I kind of agree with both angles. My take is that AI generated images need to be more in the average of available information to match expectations of people, who normally have limited knowledge. AI generated images come from a sample of the real, too. (Sorry this feels off topic).