Researchers have trained a large language model to work with badly damaged ancient Greek papyri, according to a report from WIRED. The model, named Apollo, is designed to suggest plausible missing text where physical fragments have been lost or worn away. Rather than simply translating intact passages, it reconstructs what is likely to have been written in the gaps.
The project highlights how AI tools are moving beyond modern language tasks into specialized humanities research. Because papyrus fragments are often tattered and irregularly shaped, the model was adapted to handle incomplete lines and unusual layouts. The developers hope it will help historians uncover new details about everyday life in the ancient world.
While the tool is still a research effort rather than a commercial product, it demonstrates a growing niche for AI in archival and archaeological work. The WIRED report notes that the model is part of a broader push to apply machine learning to historical documents that are too fragile or damaged for traditional reading methods.