Civ4 AI Survivor: Season Three

[Sullla]

Fluffball is right - this was one of the closest games we've ever had in AI Survivor. No spoilers here but several different leaders had a legitimate chance to win, and the ultimate winner prevailed in a nailbiter. Going to be a fun report for me to write over the next couple of days. Thanks to everyone who was able to turn out today and take part in the Livestream. Special thanks to JackDRB who lapped with field the top score in the prediction contest, scoring nine points higher than the next-closest entrant. Here's a bunch of links as we get ready for Game Seven coming up next week.

Next Game: Friday, 22 September 2017 at Noon EST

Overview Season Three AI Survivor Page

Current Bracket and Standings

Game Seven Written Preview

Game Seven Video Preview

Game Seven Picking Contest Entry Form

Overview Map:

The recording of Game Six can be found here. I'm hoping to have the written report finished by the end of the weekend. Thanks all.

[Wyatan]

Small suggestion (probably for the next event as changing the rules mid-event is pretty iffy): since you're editing the civs into a worldbuilder save anyway, why not have the civs begin with their actual starting techs, and only those techs (instead of the deity bonus techs + non-overlapping civ techs) ?
It would allow each civ to play to its actual strengths.

[Alhambram]

I am bit surprised to see Zara getting eliminated, but Julius Caesar did very well. In fact when you did skip AI survivor in 2016, I did a version of AI surivor myself for my own fun in 2016 and Zara was the winner.

And a suggestion for next year, maybe make opening rounds groups based upon traits instead random seed. So one group with only aggressive leaders and another group only financial for example.
In that case every leaders in fact got two chances to qualify for next round, but it might too big and busy to follow since every group would contain 9 or 10 leaders. Also this mean 11 groups in opening round, not sure whether you got time for it.

[Psillycyber]

Wow, game six was craaazzaaaay! Love it!

Just a suggestion: has anyone thought about doing an AI survivor contest with Fall from Heaven II? It could even be just a one-off free-for-all game with all of the civs on a large pangaea map. I don't know if multiple leaders from the same civ are allowed in the game in FfH2, but if not, then maybe players could vote for their favorite representatives for each civ? Or, there could be three 1st rounds each with 1/3rd of the leaders, and then the top third of the finishers graduating to one final round (sort of an expedited tournament).

I think it would be interesting to compare the Deity AI's playstyle against each other in FfH2 against what we've seen in these AI survivor competitions.

[Bobchillingworth]

I think a FFH version would more often than not end up concluding in time victories; the AI wartime pathfinding is so bad and the peaceful win conditions take so long to achieve that most AI will end up getting confounded at some point and spend 200 turns moving 1,000 unit doomstacks around in tiny circles.


I guess it might work for MNAI or something, it's certainly more competent there, although some civs are so much more mechanically dominant than others that I suspect it wouldn't really be much of a contest.


Could still be fun to lurk, though!

[Psillycyber]

Who do you think would be the most "mechanically dominant" civs from an AI perspective? I know that Calabim (gov. mansions and vamps), Lanun (eco), Kuriotates (eco and centaurs), and Hippus (fast-movers) are favorites for players. And Illians are almost OP if you allow them to use their worldspell. Which civs/leaders do you think the AI would use most effectively against each other?

Also, I wonder how much of AI stagnation that I've seen in my single player games is due to not playing on Pangaea. After all, if you play regular BTS Civ on something that is not pangaea, you end up getting pretty stagnant games too, considering that the AI has no clue how to due naval invasions. You'd think that FfH2 would actually play to the AI's strengths since it gives so many new times of collateral, and collateral+doomstacks is the only way the AI knows how to start steamrolling an opponent.

I think it would be fun to watch an AI survivor FfH2 game just to point out all of the dumb, suboptimal things the AI would do. Because FfH2 has so many more mechanics and crazy-powerful slingshots that can only be pulled off with conscious planning, I feel like there's a much larger skill gap between the best human FfH2 players and what AI would do vs. the best human BTS Civ players and what the AI would do.

[Bobchillingworth]

The Clan, no doubt about it. The AI often has difficulty expanding and improving its land in FFH; doubled Settlers and Workers through Warrens resolves both issues. The 10% tech penalty doesn't really matter with Deity economic boosts.


The Calabim are strong in AI hands as well, as the AI will generally build a lot of Vampires, and then have them all summon Specters every turn, plus they benefit a lot from the free production from Manors.


Those are the two which are consistently dangerous under AI control. After them... maybe the Elves (either faction), they won't be stymied by needing BW or whatever to improve their land, and tend to build a lot of PoL, which equals huge swarms of summons.

[shallow_thought]

The variance in FFH would be even larger than BTS. Deity setting might help, but I've lost count of the number of times I've seen AI civs killed by barbs before I could even contact them. Clan obviously score there, but miss out on a bunch of potential goodies from lairs etc.

Could be fun to watch, although I think you pretty much need MNAI and Erebus Continent script. I'm pretty rubbish, so I've actually seen the AI pull off a Tower victory once MNAI was installed...

[Sullla]

Wyatan: that's a tremendous idea! Thank you for suggesting it. I agree that it would be a bad idea to change the setup methodology midseason, but whenever we do another season down the road, we'll have to use the actual civilization starting techs instead of the Deity freebies. That should help out some of the leaders with better civs like France and China instead of penalizing them for actually having good starting techs.

Alhambram: the randomness is a huge part of the overall fun! I don't think it would be as entertaining to group all the leaders together by trait. Besides, every leader has two traits and they would therefore all have to appear twice, and trying to predict the outcome of games with 10 AI leaders would quickly become an exercise in pure chance. It's hard enough to get it right with 6 or 7 leaders at a time. Maybe as some kind of side scenario for fun at some point.

Fall From Heaven: that mod is not my thing, but by all means feel free to go ahead and see what happens.

The Game Six written report is now ready. I'll be busy tomorrow and wanted to get it done quickly before the weekend was over. If you would like to watch a 15 minute video of the critical turns from Game Six, here's the short version of how it played out when the game was decided. Thanks everyone.

[Psillycyber]

I tested spectating a game of FfH2. I just used vanilla FfH2, no mods or AI improvements. The settings were, deity, aggressive AI, raging barbs, standard map size, tectonics pangaea, and a random leader from all civs (except for Auric Ulvin, whom I used for the spectator civ).

It was a promising game. About half of the field was eliminated by turn 287. AIs were quite capable at steamrolling each other. And they were teching decently too. A couple had researched arcane lore. One had researched Commune With Nature. One was working on Pass Through The Ether. Many had finished engineering. A few of them bogged down their tech rates due to spamming dozens of tigers and/or skeletons (Alexis, I'm looking at you), but otherwise the game seemed to be still quite dynamic and up for grabs by turn 287 between a top trio of Decius of the Malakim, Capria/Basium, and Tasunke.

Funny enough, the AIs that got eliminated were, in order:
1. Falamar (by Alexis),
2. Cassiel (by Decius),
3. Cardith Lorda (by a Tebryn/Tasunke tag-team),
4. Tebryn (by a Tasunke/Decius tag-team--what a backstab on Tasunke's part!),
5. Jonas Endain (by Capria/Basium),
6. Alexis (by a tag-team of Capria/Basium, Dain, and Garrim Gyr--Alexis had been one of the front-runners after devouring Falamar, so this was quite a reversal!)
7. Hyborem (by Tasunke)+the Dragon fell to Tasunke's mass of horse archers about turn 260.
8. Einon Logos (by Decius)

The main thing that made me stop at turn 287 was that turn times were getting to be about 30-45 seconds just to process the turn. It feels intuitively like the "standard" map size in FfH2 is more like the "large" map setting in BTS. I think I can cut down on the lategame turn times by using a small map size instead of "standard." The "standard" size was already enough to allow most of the 19 civs to get a solid core of ~5 cities before hostilities, with a bit of room still left over for Hyborem and Basium to spawn.

There are 36 leaders (if you count Decius 3 times...34 otherwise). My plan is to do 3 playoff rounds of 12 leaders (+ spectator civ) apiece on a small map. If a standard map can actually fit 19 civs quite comfortably, a small should be able to fit 12.

Then, the winner + next top 3 in score/turns of survival from each playoff game will be chosen for the championship game, which will also be a 12-leader game on a small tectonics pangaea map.

For placing leaders randomly into the 3 playoff rounds while making sure that civs will not double-up in each game, leaders will be sorted into three pools.

The first pool of leaders is for those leaders who are part of 3-leader civs. There are three 3-leader civs, for a total of 9 leaders in this pool (if you count Decius 3 times). 3 leaders will be randomly chosen from this pool for each game, with the requirement that within each game the three leaders have to be from different civs (so, civ repeats will be re-drawn until a non-repeat leader is selected). Decius MUST be in each game in order to make the numbers work out, so in practice this round of drawing will involve:
1. randomly determine which civ Decius is going to lead in each game.
2. assign the remaining two spots for this pool.

The second pool of leaders is for those leaders who are part of 2-leader civs. There are eleven 2-leader civs, for 22 leaders total in this pool. 8 leaders will be randomly chosen from this pool for the first game, and then seven for the second game, and then another seven for the final playoff game (and likewise, if multiple leaders from the same civ are drawn, the repeat will be re-drawn).

The third pool of leaders is for those leaders who are part of 1-leader civs. There are five 1-leader civs, for 5 leaders total in this pool. 1 leader will be randomly chosen from this pool for the first playoff game, 2 leaders for the second playoff game, and 2 leaders for the third playoff game.

This way, every leader will have a fair crack at the championship. (Actually, Decius, that tricksy impostor, will get three tries in the playoffs. But I think it fits thematically to put him in thrice).

If Decius qualifies for the championship in multiple playoff games, that will not affect the method of choosing who continues on from the playoffs. It will still be winner + next top 3 in score/turns of survival (which would mean that there would simply be fewer leaders in the championship if Decius takes up multiple of those spots across several games). As for which civ Decius will represent if he qualifies in more than one playoff game...if others from the same civ also qualify (such as Malakim Decius and Malakim Varn both qualify), Malakim Varn will proceed to the championship, and Calabim or Bannor Decius will qualify instead. (If a tiebreaker is needed to decide between Calabim Decius and Bannor Decius, whichever version of Decius had the highest score at game's end will be chosen).

If there is somehow a case where only Malakim Decius qualifies, but so does Varn, for example, then too bad for Decius, he will have to miss out on the championship. This is only fair in exchange for getting three tries.