Researchers have announced the launch of Aeneas, a pioneering artificial intelligence model designed to contextualize, restore, and interpret ancient inscriptions—offering historians a powerful new lens for decoding the Roman world’s fragmented past. Developed in a landmark collaboration between Google DeepMind, the University of Nottingham, and partners at Warwick, Oxford, and the Athens University of Economics and Business, Aeneas is being hailed as a breakthrough in both historical research and AI.

The Everyday Writing of Ancient Rome – And Its Challenges

The Roman Empire was awash with written records: imperial decrees, everyday graffiti, love poems, and business ledgers were inscribed on stone, metal, pottery, and walls. These inscriptions provide invaluable, first-hand glimpses into ancient life—far beyond what survives in elite literary texts. Yet, after centuries of weathering, many inscriptions are now fragmented, damaged, or deliberately defaced, making their interpretation, dating, and geographical placement an enormous challenge.

Traditionally, historians have relied on specialized expertise to identify “parallels”—similar texts found across the ancient world that can shed light on a fragment’s meaning or origin. This time-consuming detective work has long limited the scope and speed of epigraphic research.

What Makes Aeneas Unique?

Speed and Intelligence at Scale

Aeneas dramatically accelerates the historian's task by reasoning across a massive, curated database of over 176,000 Latin inscriptions—uncovering textual and contextual parallels in seconds. This enables epigraphers to rapidly reconstruct missing sections, attribute time and place, and draw new connections between disparate texts and periods.

Multimodal and Multilingual Mastery

Beyond processing text, Aeneas analyzes images of inscriptions. This ability to interpret both written phrases and visual markers is unprecedented, allowing the AI to pinpoint the provenance of an inscription— even when key details are missing.

The model is not just limited to Latin: its algorithms can be adapted to decipher other ancient languages and media, including papyri and coinage. This flexibility vastly expands the historical reach of the tool.

Advanced Restoration Capabilities

Aeneas leads the field in restoring missing text, achieving a 73% accuracy rate for gaps up to ten characters and 58% accuracy even when the gap’s length is unknown—a notoriously difficult feat. Its sophisticated architecture attributes inscriptions to one of 62 ancient Roman provinces with 72% geographical accuracy and dates texts to within 13 years of the historian-established range.

These predictions are transparent: Aeneas provides interpretable “saliency maps” that highlight which segments of text and image shaped its conclusions.

Inside the Model: How Aeneas Works

Aeneas is a generative neural network that uses both transformer-based decoders for text and image-processing modules for visuals. It was trained on the Latin Epigraphic Dataset (LED), which harmonizes entries from leading databases like the Epigraphic Database Roma, Heidelberg, and Clauss Slaby.

For each new inscription, Aeneas encodes its knowledge—what’s written, where it came from, and what it resembles—into a historical “fingerprint.” It then retrieves the most relevant parallels, helping experts situate mysterious fragments within the broader ancient world.

Aeneas in Action: Resolving Scholarly Debates

To prove its capabilities, researchers used Aeneas to analyze the legendary Res Gestae Divi Augusti, Emperor Augustus’ autobiographical inscription—a text scholars have long debated. Instead of guessing a single date, the model produced a nuanced probability distribution, reflecting the two leading academic theories and offering a new, quantitative approach to an old debate.

The model’s contextual suggestions proved especially useful: for more than 90% of texts examined by professional historians in a controlled study, Aeneas inspired new lines of research or increased confidence in proposed restorations and datings.

Transforming the Humanities Through Collaboration

Historians + AI = Enhanced Results

Aeneas is not a replacement for expert interpretation—it is designed as a collaborative tool to amplify human judgment. In trials involving 23 professional historians, the best outcomes were achieved when Aeneas’ predictions and contextual suggestions were combined with expert reasoning.