In recent years, “representation learning, ” which assumes that knowledge presupposes representation, has been criticized. In this criticism “Wittgenstein'sparadox” has been used to point out the logical impossibility of “representation learning” : as long as the notion of “representation” is assumed, the paradox would happen inevitably and spoil “representation learning.” However, the paradox HERE is considered to be merely a theoretical fiction for it does not happens in normal cases, and hence, does not spoil learning. In view of the criticism, this paper attempts to show, by analyzing elaborately Wittgenstein's rule-following considerations, that “Wittgenstein's paradox” can not be seen simply as a fiction. In Wittgenstein's rulefollowing considerations, the notion of “inner-binding” and that of “outer-binding” are crutial. The former is a disposition that is built through “training”. This disposition enables one to judge the “sameness” and to be convinced of the necessity of that sameness. The latter is “institution” that is established through the repetition of “peaceful accordance”. The two bindings sometimes do not conflict with each other.This is exactly the case in which rule-following operates.
“Inner-binding” has not been formed yet in an earlier stage of “rule” learning.Therefore “outer-binding” conflicting with the inner does not appear. But the two can come to differentiate themselves from each other as learning progresses and as the inner has been formed. The conflict between the inner> and the outer is equivalent to the case which “Wittgenstein's paradox” describes. Accordingly the possibility of differentiation exists latently and the paradox is not a theoretical fiction.
The latent possibility of “Wittgenstein's paradox” indicates that learners are not in a position to distinguish the two. This suggests that learning is uncertain and learning theories are not transparent to learners.