Aeneas transforms how historians connect the past
read at source ↗ deepmind.google
Aeneas transforms how historians connect the past
Source: DeepMind Date: 2025-10-24 URL: https://deepmind.google/blog/aeneas-transforms-how-historians-connect-the-past/
Summary
Google DeepMind released Aeneas, the first AI model for contextualizing ancient Latin inscriptions, trained on 176,000+ inscriptions from EDR, EDH, and EDCS-ELT epigraphic databases. Key results: 73% Top-20 accuracy for character restoration (gaps up to 10 chars), 72% accuracy for identifying one of 62 Roman provinces, and dating precision within 13 years of historian-provided ranges. Validated in a study with 23 historians. Free public interface at predictingthepast.com; code and dataset open-sourced.
Implications
Science-as-moat in the humanities. AlphaFold addressed proteomics; Ithaca addressed ancient Greek; Aeneas addresses Latin epigraphy. DeepMind is systematically building domain-specific AI tools for the scholarly community — different domains, same strategy: become the essential computational tool for a research field and gain the associated scientific credibility.
The 73% Top-20 restoration accuracy is a working-tool number. 73% in Top-20 means the correct character restoration appears in the model’s top 20 suggestions 73% of the time. For a historian doing tedious manual restoration work, that’s a genuinely useful shortlist, not a solved problem. The model is a human amplifier, not an autonomous archaeologist.
Open data + open code is the academic adoption play. predictingthepast.com as a free public interface plus open-source code on GitHub is designed for adoption by universities and research institutions with no budget for commercial AI tools. If it becomes the standard tool for epigraphers, Google owns that academic relationship.
Watch:
- Adoption by classical studies departments and major epigraphic databases as a standard research tool
- Extension to other ancient languages (Demotic, Linear B, Sumerian) — the methodology generalizes
- Whether Aeneas-style tools become a pattern for Google’s humanities AI portfolio