Many ancient texts have become so damaged over the years that it is now almost impossible to understand them. There are large holes where pieces of papyrus, marble, or stone have been ruined, and trying to reconstruct the text from the fragments that remain is very difficult or impossible.
But now, researchers from the University of Oxford in the UK, together with the AI company Deep Mind, have trained an AI that can interpret the texts.
The researchers used texts written in Greece in the 5th century BC to train AI. Ithaca, as the AI is called, could not only make good guesses about what words were missing in the texts. It could also identify where and when the text was written.
Ithaca had an accuracy of 62 percent when it came to guessing which words were missing. Human historians ended up at anywhere from 25 to 72 percent. When it came to where the text was written, Ithaca was right in 71 percent of the cases, and it could date the texts within a 30-year span.
The researchers will now go ahead and train Ithaca in other ancient languages such as Akkadian and Maya.
