>
I think we can get away with the fully-automated approach for most of the mapsEven though it might look nice to outsiders, I firmly believe you can never get representations of cultural and human data really right if you're doing them automatically.
Such data will always be full of quirks, and representing it will always need some tweaking if you want to get it perfect. Say, moving a label a bit further left so it doesn't hide a monument, displaying a street wider than its actual designation because it's an important way, not showing at all another path because it's disconnected from the rest of the network and really not that important (or displaying the more minor path that connects this one to the rest even though its designation means it should be hidden).
I really think a lot of people who make maps do not acknowledge the fact that human data is not something that is completely normalized, not normalizable, and sometimes might not be able to be processed with a set of rules, however fine.