Disclaimer: I don't like the term AI but lets stick with it for clarity.
If you write the AI code that can mimic the behaviour of professionals, even in a very controlled environment like a game, not only Riot will be happy to implement this gamemode but most probably the US department of defence will offer you a job.
** Game designers are not able to develop such an AI without using MAJOR cheats.** Bots for example can have 100% skillshot accuracy of course, that's not that hard to do but bots can't predict how you will dodge or how you'll flash. Humans can.
Bots can probably somewhat take into account vision (at the moment I am pretty sure they cheat and see the entire map) but it's impossible to give them intuition as to what bush to ward.
What else?
Bots can execute combos, bots can track cooldowns. Bots will not be able to "play the map" or make flashy teleport plays or backdoors.
Just think about it....
20 years ago IBM created an "AI" that beat a professional chess player.
Chess is a really controlled environment where the "map" looks like this:
http://www.chess-space.com/wp-content/uploads/2012/11/CHESS.png
about a week ago Google, with it's multi billion dollar budget managed to create an AI that beat a professional Go player.
Go is a really controlled environment where the map looks like this:
https://upload.wikimedia.org/wikipedia/commons/thumb/2/2a/FloorGoban.JPG/300px-FloorGoban.JPG
The LoL map is still a pretty controlled envionment, it's certainly quantifiable and can be discretised. It looks like this:
http://i.stack.imgur.com/NrnQV.jpg
Imagine how much more complicated it'd be to write an AI that would be able to plan strategies for it.
