There's a few things that in my opinion will have to be done before we could even implement the AI at a > 0.95 approval score level. IMO what needs to be done is:
- Create a (private) blacklist that will cause the AI not to automatically approve any wikis that include certain words in their description. This is because a description could be very well articulated but be on a subject that isn't allowed by our policies or that is simply controversial and it would be best handled by a human wiki creator.
- Integrate the reason drop-down system and not take into account requests that are declined for duplicate or because database exists. Otherwise a perfect request will be 'bad' in the eyes of the AI even though it was only declined because a duplicate existed.
After these two things are done (and perhaps others if anyone has other ideas) I think we should be ready to take it to a further level by allowing requests that get a score higher than 0.95 to be created by the AI. This can be adjusted over time but should probably remain at 0.95 for a while.