One of the meta-challenges as I see it is “if we had a data set comprised of only the best information from the best experts we could make an AI that beats them

One of the meta-challenges as I see it is “if we had a data set comprised of only the best information from the best experts we could make an AI that beats them all”. Because I think that’s both true and entirely unrealistic. Like, how would you quantify who the best experts are, and how would you verify that you’re not training on wrong data from those best experts? Maybe they had an off day or a specific agenda. Even browsing through verified SO answers there are still answers that are outdated and answers that are just plain wrong. The greater the experience needed to answer a specific question, the more unlikely it is that there is a clear and correct answer given.