If humans can't do that either, and we cant't, why would you conclude that it's necessary to be able to do that in order to match human intelligence?

Or are you specifically talking about perfectly simulating human brains? Human brain emulations are only one very specific and narrow form a strong AI might take. But even in that specific subset of possible AIs, we have no real idea how precise the simulation might have to be. It might be perfectly acievable without even simulating individual molecules.