People can learn things when they need to solve problems, forging "expert-enough people" out of otherwise lower-skilled folks with incredible speed. This only happens in the face of problems that need solving, and so far AI seems to be amazing at generating "problems that need solving" by being just good enough to get going but not so good as to solve all the unforeseen or unknown problems you encounter once you are going. I don't think this pattern will result in a world of only upside or purely better solutions than before AI hit the scene, but I also don't feel like it's quite as apocalyptic as many fear.
This is a good point; I agree that it's good that we have a nuanced and realistic (as opposed to pessimistic) view of the situation. As someone else pointed out regarding the Dunning-Kruger effect, the less knowledgeable we are the worse we are at understanding our own level of knowledge. Regarding your comment, not understanding the "unforeseen or unknown problems you encounter once you are going" could be catastrophic in certain fields.
