March Madness: Civ4 AI Edition

[Gazglum]

I don't know about base BTS, but I play BetterAI and culture victories are how I most often lose to the AI. Partly because I often feel too conflicted about invading a peaceful empire that is quietly building temples and libraries, so I sometimes retire from the field instead. If there is only one Industrious AI they can really start to take off with their culture points.

[asdfion]

I'm amazed at how long it took Mansa to win. Almost all of my SP experience is with Warlords- did BTS screw with the AI logic to make the game easier or something? I know that by the end the AI development was more-or-less outsourced to Blake. Maybe it's just me wearing rose-colored glasses, but I seem to recall the Warlords AI being much better at developing an economy and much faster at winning by space.

AI bonuses were cut way down with BTS. Comparing Deity with Vanilla (who has Warlords installed?):

2 free worker -> 1
60% food requirement -> 80
10% unit supply costs -> 50
5% upgrade cost -> 50
20% inflation -> 80
20% WW -> 50

[Sian]

Certainly feel that Mehmed should get either a play-off position or one of the extra spots depending on how it pans out

He fought and died honorably, Zara just fought and limped along on the goodwill of Mansa Musa

[yuris125]

But how can he get a play-off spot if he's dead??

(not saying that Zara deserves it or anything, he was near-dead as well, and only survived because the leaders was feeling gracious)

[antisocialmunky]

Its too bad were can just set this up to stream 24/7. That would be hilarious to watch.

We should video the timelines if possible

Edit: Just caught myself up, epic game 2!

[scooter]

On the AI question - my understanding was that Blake felt he improved his BtS AI significantly enough that he could scale the AI bonuses way back and it would still perform at a fairly similar level - so the AI cheats quite a bit less than in Vanilla/Warlords but is "smarter." Whether it's actually more or less difficult, no idea. Consensus seems to be Warlords was most difficult because it still had vanilla-level AI bonuses with Blake improvements mixed in.

I thought that the BtS AI knew it needed to chop more aggressively than these guys are doing, though. Seems to not be the case. It has literally been years now since I've played out a game in SP, though, so I really don't remember what it was like.

[yuris125]

As far as I remember my last game of vanilla, vanilla AI doesn't know how to improve tiles at all. BtS AI at least uses _some_ workers. Not good or anything, and priority of chopping seems to be low - but it does use them

[Sullla]

It's definitely true that the expansions watered down Deity as a difficulty level. The original release version of Civ4 was nearly impossible to beat on Deity. The numbers were worse as asdfion pointed out, but you also have to keep in mind the fact that higher difficulties had lower health and happiness caps back then. Your non-capital cities would only have enough happiness to reach size 3, and they'd only have enough health to reach size 1! And the expansions also added all sorts of things that the AI was never programmed to understand, like corporations and espionage and so on. Deity isn't what it used to be.

But the other thing that no one's pointed out so far is that these games end slower because there's no tech trading enabled. Every AI has to research every tech by themselves if they want to win via space. If they're all flipping techs around, and especially if there's a human player in the field to drive the worldwide tech pace, then the ending dates will finish notably sooner. That said, this game still finished on Turn 322 out of 500 total, nowhere near the max limit. We've gotten too used to humans optimizing their way through the tree, as opposed to AIs who don't have a clue what they're doing.

OK, here are the results from our Game Two predictions:



Congratulations to Tyrmith for turning in the highest score of the game, nailing the winner, number of surviving civs, victory type, and victory date within 10 turns. The popular Mansa Musa pick for first place was a big winner for a whole lot of entries. In contrast, Becko was the only person to pick Justinian for the runner-up spot, and no one had Willem as the first to die. Roosevelt disappointed a whole lot of people by dying second instead of first. Not a bad pick there, just the timing was slightly off. Side note: Twinkletoes and Yell0w correctly picked both leaders who would advance, only they picked them in the wrong order. I gave them half credit for this, 5 points instead of the 10 points that would come from getting them in the correct order. It seemed like they deserved more than zero points there.

Most people picked too high on the surviving civs. Hashoosh had no one dying at all, which was a wee bit off. Diplomatic and Spaceship were the popular victory types, and they were both good selections, as this very nearly turned into a space win for Mansa Musa. Kind of dumb luck what the AI will choose to do at the end of the game. Most victory dates were too early in their prediction, only a couple of people overshot the mark. Congrats to Becko and Brian Shanahan, who were the closest to the Turn 322 finish and only two turns off each.

I'll put the leader draw for our next game and the updated bracket in the next post.

[Sullla]

Leader draw for Game Three is generally on the weaker side. None of these AI leaders stand out as big favorites to me, which should make for a wide open field. Going to be tough to predict this one...


Alexander

Alexander is among the most warlike of the AI leaders in Civ4. He has Aggressive and Philosophical traits, an odd pairing that doesn't lend itself to much synergy. As a Greek leader, he brings the Phalanx and Odeon, with the unique unit axe that doesn't die to chariots being the stronger of the pair by far. Alexander the AI has military and growth flavors, with a heavy emphasis on training units (8/10). He also has a very high aggression rating at 8.6 out of 10, putting him amongst the most militaristic leaders we've seen thus far. Alexander is likely to go to war early and often. Whether that will be a successful strategy is hard to say.


Boudica

Boudica is one of the rare female leaders. Did you know that only 6 of the 52 leaders in Civ4 are women? It's not very inspiring, we definitely could have done a better job here. (The original Civ4 was notably better, with 5 out of 26 female leaders, but Boudica is literally the only woman included in either expansion, 1 out of the 26 leaders added. Bit of a sausage fest.) Anyway, Boudica has the game's most offensively oriented trait pairing with Aggressive and Charismatic traits. I hope the AI Boudica uses them better than poor Spanish Apolyton did in the Intersite Game. Boudica also brings along the Gallic Warrior and Dun from her Celtic civ. Boudica AI has military and growth flavors, with an aggression rating of 8.8 out of 10. She's very similar to Alexander AI in nearly every way, with slightly less focus on building units and wonders. If the two of them share a religion, it could be a match made in heaven.


Isabella

One of the other scarce female leaders, Isabella is infamous for being one of the game's most fanatical religious zealots. Isabella has a solid pairing of Spiritual and Expansive traits, and brings the Conquistador and Citadel for unique items. She's better known for starting with Mysticism tech by virtue of playing as Spain, however. Isabella will likely grab a religion of her choice at the start of the game and then defend it to the death against all comers. Unlike nearly every other AI, Isabella only has one flavor: RELIGION. She is obsessed with the true faith, and hates all those who fail to convert. Isabella also has the highest possible rating for "demand that you convert to her religion" (10/10), and her favorite civic is Theocracy. She's one of the easiest AIs to understand, the crusader obsessed with religion above everything else.


Joao

Joao has the most expansion-focused trait pairing in the game, with Expansive and Imperialistic traits. This has made him a desirable pick in some of our Multiplayer games when the goal is to spread out across the map like the plague. I'm uncertain if Joao the AI will be able to do the same thing, this will largely determine his success or failure. Joao's Portuguese civ brings along the Carrack and Feitoria, water-based unique tools that have limited success on a Pangaea map. Joao AI gets science and military flavors, along with average numbers for virtually everything across the board. He has a 4, 5, or 6 out of 10 in almost every category, with the one exception that he likes to make a lot of tribute and religious demands in diplomacy. Joao is pretty much your average AI, rather bland in how he's set up. Not too much going on here.


Louis XIV

Unlike the last leader, Louis has a very distinct AI personality. He brings Creative and Industrious traits, along with the Musketeer and the Salon from the excellent French civilization. Louis is well known in the Civ4 community for being obsessed with wonders, as he has a 10/10 rating in that category. His AI flavors are culture and military, with more of a lean towards the former. Despite his wonder-whoring ways, Louis can be quite formidable in military endeavors as well, since he has a solid unit emphasis rating and above average aggression (6.3 out of 10). Among this group of generally weak leaders, he's probably the tentative favorite if he can avoid getting dogpiled and rolled over. Expect to see Louis clean up most of the wonders.


Qin Shi Huang

Qin was one of the mightiest leaders in the original version of Civ4, back when he paired Financial and Industrious traits with the Chinese civilization. Sadly the expansions were not kind to him, as Qin was changed to the vastly inferior Protective and Industrious pairing instead. Swapping the best trait in the game for the worst, ouch. Qin does still have the Chu Ko Nu and the Pavillion unique items from the Chinese civ, but loses out on a lot of China's power due to the free Deity starting techs. (China is the only civ in the game to start with Agriculture and Mining techs, widely considered the best pairing. On Deity, all of the AIs start with Agriculture, The Wheel, Hunting, and Archery techs.) Qin the AI has production and growth flavors, along with an emphasis on building wonders (6/10) if not nearly so much as Louis. He's pretty average across the board otherwise, with a modest aggression rating of 3.9 out of 10. Qin is another AI without much in the way of a defining feature.

Frankly speaking, this is a fairly unimpressive group. These are mediocre to weak leaders paired with mostly underpowered civs. In a way though, that helps to create some of its own drama. This game is wide open, with no clear favorites. Anything could happen, and two of these leaders are guaranteed to move on to the next round. It's like that one World Cup group that somehow manages to have all weak teams in it, right next to the Group of Death with far superior talent. With a lot of militaristic leaders, this could well be our first game to end by Domination. How's this one going to play out?

The map, spoilered for size:

Here is our current bracket:



Suryavarman and Mansa Musa have gone through as the top seeds from their games, Brennus and Justinian in second place. Peter and Zara sit in the potential wildcard game for the moment. As for Mehmed, I agree that he played a strong game, but unfortunately it ended in his death. Leaders who can't even keep their civs alive don't get to move on to the next round. Dying means that you're disqualified from the playoffs. I think it's fair to have that as a bare minimum for an "AI Survivor" competition.

Same predictions for Game Three:

Winner:
Runner Up:
First to Die:
Number of Civs Surviving:
Victory Type and Victory Date

Two days to get your predictions in, then we'll post the results on Sunday night. Have fun.

[dazedroyalty]

Winner: Isabella
Runner Up: Qin
First to Die: Louis
Number of Civs Surviving: 3
Victory Type and Victory Date: Domination, T329