“A Go-playing AI requires millions of games to learn, whereas a human grandmaster plays only thousands or tens of thousands of games in their lifetime, highlighting the data inefficiency of current AI.”