damaged fragments due to severe damage, making their interpretation a slow and difficult process for historians.
Historians traditionally deciphering such inscriptions is a painstaking process that requires them to manually find “parallels” or similar inscriptions to fill in missing texts and contextualize the content. In this context, parallels are similar words, phrases, or contexts that hint at original meanings and dates. This approach can take months or even years to produce reliable interpretations.
The AI model Aeneas, named after a Trojan hero in Graeco-Roman mythology, was trained on the Latin Epigraphic Dataset (LED), the largest collection of Latin inscriptions ever assembled, comprising over 176,000 Latin inscriptions. As it is a multimodal AI system, it analyzes both the textual transcription of inscriptions and images of the physical artifacts. This dual input enables the tool to restore missing text, often with unknown gap lengths, with 73% accuracy for gaps up to 10 characters, and 58% for unknown gap lengths.
In extensive trials of the AI technology that involved 23 historians, Aeneas proved most effective when used collaboratively, as the accuracy in dating and interpreting inscriptions improved significantly. “Aeneas’ parallels completely changed my perception of the inscription. It noticed details that made all the difference for restoring and chronologically attributing the text,” one historian under anonymity said.
Beyond restoration and dating, Aeneas also helps historians quickly find relevant parallels from the vast database, by ranking similar inscriptions by relevance and narrowing down historical contexts. This especially helps streamline research workflows.
Aeneas has already been applied to significant historical texts such as the Res Gestae Divi Augusti, Emperor Augustus’s autobiographical record of his achievements. When asked about the dating of this inscription, the AI model generated a probability distribution that highlighted two plausible periods and reflected existing scholarly debates, providing nuanced insights that historians can explore further.
With the aim of promoting open research and education, Google DeepMind has made Aeneas freely accessible to the academic community via the interactive platform predictingthepast.com. To also support further research and innovation in this field, the tech giant made the AI’s code and dataset open-source.
This fusion of machine learning, archaeology, and history signals a new era where AI aids scholars to fully uncover the history of the past by contextualizing fragmentary ancient texts and unlocking richer and accurate understandings of history through technology.
