A team of researchers co-led by Google DeepMind and the University of Nottingham has introduced Aeneas, a groundbreaking artificial intelligence model capable of contextualising ancient inscriptions. Detailed in a new Nature publication, Aeneas enables historians to draw connections from a wide array of historical evidence with unprecedented speed and depth. Traditionally, identifying significant parallels between ancient texts has depended heavily on expert memory and laborious manual searches, but Aeneas drastically shortens this process by instantly analysing thousands of Latin inscriptions for similarities, patterns, and provenance.
The model’s adaptability extends beyond Latin inscriptions, able to process other ancient languages, scripts, and even forms of media such as papyri and coinage. Developed in partnership with the Universities of Warwick, Oxford, and Athens University of Economics and Business, Aeneas is part of a broader effort to explore the potential of generative artificial intelligence in revolutionising the study of ancient texts. Its training involved the creation of the Latin Epigraphic Dataset (LED), a machine-actionable resource comprising over 176,000 digitised inscriptions meticulously curated and harmonised from decades of scholarly work on Roman history.
Aeneas delivers multiple world-first features: it performs deep parallels searches, turning each inscription into a historical fingerprint and linking it to broader contexts; processes multimodal input, combining textual and visual information to deduce geographical origins; restores gaps of unknown length in damaged inscriptions—a first for models in this field; and achieves state-of-the-art performance in reconstructing texts and predicting their origins. Notably, Aeneas is designed to help researchers assign meaning to fragmentary texts, reconstruct lost information, and gain richer historical insight. Eminent classicist Professor Dame Mary Beard lauded the model as a transformative advance, surpassing reliance on individual intuition and opening up entirely new horizons for the study of Roman epigraphy.
The full study and dataset are publicly available, and the research is expected to transform traditional historical methodologies by providing immediate, expert-level contextualisation at scale, fostering a deeper, more comprehensive understanding of ancient civilisations.
