They say it's easy to bash, hard to offer solutions. This thread is for solutions.
In order for things to start improving, I am posting some things and maybe people who are in charge will learn something. As a chess master and computer strategy games expert, I know what I'm talking about. I don't know how can AI combat performance improve directly. I do now how to hugely improve it's overall performance via improvement of economical and diplomatical features.
*Important note* good AI =/= handicaps. Deity players please skip this thread if you don't have anything relevant to say. You win on Deity? Good for you. That only shows how much truth is below. Because if an AI that gets such huge advantages manages to lose, it only shows he's totally useless. What is below applies very well for Prince-Immortal so it really needs to be fixed asap.
Problem 1 : Massive Gold Theft Exploit.
I can, and do steal 100.000+ gold during a huge game with 12 civs from the other 11 AIs. With this gold, I buy Markets, Universities, Castles, settlers, workers. The task difficulty of defeating the AI drops dramatically because of this.
Problem 1: Easy solutions
a. When trading, AI will NEVER put ANY value on Gold Per Turn AND it will NEVER put ANY value on Luxuries unless his happiness level is <0.
Why this works: The task difficulty (TD) will increase because of this fix, as players will no longer have the advantage of thousands upon thousands of extra gold during the game. It will BE harder to defeat the AI, even if he will remain the same combat-wise. Reasons not to do what I propose : NONE.
There's plainly an exploitable issue, but I think this solution is too strong. Firstly, trading gold for luxuries you don't need is a perfectly valid tactic in itself - it's exploitation that's the problem. Effectively barring it altogether reduces the options available in diplomacy. The AI already has the ability to weigh deals by its diplomatic status; perhaps only allow gold trading with friends (rather than just give them a better deal), or when suing for peace? Correspondingly, the amount of gold a civ will grant should go down - you can currently get 240 for a single unwanted luxury with a friend, which is almost exactly the cost of an early-game Research Agreeement that is plainly far more powerful.
Gold per turn value should be calculated based on the overall amount of gold that would be granted over the lifetime of the deal (so that if a civ will trade you 240 gold as a one-off payment, it would be willing to accept a deal of 8 gpt over 30 turns but no more). This is an advantage to the AI because it makes the player (as long as said player can do elementary mathematics) less likely to declare war before the deal's up, and what's more it will cost the AI less if it plans to declare war itself during the lifetime of the deal. So it may be in the AI's interest to preferentially offer/accept gpt terms.
Finally, you missed strategic resources - the AI should check the units it has access to, and will only be willing to trade (at all, not just for gold) if it detects units or buildings that need that resource (no more "I'm in the Atomic Era - could use more horses"). This calculation would have to ignore buildings (since it will always have access to circus or forge late in the game, but doesn't actually need horses or iron as these buildings will already have been constructed in cities that want them), so it may suffer slightly on this front, but a compromise is needed somewhere.
It should also be programmed to check whether it actually has any of resource X remaining, and should only accept a trade for gold if it doesn't, since until it runs out it's able to support any new resource-demanding units it wants.
Problem 2 : Irrational War Tributes
I can, and do Declare War on everybody I meet, in order for them to give me everything that they have after 10-15 turns, with the sole condition that I am up in the army strenght. In a game with many AIs, this gives me THOUSANDS of extra gold from the Tribute the AI pays me. It also gives me their cities, later in the game, along with strategical resources and luxuries.
The problem is that AI sees me as a threat even if I don't even know or care about where he is and what he's doing.
Problem 2 : Easy Solutions.
a. AI will never offer tribute to the human player, nor accept anything that the human proposes to him. (Later edit : except peace at no cost, of course). (Doing so would only worsen his position, as the human will make the deal and still continue to destroy him after 10 turns.)
Why this works: AI stands at a loss in the system of tribute. He does not accurately understand his chances during the war. Paying things, including cities, can only help to hasten his demise. Especially when making them for absolutely no logical reason whatsoever.
All of which is true if the AI's facing a domination player, but this is not invariably the case - mostly in my games the AI sues for peace after declaring war but then losing said war, and I'll only prosecute it further if the AI is a threat. On the other hand, I won't accept a peace deal that gives me nothing if I'm in a strong enough position to make demands. If they give me what they have I'm less likely to go after them again because everything they give me weakens their position, making them less of a threat/rival and so giving me less incentive to attack them - exactly as the AI is programmed to expect. There are also specific cases where this kind of deal makes sense - if the AI is going to lose a specific city anyway, for instance, and offers you a peace deal for peace + that city and nothing else.
This is an extremely tricky problem, and not one I can recall any Civ game really cracking, but banning the tribute system is not the solution. And, once again, there's a difference between an exploit (such as declaring war repeatedly in order to steal their assets) and a valid strategy (which declaring war to obtain a favourable settlement that will weaken a rival should be). However you design an AI it's going to be exploitable by people determined to exploit it.
G&K certainly strikes a better balance than vanilla, in which peace was either on their (inflated) terms or a complete capitulation on the AI's part. You're right that it still isn't good enough, but it's probably as good as it can reasonably be at this stage - it works fine for people who aren't actively out to exploit the system, and frankly it's not the fault of the AI if you've worked out how to exploit, go out of your way to do so, and then complain that it's too easy to exploit.
If it's more fun to play in a way that doesn't exploit the AI, simply play that way, because nothing you can do to program an AI is going to resolve this kind of issue without banning entire playstyles outright, and that's just bad design (as with Civ IV killing ICS by actively making it impossible, rather than by making it an undesirable strategy).
I think one solution that will cover a lot of these without AI adjustments is simply reducing the gold bonuses the AI gets at higher levels. It appears not to actually need or make a great deal of use of the gold advantages it has, since you routinely see AIs at higher levels with 3,000 gold in reserve or more. If an AI has less gold, then the player is inherently constrained in what he can demand - if I give this AI luxuries for gold, I can buy a research agreement! Except that I've just grabbed all the AI's gold and none of my friends can now afford a research agreement...
How about adding more diplo options that demand payment by the AI (such as a trade agreement option, as in Civ IV?), so that there are more demands on the AI's gold that the player will benefit from. That way it simply becomes less attractive to extort all the gold you can from bad trades, because the AI won't have enough left to support the good ones.
Problem 3 : Paying Something for Nothing.
Let's face it, 99% of the times, the open borders that you sell for 50 gold is completly and absolutely useless for the AI. He doesn't need it, it doesn't help him, he doesn't use it. Getting 50 gold / AI means 500 gold / 10 AI , means 1 Library or 1 Market bought with money obtained for selling nothing. This can only hurt the chances of success of the AI.
Problem 3 : Easy Solution.
AI will ONLY trade Open Borders for Open Borders and nothing else. That is simple, effective, logical, and sensible. Reasons not to do this : NONE!
Agreed - I have no idea why this was changed since open borders were always mutual in previous versions of Civ. Alternatively, open borders should provide some kind of diplomatic benefit that makes them worth some cost (probably reduced - why is open borders worth 50 while the more useful embassy - which does include a diplomatic modifier - is only worth 25?).
Problem 4 : Bad Trades.
Imagine you're playing Duel against 1 other good player. At some point, he has 2 resources of FURS. He offers you the deal : you give him 240 gold and he gives you 1 FUR. Do you accept? If you are like me, (good player) , You immediately click REFUSE.
He gets something, but loses nothing. This is a BAD trade.
During the game, you can make TENS of THOUSANDS of gold by selling luxuries. Who profits? It should be obvious. The one who gets to buy Universities, Public Schools, and the like with the money obtained from the deal.
The only exception is if you need that because you a. have unhappiness. and b. will go into "we love the king" . AI already cheats his way to having happiness so he does not need to make such bad trades.
Problem 4 : Easy Solution
a. AI will only buy luxury resource IF : 1) he has unhappiness. 2) He will have "We love the King" in the CAPITAL.
See above. Does the AI benefit from Golden Ages? I recall reading that it doesn't, but if it does that complicates the equation. I don't think restricting it to the capital is a good plan - trading for We Love the King is a good plan anywhere, just reduce the gold he's willing to pay. As in:
1. Do I have unhappiness? Will pay X
2. Do I have We Love The King? Will pay Y. Will pay Z (higher) if same resource needed for We Love the King in two or more cities.
The AI should be programmed always to offer a lux-for-lux trade first, if possible, and only to offer gold if that is refused.
Also, the AI knows when the player has severe unhappiness (hence the contacts "I can hear your people wailing from my empire"), so it should actively offer luxuries to the player (if not hostile) for gold or luxuries.
You present interesting ideas, but mostly too narrowly-focused on your particular (and aggressive) style of play, and consequently you suggest changes that are too limiting for other playstyles or for players not actively out to exploit the system.
So Hwacha is bugged (gets +200 vs cities) when upgraded from the catapult. So you are saying that instead of it being fixed, it's better for people to just not use it anymore. Correct?
While it's not ideal, in actual fact since learning about that I do refuse to use Hwach'a against cities because it's exploiting a bug. Of course it needs to be fixed, but this is a simple binary issue - a feature that was not intended has found its way into the system and can easily be removed. The AI is not faced with binary situations in the cases you outline above, and your proposed binary solutions are consequently not appropriate. It's appropriate to remove the city attack bonus from the Hwach'a because it can be exploited. It's not appropriate to remove gold trading from the AI because it can be exploited - the situations simply aren't comparable.