When a simple AI was taught to play Tetris, the best solution it found was to pause the game just before the screen filled up so it would never lose.