I think what they need to do with the AI is have it recognize that it isn't human.
Really, from what I understand of pre-Civ 5 hype, they said they were going to make it act more like a human.
Well, with current technology, that's like trying to make an airplane out of concrete. Something will fail.
As far as I can understand it, they gave the AI a set of directives that roughly resembled how humans act (getting wary/angry when people settle where you really wanted to settle for instance). Problem being that the AI doesn't know how to react appropriately to those directives, understand the reasoning behind those directives, or even carry them out very well.
The human typically has problems when you start settling really prime locations near him, get a close-by border town and start building up an army there, or when you're trying to box him in. He might declare war on you, he might watch you a little more closely, he might decide that it's not worth it and go settle elsewhere. But if he knows what he's doing, he can be expected to be reasonable and intelligent about how he carries out the basic idea of "Watch the folks settling near you"
The AI doesn't do that. It mostly only understands that it must hate people who settle on what it considers "its" land.
For that matter, it tends to have really absurd definitions as to what the AI considers as belonging to it.
More or less, the AI will hate you from thereon because you settled in a reasonable spot, if not for a variety of other reasons. You're more or less waiting for when the volcano erupts.
Meanwhile, with the human, you may have had the opportunity to make some kind of lasting alliance with him, at least -- even if you committed a minor infraction like settling an area he was kind of hoping to settle and otherwise prove trustworthy.
Heck, you can't even keep an alliance with the AI because the way it acts is essentially designed to be shifty and backstabby like humans only CAN (but not necessarily will) act like.
tl;dr, there's a fundamental flaw in the AI's goals. Unless we start getting something almost self-aware, or at least significantly more powerful than today's Civ AI, trying to make it emulate a human is a losing game on all sides.