When you consider all the refinement through reinforcement learning managed by labelers and domain experts, it is indeed a simulation of the intelligence of those labelers.