Not necessarily. For Go it is easy to simulate, hard to win. Many other things are easy to simulate at low level but have complex higher order behaviour (such as AutoML and ML graph compilation for GPU). A learning approach based on simulation is capable of finding super-human solutions, given enough compute.