Civ4 AI Survivor: Season Three

[Brian Shanahan]

I am just after finishing reading the cap to the season, and I am wondering if we, for a competition, instead of going with an open draw in the first round (where anybody can end up in the same game), have enough data to run some sort of seeding system?

It would likely make the games somewhat more predictable I think, but it may make for a much stronger playoff field.

I think it's a lot more fun to have it as an open draw, especially when you consider an extremely low amount of possible combinations have been played. And it feels to me like we're still low on overall sample size, especially given this year had a fair deal of upsets.

Perhaps more importantly, I think a seeding system might kind of force people to stay kinda same-y in where they are, given the matchups they're being put in. And some unpredictable matchups are fun.

Oh, undoubtedly open draw is more fun, hence why I removed reference to uising seeding for Season 4 when typing. But it might be interesting to run as a side competition to see how random all-AI games are.

[antisocialmunky]

Just for fun, what would the ultimate AI be? fin + imp, spi, or org?? The leader would be a low-middle peace weight, medium-low aggressive, medium unit builder, with growth and science flavors ish?

I think Mansa Musa + ___ is probably the best leader.  Either Mansa Musa or Byzantium or Mansa Musa of Ottomans would be great.  I think he also has a natural pairing with England.  Basically you give Mansa Musa an extra happiness UB to get extra pop to snowball faster and a UU that is good against previous eras and he becomes much less of a punching bag during the era he gets invaded in (around Knights).  Not only will he get to the UU faster than the rest of the field, the UU can shred anything in the previous era.

I am just after finishing reading the cap to the season, and I am wondering if we, for a competition, instead of going with an open draw in the first round (where anybody can end up in the same game), have enough data to run some sort of seeding system?

It would likely make the games somewhat more predictable I think, but it may make for a much stronger playoff field.

I think it's a lot more fun to have it as an open draw, especially when you consider an extremely low amount of possible combinations have been played. And it feels to me like we're still low on overall sample size, especially given this year had a fair deal of upsets.

Perhaps more importantly, I think a seeding system might kind of force people to stay kinda same-y in where they are, given the matchups they're being put in. And some unpredictable matchups are fun.

I think it might actually be interesting to do a tournament seeding system where the AIs that made it a certain amount through the bracket get seeded evenly to the next season's groups.  That way you avoid groups of death.

[Sullla]

Here's one of the side projects that we discussed on the Livestream related to AI Survivor. What would happen if we replayed some of these same games with "Preserve Random Seed" turned off? Would the games play out in similar fashion, or would the results be completely random? I decided to find out using Game Four from this season as the starting point. Here's the link to Civ4 AI Survivor: Alternate Histories #1.

[Dark Savant]

Here's one of the side projects that we discussed on the Livestream related to AI Survivor. What would happen if we replayed some of these same games with "Preserve Random Seed" turned off? Would the games play out in similar fashion, or would the results be completely random? I decided to find out using Game Four from this season as the starting point. Here's the link to Civ4 AI Survivor: Alternate Histories #1.

I think even that's reasonably convincing evidence to say that the results aren't at all "complete random".  I think it's a good deal more predictable than baseball for an individual game, which is an example you've used before.

[Fluffball]

I would be VERY curious to see how a bunch of replays of the English games worked. Lizzy and Vicky are insanely strong techers, and their blown victories were... memorable.

[yuris125]

I'm more curious about the game De Gaulle won tbh. That result has to be a fluke

[Kuro]

Do you still have the games from previous seasons? Because, if so, I would nominate Season 2 Game 5 as one to see: Shaka and Tokugawa winning HAS to be a fluke, right? Or are the high amount of weak leaders (Willem, Frederick, etc) meaning they actually come out ahead and this is a situation where those kinda warmongers might usually pop up?

[Sullla]

Yes, I do have the savegame files from previous years and that might be a game worth investigating because the field of AI leaders was so terrible by every metric we have available. It graded out as the worst game ever following Season Two, and then again as the worst game ever following Season Three. If I find time to keep doing more of these investigations, that game would be a good candidate.

[Kuro]

Yeah, I think it's very unique as a game and so would be very interesting to look at.

By the way, if I may make a recommendation for your point system, I think that surviving until the end of a game should give 1 point, and then bump up 1st/2nd to 6/3 points respectively to keep the point differential the same for those. My logic is that there is value in moving on to the Wildcard game/surviving compared to elimination (See: Zara in Season One, who went to the Wildcard and then was dominant and Kublai/Pacal this year) and that the AI Survivor competition should punish people who repeatedly fail to even survive. Sitting Bull and similiar leaders who have done nothing but be eliminated should be below AI who at least display the capability to move on and make something in theory happen such as Washington and people like Brennus shouldn't be below De Gaulle: Brennus has been in the playoffs every year but is below De Gaulle because he has a single win despite otherwise being a horrible leader because being eliminated isn't punished.

Plus, there have certainly been third place winners deserving of some kind of points because 2nd and 3rd have been very close to advancing, and it feels like they should get something as well, even if in this case, it is just a point to put them in front of people who fail to do anything at all. The actual differential between 1st and 2nd finishes remains the same, so it should still highly incentivize finishing Top 2.

[RefSteel]

My logic is that there is value in moving on to the Wildcard game/surviving compared to elimination (See: Zara in Season One, who went to the Wildcard and then was dominant and Kublai/Pacal this year) and that the AI Survivor competition should punish people who repeatedly fail to even survive.

I get this, but under your scoring system, round 1 losers who go on to win the Wildcard game would get more points than the round 1 winners who beat them, which is one of the things the new scoring system was meant to fix. Under the current system, S1 Zara and S3 Kublai/Pacal aren't penalized anyway: They get just as many points for gold/silver in the Wildcard game as they would have for a first-round game. Also, surviving a round 1 game as an also-ran nobody (+1 point) and then surviving the wildcard game as an also-ran nobody (+1 point) would be almost as good as finishing second in the opening round (+3 points) which is exactly what we don't want to see!