Here's a report of the war decisions of several AIs in my recent game. This was a marathon game on a gigantic (200x132 plots) map with 34 civs, using BetterAI 0.41. I'm just going through the relevant civs and describe their decisions:
1. Mehmed
He really played a good aggressive game. As soon as he had Feudalism, he started to vassalize first Churchill, then Sitting Bull, then Suryvarman, then Joao. This would probably have gone on but I had to stop it.
I attacked him and Joao at four cities at the same time. Turned out that Mehmed had a massive army in one of these four cities. So while my other three armies captured one or even two cities each, the fourth army was slaughtered by Mehmed's forces, who then advanced into my territory, quickly capturing one of my cities, and advancing to the next (Thebes). I tried to mount some defenses there, but would've certainly lost the city (which would've hurt because it had the Pyramids). However, Mehmed spent two turns reducing the defenses instead of attacking the city. This was enough time for my other three armies to capture their cities. I then traded one city back to Mehmed for peace. In the end I had won 3 cities, lost one, and prevented a huge army from wrecking my only scarcely protected mainland. I think he made a bad deal by agreeing to my peace offer. I also think that he should've pressed his advantage at Thebes, quickly taking it instead of reducing the defenses first. He had an army of about 50 units there, against my mere handful of defenders, so reducing the defenses first may have saved the lives of some of his units, but also gave me time to move more defenders into the city, and capture cities with my other armies. Mehmed would've been better off sacrificing some weaker units and taking Thebes quickly - not only would that have strengthened his position in the peace talks, it would've also given him access to the Pyramids (and taken it away from me).
Mehmed also had another problem: At some point his research totally broke down. He had been among the tech leaders on the continent when he started his wars, but at some point he seemed to have stopped research altogether. In the battles above, I had riflemen and fought against his janissaries and cuirassiers. I then upgraded to Infantry and attacked again. It took him *ages* to get Rifling - by the time he finally had it, I had totally destroyed Joao and Suryavarman, and captured about 30 cities. Perhaps the maintenance costs for his empire (with four vassals) were eating up his research. He should have been able to direct his vassals' research to compensate for that, but I'm not sure whether the AI does that.
What went good, however, was the way Mehmed assembled his forces. After I defeated his army at my borders, for a while I only met his city defenders. Only after destroying Joao and Suryavarman I met another *huge* force of mounted units. It seems that instead of building those and throwing them at me piecemeal (as it would've happened in previous versions of Civ), he actually assembled a large force to counterattack with it in one single move.
One thing that left me wondering, however, was that he never attacked Tokugawa. After vassalizing Joao, Mehmed had surrounded Japan on three borders. I doubt that Tokugawa would've been able to withstand a three-front war, so in a way he was a low-hanging fruit for an already aggressive Mehmed. I don't know whether the AI takes such details into account when planning a war ("I should attack Tokugawa because with my vassals I can draw him into a three-front war"). I also don't know the relations between Mehmed and Tokugawa at this point, they may have played a role too, but I don't think they were excellent, the two had diferent religions after all.
2. Napoleon
Napoleon was boxed in badly, but at some point he managed to pass through Suryavarman's territory and declare war on India (which lay on the other side). He took two cities from Ashoka, who quickly capitulated. However, 30 turns later (which is not much on a Marathon game) he revolted and became independent again.
This means that at the time when Napoleon accepted Asoka's capitulation, Asoka must have been pretty close to the two 50% thresholds already which allow forced vassals to break free. The question is: Does the AI take this into account? Should Napoleon have pressed on and perhaps captured another city, assuring that Asoka wouldn't have such an easy time jumping over the 50% threshold?
Of course, it might have been a smart move by Napoleon if he didn't see a chance to win any more cities from Asoka. Napoleon couldn't reinforce his army quickly (his reinforcements would have to cross the whole Khmer territory again), so accepting a short-lived capitulation would have been preferable if the other option was to be crushed by the backlash. the question however remains whether the AI is perhaps too willing to stop attacking and accept a capitulation if the new vassal is very close to the threshold that allows him to break free again.
3. Suryavarman
Suryavarman played a very strong start. He was the first of all 34 civs to start a war, and in this war he captured four cities from Charlemagne, including the holy city of Christianity, the first religion founded in the game. It looked as if he saw a weakly defended holy city and capitalized on the opportunity, which would've been rather smart.
What I'm not sure about is whether it was a good idea of him to leave Charlemagne two cities. I've seen this on the other continent as well, where Willem and Stalin were reduced to two cities each. The AI seems reluctant to totally destroy another civ.
I'm not totally sure about this (I don't ever read strategy guides, part of the long-term fun that I have with Civ comes from figuring things out myself), but I think that in some situations it will be better to destroy another civ if you can. By destroying the other civ, you at once get rid of any "foreign population" unhappiness in the cities you captured. You can also draft (and whip?) sooner in them since the foreign part of the population will be removed from the calculation of ethnicities in a city - for example, if you had 10% of the population in this city already, and the home civ had 90%, then you won't be able to draft there for a long time - unless you eradicate the other civ, in which case the city is 100% yours and you can draft as soon as the city comes out of revolt, which is a handy way of making use of population that would have starved otherwise. This is a very powerful method for fueling further expansions, but it only works if you're eradicating your enemy. You also don't get higher maintenance costs (as you would for having a vassal) and you don't have "culture fights" (as you would if you made someone your vassal - although your city will keep its fat cross under its influence, it's still susceptible to revolts as long as your vassal has much more culture there).
On the other side, not eradicating the enemy means that you don't have to pay maintenance for more cities, and if you make him your vassal, you also get bonus happiness, can direct his research, and have an automatic ally for future wars.
Still, if the enemy only has two cities left, then he usually won't be able to provide many units or have much research. And the happiness you get in your cities is countered by the unhappiness you get in the conquered cities. So it seems to me that eradicating the enemy is preferable in many situations, but the AI doesn't seem to do it. In my game, it only happened with one civ (Charlemagne, who was already down to two cities, was destroyed by Gilgamesh, who declared war on his master).
4. Gilgamesh
Speaking of Gilgamesh, here's something strange. Towards the end of the game, Gilgamesh was building a *massive* military force, he was leading the power graph. At one point, he finally declared war on his neighbour Zara Yaqob. Gilgamesh had his vassal Wang Kon at his side, while Zara Yaqob was joined by his vassal Mansa Musa, and also by Saladin and Shaka.
I was surprised that I only got messages about Gilgamesh losing cities. He lost three cities in the first couple of turns after the war started, which was odd considering that he was leading the power graph and had started the war, so he shouldn't have been unprepared. I didn't see him achieving a single victory.
At that point I decided to invade the continent and subjugate Zara Yaqob, Mansa Musa, Saladin, and Shaka. So I took a lot of pressure from Gilgamesh. Afterwards, I decided to go after Gilgamesh myself, who was still showing strong on the power graph. However, my forces simply sliced through his cities, although he had the technologies for infantry and tanks. But I saw only a single tank, and his defense forces seemed way too small for his position on the power graph. I had reduced him to three cities when I found the reason: At one remote island at the edge of an archipelago, he had assembled a force of 22 destroyers, 22 transports, 20 infantry units, and some cavalry.
Now this force would've been nice in a counterattack against me, but he never used it. The units simply sat at their island until Gilgamesh capitulated.
I'm wondering about two things:
a) Why didn't he use this force against me?
b) Did this remote force perhaps lead him to overestimate his actual power on the mainland (based on the power graph), so that he declared war on Zara Yaqob who was actually *stronger* on the mainland?
5. General Observations
5a) Unit upgrades: The AI was still a bit reluctant to upgrade its units. When I finish researching a key military technology that lets me upgrade my units, I usually halt research for a couple of turns, throw everything into making money, and upgrade my forces. This usually pays off since it makes me a less likely target of attacks, makes me better prepared if I'm attacked nevertheless, and gives me a better position in diplomacy. The AI however doesn't seem to do this, or at least not aggressively.
5b) AI production: In the industrial era, my production absolutely skyrocketed until it was several times as large as that of the next best AI. Many factors contributed to this (playing Frederick, having Assembly Plants, managing to build the Three Gorges Dam, having a large empire), but I still got the impression that the AI didn't give production a high priority. Due to this, they were never able to build a meaningful amount of tanks - tanks are a very powerful unit which can't be gotten by upgrading others, so tanks are the point where having high production translates directly into military power. However, the only two enemies I fought that had Industrialism managed to field exactly one tank each (Gilgamesh may have had some more since he had this war against his neighbors before I reached him).
5c) Tech Trades: The AI doesn't see that making a lot of "bad" trades can add up to a substantial advantage. Example: An AI could trade a technology to five other AIs, but doesn't do so because none of those have anything to offer except 1000 gold each, and the technology is worth much more. So the AI won't trade its tech. Then the human player comes along, researches the same tech, and trades it to all five AIs for 1000 gold each. As a result, the human player now has 5,000 gold more, and probably better diplomatic modifiers with five other civs, whereas the AI who researched the tech first has neither.
I'm aware, however, that changing this would be a two-edged sword. It might all too easily come across as "The AIs are giving themselves easy trades in order to screw the human over". And I'm not even sure if it's *really* wise to trade technologies at a discount. But I get the impression that in my games I rely a lot on this technique to catch up in techs, or make money fast, and it seems to help me, so I wonder whether the AI is a disadvantage by not using it.
5d) Resource trades and Corporations: How does the AI value resources that it trades to me for use in my corporations? I get the impression that the AI gives such resources very little value. They usually give me several resources for one of mine, or they add gold per turn to the deal. This is not in proportion to the use I get from these resources when I have an empire with well-spread corporations - or am I missing something?
As usual, thanks for reading, and if you have questions, just ask.
I hope that the AI questions I have raised can be answered.