Experts discuss the limitations and breakthroughs of neural nets in gaming, highlighting the success of AlphaGo and AlphaZero, which beat world-class players without any supervised learning on expert games, using only self-play.