Self play as a concept was first introduced in backgammon research by Gerald Tesauro's paper on training an expert using self play. This term later gained traction in AI research and has become a more general and powerful concept.