Civ4 AI Survivor 2: The AI-stralian Outback

[spacetyrantxenu]

Also I'd vote for a balanced map even if its just the same one for every game - I'm interested in the AI personalities rather than the vagaries of the map script.

This is probably the most important part in setting up a true test of how the AIs play. I think last time around there were some maps where the human/observer civ was moved out of the map, leaving a gap of unclaimed territory for the human's expected neighbors to benefit from. It might be better to roll a map for the intended number of civs and then edit in the human civ after the fact in the WB file. It takes a few more minutes per game to set this up but would likely give a more "fair" contest for the AIs to win by merit/lose by ineptitude rather than having the map make as much of a difference.

A mirrored type of map would be interesting too, just to see how the different civs treat the terrain differently. We humans may crave variety but they AIs will play on what we give them. Thanks for launching a new tournament, last year's was fun.

[Tasunke]

xenu makes some good points. I don't much care whether it's arid or temperate but I do think that temperate might take ... less time to finish ... possibly.

I do think it is important that no civs be separated completely by any body of water from any other civ ... or at least that it is one connected landmass. I would prefer something like a standard/large Mirrorland map ... or barring that, perhaps Pangea.

[Sullla]

Thanks for the feedback. Let me try to answer some of your comments in turn.

Speaker, that sounds like a good idea. The ideal situation would be a setup where anyone could edit an online spreadsheet and add their own prediction contest entry, without the possibility of one person going in and erasing everyone else's stuff. I'm guessing that it's possible to set things up that way.

Westbrook, the goal as always is to generate maps that are reasonably fair. I will try to avoid maps that place AIs in the center, but I won't rule them out entirely, not least because there were quite a few cases where centrally located AIs did extremely well. It's definitely a judgment call. As for why Mansa failed in the final last year, I honestly have no idea. He just didn't expand... Not sure what the AI was doing. They can do some really weird stuff at times.

Old Harry, the Deity difficulty is mostly chosen to make the games go faster. Even with the highest setting with all the tech discounts, the games still take 300+ turns and at least 4 hours, often longer than that. Reducing the difficulty slows things down noticeably. If we went down to Monarch or Prince, there might even be games where no AI won a victory type before the 500 turn limit ran out. Very nice idea, but unfortunately not practical for this setup.

Also, when I did a balanced map for the championship game last year, most people seemed to prefer the random maps. I think it would get a little boring reusing the same map 13 times in a row, or a series of perfectly mirrored maps. I guess I happen to like the randomness, I think it's a large part of the fun. This isn't designed to be a super serious competitive MP experience or anything like that.

Brian Shanahan, yes you can submit a random number generator prediction alongside a real one. Only you though - I don't want 25 predictions from random.org filling up the tables.

I also should have specified initially that all of the maps are single-continent Pangaeas. The AI does not handle overseas invasions very well, and when I tested archipelago maps, it was very rare to see anyone eliminated. Oh, the AIs would declare war, it was just that no one would ever die, and that's pretty boring. The AIs need to be able to reach one another to make this work.

[Wyatan]

Just a quick thought, not for this year's tournament because it's too late, but for a potential 2016 edition: iirc, the AI "personnalities" are driven by a bunch of XML values.
Let each person pick an AI, and submit a customized "personnality".

[Gavagai]

Sulla, may you consider using K-mod for this tournament? It has a much better AI than base BTS.

(As an example, K-mod AI can handle naval invasions quite well, in my games I saw civs totally conquered by overseas AIs.)

[flugauto]

Sullla, would you try "Random Personalities" and "Unrestricted Leaders" this time?

[Rowain]

Sullla, would you try "Random Personalities" and "Unrestricted Leaders" this time?

Wouldn't that destroy the purpose of this game?

[flugauto]

Sullla, would you try "Random Personalities" and "Unrestricted Leaders" this time?

Wouldn't that destroy the purpose of this game?

Not at all. Not at all.

[Twinkletoes89]

I think random personalities would be good fun and make it a little more interesting to bet on.

Unrestricted leaders is a no-no for me though - that adds more shuffling for Sullla for very little gain - plus there's then a real chance of some really awesome or really terrible combos

[spacetyrantxenu]

Random personalities would make it more of a guessing game of "which AI is Mansa roleplaying this time" rather than being able to make any kind of evaluation of how the "Mansa" AI played out its given start. Which, to me, seems to be the point of this tournament -- seeing which AI in a given iteration reigns supreme. If a randomized Mansa (who is really running the Monty AI) bulldozes everyone yeah that's fun, but we won't really get any feedback about it because we won't know what AI conventions were guiding the actions. That's kind of a different contest, like the one Wyatan was suggesting.

As for Gavagai's suggestion of K-Mod, it could be interesting to see smarter AI actions. One question there as I haven't really used the mod except for one quick playthrough, does it change anything other than AI logic/programming? As in, does it change the flavor/preferences from the XML? Because if it does that would go right back to muddying the waters in understanding why a particular leader succeeded - was it because his XML personality works better with smarter AI (whatever that would be...invasions/tactics/whatever) or something else? I think a more fun use for K-Mod if someone is up to the challenge would be to run the same matches Sullla is running in BTS under K-Mod, as a shadow game/tournament. Contrasts would certainly be interesting anyway. But that's a lot of work and a lot to ask.

Anyway,