The incorporation of self-play in combination with other ideas such as deep learning is necessary to create an AGI system. Self-play allows systems to learn by exploring the world competitively against entities that have similar skill levels, leading to incremental improvement.