Basically, in order to communicate perfectly, we would have to know everything the other knows, and also that he knows that we know, and vice versa. But ultimately, this leads to an infinite regress, because we would have to check for every possible assumption someone could have. (A knows that o, B knows that A knows that p, A knows that B knows that A knows that p, and so on)
“Knowledge of this infinitely regressive sort was first identified by Lewis (1969) as common knowledge, and by Schiffer (1972) as mutual knowledge.' The argument is that if the hearer is to be sure of recovering the correct interpretation, the one intended by the speaker, every item of contextual information used in interpreting the utterance must be not only known by the speaker and hearer, but mutually known.” (18)
Because we can’t check for every single assumption implicit in an utterance there is never any guarantee that we might understand each other. To cut a long story short, basically this means that there must be some other way of understanding each other, which doesn’t presuppose that we assume that the others has knowledge of this and that sort, and that the other assumes that we have certain knowledge of some sort or other. The question would be which principles we actually use in understanding each other.