Yeah, the more I learn about this the less I'm impressed. Seems like they wanted to steal DeepMind's/Blizzard's thunder and hacked together a demo that isn't quite as advanced as they made it out to be.
MC sampling is an essential tool (when used together with simulation and neural nets). Don't dismiss it yet, it has a lot to offer to deep learning. For example, if you don't have enough data, you can create it by brute forcing / simulation.