The problem with the AI is that it too often does blatantly stupid stuff.
I think there are two parts to a strategy AI which affect the players opinion of it:
1. Does the AI do stupid things?
2. Does the AI provide a challenge? (without cheating to a ridiculous level, I would add).
Most strategy AI's struggle to really nail #2 because it is very hard to write an AI that can compete with a player in a complex game. LH is probably about on par with similar budget games in this respect. I think the current (not initial release) versions of both Civ 4 and Civ 5 AI's are noticably better than LH but that is a reasonably high standard as Civ is a big budget franchise.
However some strategy AI's full down at step #1 and that is when the player (and reviewers) really pans the AI. I think WoM/FE/LH have had a chronic problem in this area and that has pulled down the player's opinion of the AI below where it would otherwise warrant. LH is much, much better now, but even so the AI still does some of same stupid things it did in WoM/FE.
Based on Brad's comments in the last month or so its even becoming fairly obvious why this happened. Some of the library functions that the AI relies on just didn't work properly. It doesn't matter how well the AI is written, if the data going in is bad then result is going to be bad (or at least bad sometimes). Based on my recent LH play I suspect that some of the library functions are still broken - the AI shows definite signs of intelligence at times but then it will drop a real clanker.
Needless to say it is still my number one priority for LH. I would love the AI to improve further.