Currently, artificial intelligence and machine learning are found in many areas, ranging from technology to science and home automation. But that’s not all, because artificial intelligence also proves to be very useful to help humans both in their usual tasks and in specific work.
Pythia is an artificial intelligence system designed by DeepMind researchers to help scientists recompose and understand fragmented texts in ancient Greek on tablets made of stone, clay and metal, dating back to 2,700 years ago. These materials are invaluable to scientists in that the inscriptions engraved there are thousands of years old and represent a treasure trove of history, literature and anthropology.
Although these tablets are valuable, researchers still have the difficult task of recomposing and deciphering them. Only, as they are 2,700 years old, many of them have been crumbled, spoiled, with missing pieces or incomprehensible inscriptions.
To design Pythia, the DeepMind team first created a software that could convert the largest digital collection of inscriptions in ancient Greek into text that can be integrated into an artificial intelligence system through learning. automatic.
Subsequently, DeepMind just had to create an algorithm that could guess more or less accurately sequences of potential letters that can nest in the white on tablets. To test the effectiveness of the software, PhD students and Pythia each received texts with parts artificially excised. It turned out that the researchers had found the right answer 57% of the time while Pythia got away with 30% correct answers.