SCIENCE UNDER AUTOMATION
Contextualizing ancient texts with generative neural networks
Historians judged Aeneas retrieval useful in 90% of cases. Human–AI teams outperformed either alone on restoration and geographical attribution.
What this result supports
These are inscription tasks with historians evaluating context. Excavation and independent historical interpretation remain separate work.
Evidence record: Peer-reviewed expert workflow evaluation. Checked: 2026-10-21.
Where this evidence is used
Historical & archaeological research
- Detect candidates in images or texts: Image screening and inscription retrieval provide useful candidates from existing material.
- Compare and attribute candidate evidence: Aeneas improves expert workflows, with human–AI collaboration outperforming either alone on key tasks.
- Check whether a candidate is historically meaningful: Historians assess contextual parallels and researchers confirm geoglyph candidates in the field.
- Interpret provenance and cultural context: Contextualization is evaluated with historians; autonomous historical adjudication is not established.
- Resolve competing explanations with independent evidence: Physical confirmation and historical judgement remain researcher responsibilities in the demonstrated workflows.