Self play is used in gamp playing AI in order to grow the available data of games. Game playing systems can be trained using records of past human games, for example, extensive records of chess competitions, However, this apprach is limited by the available data. Systems such as AlphaGo and its successors grew their available data by pitting AI aganst AI, this being able to create vast numbers of entirely machine-generated games. This data can then be used to train better game playing AI, and these used to generate yet more games.
Used on page 234
Self play over a (small) Go board