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
i feel like it's important to win a couple of times while learning to give the human brain some short term reward to keep going. I'd surely try this first on something low stakes if a kid gives up. You don't want to give a kid a mindblock from something `important`