New AI Learns Ancient Greek to Decode Tattered 2,000-Year-Old Scrolls

The Austrian Academy of Science, working with French AI lab Mistral and tech firm Sail Reply, is releasing Apollo — the first advanced AI chatbot trained specifically on Ancient Greek. It learned from roughly 600 million words drawn from manuscripts, papyrus fragments, and stone inscriptions. Starting Wednesday, academics anywhere can use it for free.

The problem it solves is genuinely hard. Ancient Greek was written without spaces between words, and many scrolls are torn, faded, or missing chunks. Filling those gaps normally takes a rare expert — one of maybe a few dozen people worldwide — who can date the document, recognize the dialect, and choose words that fit historically. Apollo absorbs that skill. Show it Homer, and it responds in Homeric Greek; show it a Doric inscription, and it switches to Doric.

In practice, scholars can ask Apollo for the three most likely words to fill a gap, speeding up work that once consumed hours. Experts caution it won't magically surface lost plays by Sophocles — most unrestored papyri are mundane: personal letters, marriage contracts, tax records. But those ordinary documents, decoded faster, add up to a richer picture of how people actually lived.

The catch: AI works in probabilities, so it can guess confidently and wrongly, potentially polluting the historical record. Apollo is designed to offer a menu of options rather than a single answer, leaving the final call to a human. "Human competence needs to remain," says historian Anna Dolganov. "If we become totally reliant on AI transcriptions and interpretations of historical material, that's when the problems start."