Sullla's AI Survivor Season Seven - begins today!

[Plains-Cow]

REMINDER:

Sullla says no game this Friday. Will resume next Friday.

 

[NickGabben]

Civ 4 Survivor probably reached its final form with Season 5 abolishing the Apostolic Palace from the competition.



A quick overview of the results after getting rid of that pesky wonder:




Season 5 had an average of 11.1 wars and ended on turn 336.6 (Cold Map Setting)
Season 6 had an average of 11.9 wars and ended on turn 327.5 (Tropical Map Setting)
Season 7 had an average of ???? wars and ended on turn ???? (Temperate Map Setting)

Of the 29 games played so far in the latest seasons, there were 14 spaceships, 7 dominations, 7 culture wins, and 1 diplomatic victory.


Season 7 has flowed with blood, and so far only Shaka has not been totally eliminated by the top 2 winners.
If he ends up in the wild card game by himself, Sullla might just put in 6 Shakas to fight it out

The top 2 finishers in the Wild Card game end up in Playoff Game 2 and Playoff Game 3 I think.
That way the 3 playoff games and the final all have 6 civs.
The most a civ can show up in a season is 4 games if they botch their 1st game and don't die, get 1st or 2nd in the Wild Card Game, get 1st or 2nd in a Playoff Game, and then appear in the Championship Game.

That's a nice summary! Would you happen to have similar statistics for the first seasons, too (with the AP)?

 

[Kaitzilla]

That's a nice summary! Would you happen to have similar statistics for the first seasons, too (with the AP)?

I'll try to make it to make it tonight.


I'm having trouble finding all the records of Season 1.

Civ4 AI Survivor

The spreadsheet in located at the end of the conclusions page, but does not appear to contain any info besides the 2 winners each game and perhaps the 1st eliminated.

I also failed to find the thread for it on Forgotten Realms.
And I was also unable to find any video record of Season 1 on YouTube or Twitch.

I can reconstruct it from Sullla's detailed written accounts of each game I think, but it will take time.

 

[Plains-Cow]

Alternate histories video drop and discussion in this thread:

Sullla's CIV IV AI Survivor Season 7 Alternate Histories Posted!

There was no game today (Sullla on a trip for work), so he posted some alternate histories! Didn't see anyone posting about it, so I made a thread and will do some cross-linking work. Here's the video to discuss!

forums.civfanatics.com

 

[Kaitzilla]

That's a nice summary! Would you happen to have similar statistics for the first seasons, too (with the AP)?

Right, so I got all the info except for the number of wars which is quite difficult to tally up.
If anyone wants to read through Season 1 and find out how many wars there were, let me know.

So, I'll just say that good is better than perfect and post what I got.




With Apostolic Palace (Religious Victory is not possible since the observer cannot ever get a religion)
Season 1 had an average of ???? wars and ended on turn 318.7 (Temperate Map Setting) (Huts enabled, free Agriculture, Wheel, Hunting, Archer)
Season 2 had an average of 14.7 wars and ended on turn 326.6 (Arid Map Setting) (Huts enabled, free Agriculture, Wheel, Hunting, Archer)
Season 3 had an average of 14.5 wars and ended on turn 331.7 (Tropical Map Setting) (Huts enabled, free Agriculture, Wheel, Hunting, Archer)
Season 4 had an average of 13.5 wars and ended on turn 319.9 (Rocky Map Setting) (Huts enabled) (2nd observer civ added to view AP votes)

Without Apostolic Palace
Season 5 had an average of 11.1 wars and ended on turn 336.6 (Cold Map Setting)
Season 6 had an average of 11.9 wars and ended on turn 327.5 (Tropical Map Setting)
Season 7 had an average of ???? wars and ended on turn ???? (Temperate Map Setting)

It hurts deep inside to see Lincoln won a game and Charlemagne won a season.
The first 3 seasons allowed all the deity AI to have free Agriculture, Wheel, Hunting, and Archery.
A nice boost!

**Edit**
Ahh, I see the explanation in Season 2.

More important are the structural changes to the reporting and the picking contest. I won't be writing reports for the AI Survivor games this year. It's simply not possible for me to do another 13 reports while also working full time. I actually think the written reports are better for these games, but it ain't going to happen, sorry. In lieu of that, I'll be streaming the games this time around, and then archiving the recordings on Twitch so that anyone can watch them later. This has the advantage of making the games interactive, allowing viewers to watch in real-time and interact with the gameplay. The downside is that the reports are summaries that can be read in 15 minutes, while the recordings will last for three to five hours. It's a tradeoff. Regardless, we're switching to the video format this year for better or for worse.

We're also going to switch the picking contest entry system as well. Rather than making forum posts at Realms Beyond (and then relying on me to combine them all into an Excel spreadsheet), this year I'll be putting everything online using Google Docs. The current bracket will always be visible online, and to make an entry for each game, contestants will fill out a short Google Forms document. Of course, there's still a Realms Beyond form thread to discuss the competition as well. I will update this webpage as well as my Twitch livestream throughout the season after each game. Here's a big list of the relevant links below. Enjoy the new season of AI Survivor!

 

[NickGabben]

Many thanks! I didn't realize you'd have to dig through the results manually (sorry).

It looks like the AP was good for one additional dogpile, or how were the religious wars counted? As individual DOWs or just once as an "AP declaration"?

 

[Plains-Cow]

Many thanks! I didn't realize you'd have to dig through the results manually (sorry).

It looks like the AP was good for one additional dogpile, or how were the religious wars counted? As individual DOWs or just once as an "AP declaration"?

The AP is gone this season and following seasons. Been gone for a few now. It's considered a cheesy wonder that ends wars and re-assigns cities, making domination almost impossible. It also makes the game "less fun" according to Sullla and the previous viewers. I agree with them.

 

[NickGabben]

Yes, I know that, but my question refers to the previous seasons when it was still in play.

 

[Plains-Cow]

Yes, I know that, but my question refers to the previous seasons when it was still in play.

Every war declaration noise counts as a war. Those are the rules. So however many declarations the AP caused increases the count.

 

[lymond]

I'd say another big factor regarding AP and war count is forced peace. With AP active, the resident could call a vote for peace with X, which often passed. Forced peace resulted in fewer sustained wars than we see post-AP, especially with the warmongers that get stymied by this AP resolution. Fewer sustained wars, and outright first kills, result in more war declarations over the course of a game.

 

[Kaitzilla]

Many thanks! I didn't realize you'd have to dig through the results manually (sorry).

It looks like the AP was good for one additional dogpile, or how were the religious wars counted? As individual DOWs or just once as an "AP declaration"?

Here is a video from the Wild Card Season 4 where "Declare War on the Infidels" passed at the 1 hour 54 minute mark.

It counted as 3 wars it looks like.


Also, the AP actually eliminated a civ in Season 4 Playoff Game 2.
The AP was pretty amazing how many wars it shut down that season.

 

[NickGabben]

Thank you! Yes, that's pretty clear-cut, then. I think I have only watched non-AP seasons, but maybe I should give the older ones a spin.

 

[NickGabben]

I have watched a couple of the season 4 games now and my opinion on the AP highly depends on whether the votes go my way or not. Justinian used it pretty well in one game (probably close to how it was intended by the designers) and Brennus completely abused it in another with stopping wars and reassigning cities. In general, I do agree the game is better without it and you can kind of see Sullla's transformation from being for it "it's part of the game" to being against it "it's stupid how one wonder can completely ruin the game".

 

[Kaitzilla]

Game 5 of 13 is now in the finish pile.

I can hear brains out there scheming, trying to suss out the future.

How about AI tech choice?
Aetryn over on Realms Beyond dove into the code that no one but perhaps Manco Capac was brave enough to challenge with tech flavors.

Civ 4 AI Survivor Tech Valuation (warning, EXTREME MATH)

And the weak Civ Illustrated attempt:

10) Flavors describe what kind of techs a civilization pursues. The flavor strength was converted from numbers to words as follows:

Flavor
2->Low
5->Medium
10->High

There are many factors that determine what technology a civ will pursue such as a random variable, a free tech or free great person (massive influence!), unlocks a religion(bigger factor if favorite), unlocks a unit, unlocks a wonder, unlocks a civic(bigger factor if favorite), the strategy the AI is pursuing, planning a war etc. It is thought that the flavor of the AI affects the tech choice perhaps 33% of the time.

All the calculated values for the various AI based off their flavors and the flavors of the techs can be found here.

Just looking at those results, it becomes apparent that Joao, Lincoln, Peter, and Tokugawa like Education more than the rest of the AI and are more likely to tech it.

Running through a calculation: Hatshepsut has religion flavor of 2(Low) and culture flavor of 5(Medium). Aesthetics has religion flavor of 2, science flavor of 6, and culture flavor of 8. So Hatshepsut calculates Aesthetics to have a value of 2*2+5*8=44. Mathematics has gold flavor of 2, production flavor of 1, military flavor of 1, and science flavor of 10. So Hatshepsut calculated Mathematics to have a value of 0. She will tend to tech Aesthetics before Mathematics.

Since flavors only influence what techs are researched to a limited degree, the usefulness on this info isn't as great as it could be. However, there is no denying that on average Isabella techs religions and Genghis Khan goes after Military Tradition, Rifling, and Artillery when they get the chance.

AHHhhh too much math!

Right, I'll wait for the executive summary.

Next, where do AI settle their 2nd city at the start of the game?
Does the Settler just follow one of the two scouts from the capital or what?

Hmm. Keyboard left and right arrow moves Youtube backwards and forwards 5 seconds.

Season 7, Game 5 (2nd City status)
Churchill - Founded 4 tiles away with 4 visible resources in BFC (big fat cross) [Banana+Gems+Gems+Dye][6 River tiles]
Roosevelt - Founded 4.5 tiles away with 2 visible resources in BFC [Corn+Cow][5 River tiles]
Asoka - Founded 4.5 tiles away with 1 visible resource in BFC [Pig][16 River tiles]
Cyrus - Founded 4 tiles away with 3 visible resources BFC [Corn+Rice+Cow][0 River tiles]
Boudica - Founded 5 tiles away with 2 visible resources BFC [Pigs+Spices][0 River tiles]
Kublai - Founded 5 tiles away with 2 visible resources BFC [Corn+Cow][0 River tiles]

Season 7, Game 4 (2nd City status)
Gilgamesh - Founded 4.5 tiles away with 2 visible resources in BFC (big fat cross) [Cow+Deer][1 River tile]
Qin - Founded 4.5 tiles away with 2 visible resources in BFC [Corn+Cow][0 River tiles]
Wang Kon - Founded 5 tiles away with 3 visible resources in BFC [Wheat+Sheep+Spices][2 River tiles]
Sitting Bull - Founded 5.5 tiles away with 5 visible resources BFC [Pig+Ivory+Dye+Dye+Dye][9 River tiles]
Mao - Founded 5.5 tiles away with 1 visible resource BFC [Corn][6 River tiles]
Joao - Founded 4.5 tiles away with 2 visible resources BFC [Cow+Wine][19 River tiles]
Peter - Founded 5 tiles away with 3 visible resources BFC [Sheep+Wine+Whale][12 River tiles]

Season 7, Game 3 (2nd City status)
Zara - Founded 4.5 tiles away with 2 visible resources in BFC (big fat cross) [Dye+Incense][16 River tiles]
Victoria - Founded 5.5 tiles away with 2 visible resources in BFC [Dye+Spices][0 River tiles]
Ragnar - Founded 4 tiles away with 2 visible resources in BFC [Corn+Fish][0 River tiles]
Alexander - Founded 4.5 tiles away with 2 visible resources BFC [Pig+Rice][1 River tile]
Justinian - Founded 4.5 tiles away with 3 visible resources in BFC [Spices+Spices+Wine][16 River tiles]
DeGaulle - Founded 4.5 tiles away with 2 visible resources in BFC [Pigs+Wine][14 River tiles]


From what I can tell, if the scouts or a Settler don't unfog a resource, they won't consider it for settling.
This was pointed out by Sullla in Game 3 with Zara.

Squinting at the Scouts, it seems like the 2nd settler usually heads towards the 1st visible food resource between turn 1 and turn 2, but not always?

*Edit*
Flavors link fixed in Civ Illustrated.

 

[Thrasybulos]

Regarding scouting and settling... I just don't know.
I thought for very long that where the initial scouting took place was a good indicator of where the second city would be founded, but having replayed a lot of the games from the WB file (so different initial moves for the scouts), it's not as clear cut.
Sometimes the AI will always go for the same spot, not matter what. For instance, HC in S4 Playoff 1 always goes straight for the spot 5N when the ressource cluster is far enough that it's not scouted everytime.
Hannibal always goes for 3N 3W, but you could say it's close enough that scouting is not as much of an issue.
Izzy and Willem, on the other hand, have their settler wander around, sometimes even founding around turn 10! Willem heads south, then turns back and settles 4W. But not always. While Izzy's second city site varies from game to game.

 

[Saxo Grammaticus]

Out of curiosity, what is the general consensus on early vs. late victories? I consider sub-300 to be the gold standard for better AI, with 253 as the lowest I have turned up in tests for this season. I think of 300-350, or even up to 400, as being where most AI on most maps tend to win. From 350-400, but especially upwards of 400, I think of tortured tech rates and geopolitics or good days for bad AI. Does that sound about right?

 

[Fippy]

I always wonder too ~350 is late already?
Maybe for domination it's not that late.

 

[Saxo Grammaticus]

Off the top of my head, 250 is an excellent date for domination, 300 good, 350 the border of competence/incompetence, and anything later likely a heroic attempt to clean up a very broken world.

 

[Kaitzilla]

All this scheming is a lot of work.
What if we just trained an AI to make our pick for us? Something better than random.org
CivGPT


Or maybe the human element can do it in 5 seconds?

The idea is to take the #1 point scorer from alternate histories and then look at the screenshot at the start of the game for that civ.
Maybe there is a pattern in the starting land powering the winning AI over an average of 20 games.

Season 6 Alternate History winning civ starting capital:

Spoiler :

Game 1 (Pericles Alt. History winner with 81 points far ahead of number 2's 31 points)


Game 2 (Napoleon Alt. History winner with 107 points far ahead of number 2's 53 points)


Game 3 (Gilgamesh Alt. History winner with 63 points slightly ahead of number 2's 53 points and number 3's 52 points)


Game 4 (Catherine Alt. History winner with 76 points ahead of number 2's 52 points)


Game 5 (Bismarck Alt. History winner with 99 points far ahead of number 2's 41 points)


Game 6 (Churchill Alt. History winner with 63 points slightly ahead of number 2's 45 points)


Game 7 (Mansa Alt. History winner with 65 points slightly ahead of number 2's 42 points)


Game 8 (Justinian Alt. History winner with 75 points slightly ahead of number 2's 60 points)


Wild Card Game (Huayna Capac Alt. History winner with 83 points far ahead of number 2's 44 points)


Playoff Game 1 (Mansa Musa Alt. History winner with 99 points far ahead of number 2's 54 points)


Payoff Game 2 (Pacal Alt. History winner with a crushing 117 points far ahead of number 2's 63 points)


Playoff Game 3 (Augustus Alt. History winner with 66 points slightly ahead of number 2's 56 points)


Championship Game (Huayna Capac Alt. History winner with 67 points slightly ahead of number 2's 47 points from Pacal)
All starting positions were identical, so this was purely on the strength of the AI

 

[Saxo Grammaticus]

Wow, thank you for your analysis! Both here and elsewhere.

At first glance, I thought you had already run these games by an AI

From looking over alternate histories, I have been most struck by how competitive these maps are, where usually there is no decisive advantage. I find a 50-point spread there to be a lot more significant than a 10-point one, to which I would ask for a better metric than explaining the point leader, perhaps an AI-devised multipolar theory...

Clearly, rivers play a role! Intuitively, I also think that distance between capitals could be a useful metric, as some starting positions are going to get forward-settled no matter what. On diplomatic metrics, clearly I defer to you!