Google DeepMind has introduced Aeneas, an AI model designed to help historians interpret ancient Roman inscriptions. This multimodal neural network processes both text and visual information to restore damaged texts, determine geographical origins, and identify contextual parallels across over 176,000 Latin inscriptions. Aeneas represents a significant advancement in digital humanities by automating what was traditionally a time-consuming manual process.

The model achieves state-of-the-art performance with 73% accuracy in restoring damaged texts and 72% accuracy in attributing inscriptions to specific Roman provinces. Unlike previous systems, Aeneas can handle gaps of unknown length in texts and produces probabilistic dating results rather than fixed dates. These capabilities were demonstrated when the model analyzed Emperor Augustus' Res Gestae Divi Augusti, capturing scholarly debates through quantitative date distributions.

Aeneas was developed in collaboration with the University of Nottingham and other academic institutions. Researchers can access an interactive version at predictingthepast.com while the code and dataset are open-sourced. A study involving 23 historians showed Aeneas improved restoration, dating, and geographical attribution tasks when used alongside expert knowledge, demonstrating how AI can enhance rather than replace human expertise in historical research.