Don't know if this would really happen much, unless we are saturated with labelled datasets. They are used for labelling and not directly responsible for the actual end classficiation which is what AI is aiming to solve. In fact, AI needs labelling to work on these classficiations