Usually those kinds of ai I've seen tend to win by having flawless micro(they "know" their range with pixel precision, their exact damage output, etc).
But when we look at the macro, and things we associate with thinking and smartness, we see a bleaker picture : they make a lot of mistakes and struggle to adapt their strategy.
How is the research progress regarding AI long term planning and decision making?
We are actually building in a team layer of AI that is essentially the macro, and it can influence the micro of the champion AI. This layer will not have cheats but access to utility functions ("How much approximate damage can be done in the next 2 seconds?") and a memory bank ("their last known position was X"), somewhat similar to humans where you go "he likely backed to base since I haven't seen him in X time and mid isn't pushed up, so probably not ganking mid, and we have a ward on Shelly so they aren't taking that" which we can fudge read accuracy at both the team AI layer and champ layer (like we make an error role that can be configured for different skill levels /types of play).
Right now we are just starting on the foundational bits of this layer, but we want to test this one out during our 2 weeks on PBE test!
Read more