Let us take an known example for the zero-knowledge proof like the Graphisomorphism (NP). I know there are different versions how to formulate an zero knowledge example for this problem but for this question it doens't matter.
I would like to know why it does make sense from a theoretical point of view to choose such a problem and ask somebody of whom I know already he can solve it if he can? As I read on wikipedia the prover in an interactive proofing system is in a higher complexity class than the problem (all mighty is atleat EXPSpace) then we know before trying to let the prover prove something that he will succeed.
So why is the prover chosen to be all mighty? Shouldn't it be possible that the result of the interaction reveals the complexity class of the prover (or atleast makes considerably assumptions possible, nobody knows if graph isomorphism has a P algorithm and because the problem isn't complete it would follow no collapsing of NP to P) like if the prover is able to solve an NP complete problem the prover is atleast in NP?