Could it make sense to leverage the large and global user base to help to train the autogen building and tree AI?
At its most basic form, perhaps users could click on an existing building and see the AI’s top few predicted buildings (with its current choice at the top, I imagine). If a user thinks that a different autogen building from the list would be more accurate in its place, a user could select it and effectively vote for it to replace the existing building. Hopefully this could result in a more accurate world - both in the locations that users have “edited”, but also in unedited areas if the AI is able to train itself better due to the user interventions elsewhere.
A more ambitious (but also more difficult) version would be if users were able to make more detailed edits to the trees and buildings, ie not being forced to select from a couple of AI suggested autogen buildings. I guess it would be a bit like having add-ons for the whole world (where edited) automatically downloaded and installed. The downside of course is that it would be more at risk of vandalism or simply conflicting visual styles - there would need to be a robust trust system of some kind in place.
In both above versions, I would of course imagine this “user edited” world version to be optional, a bit like we can toggle on or off photogrammetry currently.
This was just a (half-baked!) idea I had when thinking about the AI and would be interested to hear any thoughts about it