Civ4 AI Survivor Season 5: Game Eight Alternate Histories


Introduction

Game Eight Alternate Histories Spreadsheet

One of the recurring features of past seasons of AI Survivor have been our "alternate histories", running additional iterations on the same maps to see if the same events would play out again. Game Eight saw Willem eventually pull away from the pack with some smart conquests as Genghis Khan rampaged through two Persias before eventually falling apart from overaggression. Was that something which would unfold in each game? This was a topic that called for more investigation with alternate history scenarios. Following the conclusion of Seasons Three and Four of AI Survivor, I had gone back and investigated some of the previous games and found that they tended to play out in the same patterns over and over again. While there was definitely some variation from game to game, and occasionally an unlikely outcome took place, for the most part the games were fairly predictable based on the personality of the AI leaders and the terrain of each particular map. Would we see the same patterns play out again and again on this particular map?

The specific inspiration to run these alternate histories came from Wyatan. He decided to rerun the Season Four games 20 times each and publish the results. The objective in his words was twofold:

- See how random the prediction game actually is. There's a natural tendency when your predictions come true to go "See! Told you!", and on the contrary to dismiss the result as a mere fluke when things don't go the way you expected them to (pleading guilty there, Your Honour). Hopefully, with 20 iterations, we'll get a sense of how flukey the actual result was, and of how actually predictable each game was.

- Get a more accurate idea of each leader's performance. Over 5 seasons, we'll have a 60+ games sample. That might seem a lot, but it's actually a very small sample, with each leader appearing 5-10 times only. With this much larger sample, we'll be able able to better gauge each leader's performance, in the specific context of each game. So if an AI is given a dud start, or really tough neighbours, it won't perform well. Which will only be an indication about the balance of that map, and not really about that AI's general performance. But conversely, by running the game 20 times, we'll get dumb luck out of the equation.

Wyatan did a fantastic job of putting together data for the Season Four games and I decided to use the same general format. First I'll post the resulting data and then discuss some of the findings in more detail. Keep in mind that everything we discuss in these alternate histories is map-specific: it pertains to these leaders with these starting positions in this game. As Wyatan mentioned, an AI leader could be a powerful figure on this particular map while still being a weak leader in more general terms. Now on to the results:

Season Five Game Eight

Game One | Game Two | Game Three | Game Four | Game Five

Game Six | Game Seven | Game Eight | Game Nine | Game Ten

Game Eleven | Game Twelve | Game Thirteen | Game Fourteen | Game Fifteen

Game Sixteen | Game Seventeen | Game Eighteen | Game Nineteen | Game Twenty



(Note : "A" column tracks the number of war declarations initiated by the AI, "D" the number of times the AI is declared upon, "F" the points for finish ranking, and "K" the number of kills.)

I'm not exactly sure what I was expecting to see from re-running Game Eight twenty additional times but it certainly wasn't this! Willem ended up submitting the most dominant performance in AI Survivor history, taking first place 18 times in 20 games along with the victory in the real match that we watched on Livestream. He won the first FOURTEEN games in a row before finally having his streak broken in Game #15 by Cyrus and then taking another loss to Genghis Khan in Game #19. I've run a bunch of these alternate history scenarios by now and I'd never seen anything like this before. The only comparable setup was Season Four Game Three where Wyatan ran the alternate histories and found that Julius Caesar won 17 times in 20 games. Wyatan wrote at the time that he didn't expect to witness such a performance level in any other game and I was inclined to agree with him. But Willem actually scored more victories (18 to 17), more kills (38 to 32), and more total points than Caesar (128 to 119) in that previous game from Season Four. What happened here? How did Willem wind up so incredibly strong on this map?

There were a number of different factors working in Willem's favor that helped put him on this path to glory. For starters, Willem found himself in a very strong starting position that synergized well with his Dutch techs. While there were no seafood resources present to make use of Fishing, Willem's Agriculture tech allowed him to connect his wet rice immediately, and then there were more rice and corn resources present at his first two expansion sites. Willem would inevitably research Hunting for his ivory resource and then Animal Husbandry for his plains cows, connecting two more resources and revealing horses which were also in the immediate vicinity. The Dutch lacked copper but did have iron and stone resources present. This led to a fast Iron Working in most games and Willem building the Pyramids in nearly every match, then switching into early Representation for happiness and the specialist beaker bonus. Four different rivers ran through the Dutch homeland which allowed Willem to pick up the Financial bonus all over the place. He had high food, high production, plenty of luxury resources, and oodles of commerce thanks to his traits. This was a lush region that primed Willem for a strong opening sequence.

Willem also benefitted from a friendly diplomatic environment. Isabella was the closest neighbor to the east but she turned out to be very weak on this map, constantly struggling to expand and connect critical strategic resources. Her poor state in the game that we watched on Livestream wasn't an isolated incident, the Spanish ended up being pathetic over and over again. The presence of high peace weight Darius to the south served as a magnet to draw early game aggression from the militaristic AIs. I can't recall Genghis Khan ever launching an early attack against Willem in any of the twenty games that I watched, he was pulled towards one of the two Persias (usually Darius) in these matches. As for Cyrus, while he was stronger in many of the alternate histories than in the sad spectacle that we observed in the real Game Eight, he had a seafood capital which was a poor fit for Cyrus' starting techs (Agriculture/Hunting) and Imperialistic trait. Cyrus tended to be a third tier power and rarely managed to break out of the northwest corner of the map.

That left Willem plenty of time to do what he did best: tech away in peace. Willem usually ballooned out to a huge lead on the scoreboard thanks to claiming lots of territory with his Creative trait and pushing relentlessly ahead in technology. He was significantly more dominant in most games than in the actual Game Eight where the Mongols were pretty close in score for most of the match. Sometimes Willem would remain peaceful for long stretches of time and then come crashing into a war with superior units. At other times, Willem would join ongoing dogpiles and acquire additional land for himself at an earlier date. His most frequent targets were Isabella and Darius which is the main reason why the two of them finished in the last two spots in scoring. Willem almost always ran over one or both of their territories and snowballed to the point where no one could stop him. Darius was the only other leader who could match Willem in economy... but Darius was eliminated in 90% of these games (he was First to Die fully half of the time) and that left no one to keep up in tech. Even Willem's well-known weakness of avoiding Rifling tech largely didn't matter on this map. Oh, he still ignored Rifling tech most of the time, including going all the way to Flight and Assembly Line in one game without acquiring guns. But Willem was simply so far ahead in research that it didn't matter, with the AIs unable to punish him since they remained at the same level of military technology. There were a bunch of games where Willem jumped straight from medieval units up to infantry and artillery and then slaughtered everything in sight.

Thus Willem seized the lead at an early date in most games and never looked back. He won most frequently by Domination (13 times) while also slipping in a few Cultural and Spaceship victories as well. Several of those Domination wins were attempts at a cultural ending where the Dutch simply picked up enough tiles from running the culture slider to slip over the Domination limit. However, most of them were crushing military victories and resulted in the Dutch accumulating an astounding 38 kills in 20 matches, by far the most we've ever seen. Averaging two kills per game across that many different matches is just unreal. The Dutch could have had even more kills if they hadn't had some bad RNG luck at times in finishing off the last city of their rivals (pretty much inevitable with that much fighting). Willem's dominance also resulted in strikingly short games, with few of them even reaching Turn 300. I typically have the AI simulate 350 turns for these alternate histories and I shorted that down to 325 turns because Willem was winning so quickly in each game. He won early and often in game after game.

Readers are probably wondering what happened in the two games where Willem didn't end up winning. Game #15 was simply an outlier result where Willem had a weaker landgrab than normal and then found himself in an unusual 1 vs 2 situation against Darius and Isabella. This almost never happened as Willem was able to avoid being on the wrong side of these situations in the vast majority of games. But Darius proved to be stronger than normal in this game, and then later Willem become the target of a dogpile as he was partitioned in a 1 vs 3 war against Isabella and Stalin and Genghis Khan. It was a weird game with an extremely late Spaceship (Turn 383) from Cyrus of all people. In Game #19, the Mongols achieved solo conquests of both Cyrus and Darius to become huge but Willem had solo conquered Isabella to become equally large. When Genghis Khan went charging off to fight Washington on the other side of the world, Willem had everything lined up perfectly: he could spend the next 30 turns teching past the Mongols and then win a lategame showdown. But then Willem committed suicide by attacking Temujin while ignoring Rifling tech and that was that. He threw away an easy win for no reason at all, argh!

To be fair, while most of Willem's victories were runaway performances, there were a few close calls as well. Oddly enough these tended to happen at the end of the alternate histories sequence, almost as though Willem were losing energy and getting tired from winning so many times. In Game #18, Darius was inexplicably left alone by the other AI leaders and became a powerful economic rival to the Dutch. Willem conquered Spain and then America at a late date, and he would have lost the game if Darius intervened in those wars at any point in time. In Game #20, it was Washington who emerged as very strong and Willem nearly lost to America in the lategame. Willem foolishly decided to chase after Culture and Washington managed to eclipse him in tech, with the Americans nearly launching the Spaceship before Dutch territory just barely ticked over the Domination limit. Anyway, I think that Willem's true victory chance was somewhere in the 80-90% range for this map which remains an astonishing achievement. The fact that he could win this often from a seven AI field as opposed to the easier six AI field makes it even more impressive.

Now for a look at the individual leaders:

Leader Summaries


Willem of the Dutch
Wars Declared: 56
Wars Declared Upon: 14
Survival Percentage: 90%
Finishes: 18 Firsts, 0 Seconds (90 points)
Kills: 38
Overall Score: 128 points

The man, the myth, the legend. This should decisively kill off any notion that Willem isn't one of the better leaders for AI Survivor purposes along with his dual runs to the Season Four and Season Five Championship games. Eveything on paper suggested that Willem should be a strong leader and after poor results in the first three seasons we've finally seen that in action lately. This map was particularly favorable to Willem for the myriad reasons detailed above. I'm going to keep this section relatively short because I've already extensively covered the reasons for Willem's success but I wanted to note here how aggressive Willem proved to be. He had 56 offensive wars against 14 defensive wars in these games. This was a sign that Willem was rarely being attacked (because the other AIs feared him) and he was engaging in his own conflicts repeatedly (again, because the Dutch had a massive lead in Power most of the time). Will anyone ever be able to top this score of 128 points in future alternate history scenarios? There's always a chance but this was pretty close to maxing out the scoring possibilities under the AI Survivor ruleset. I would be surprised if we saw anyone out-score this performance barring something truly remarkable happening.


Stalin of Russia
Wars Declared: 46
Wars Declared Upon: 19
Survival Percentage: 65%
Finishes: 0 Firsts, 9 Seconds (18 points)
Kills: 15
Overall Score: 33 points

Then there was everyone else on this map. Stalin proved to be the best of the rest with nine runner up finishes and the second-most kills at a distant 15. Stalin benefitted from having a low peace weight which made him one of the less likely AI leaders to get targeted by Willem. They existed side by side quite amicably in most games, notably different from the actual Game Eight where Willem snowballed his position by attacking and consuming Russian territory. Stalin's best games took place when he was able to partition either Isabella or Darius with some of the other low peace weight leaders. He wasn't strong enough to solo conquer either of these opponents and games where Stalin tried to fight alone usually resulted in him stagnating and falling behind. He was smart enough to stay out of suicidal conflicts for the most part but lacked the economic traits to ever gain much of a technological edge.

Stalin was the favorite in the picking contest before the game began but his position turned out to be fairly mediocre in practice. The community anticipated that Stalin would conquer Washington to the east, however that was something that rarely took place. Stalin tended to get sucked into wars with Darius, not Washington, and the American leader was stronger than expected on this map. There were a lot of games where Stalin and Washington fought one another and their duel was rarely decided one way or another until late. In big picture terms, Stalin would have had a good chance at winning these games if Willem hadn't been present. He became decently strong on a lot of these maps but the Dutch always seemed to be bigger and badder and deadlier. Stalin largely played the role of junior partner to Willem and that was probably the best that anyone else could achieve on this map. He was the best choice for a runner up finish and had roughly 50/50 odds to advance to the playoffs; it was a bit of an unlucky break not to do better in the game we watched on Livestream.


Washington of America
Wars Declared: 20
Wars Declared Upon: 41
Survival Percentage: 45%
Finishes: 0 Firsts, 7 Seconds (14 points)
Kills: 7
Overall Score: 21 points

Washington was the only other leader to achieve consistent results from a finishing standpoint. He came in second place seven times and it was nearly always Stalin or Washington who secured the other playoff spot behind Willem. He did this in a completely different fashion than Stalin, however; rather than being an alliance partner for Willem, America simply had the good fortune to be the furthest nation away from the Dutch on the continent. Washington backdoored his way into a lot of distant runner up finishes by avoiding war declarations and/or Willem hitting the Domination limit before his nation could be overrun. These examples included Game #4, Game #6, and Game #10, all of which saw an America which was enormously behind wind up in second place. This was exactly what we watched on Livestream, of course, as a largely undeserving Washington backdoored his way into the playoffs thanks to Genghis Khan losing his mind. Being in an isolated part of the map isn't always very good for outright winning games but it can be very useful for cleaning up second place spots.

To be fair, Washington did have some genuinely strong games as well. The Americans conquered Stalin in Game #5 and would have been one of the top competitors in virtually any other game. Unfortunately Willem actually researched Rifling tech in that game and used it to crush Isabella and then Temujin to reach unassailable runaway status. Washington also came very close to winning Game #20 thanks to Willem inexplicably chasing after culture rather than pushing for an easy Spaceship win. Washington was unable to win either of these matches but he came much closer than most of the other AI leaders. Overall Washington remained his usual pacifistic self, starting few wars and failing to accumulate more than a handful of kills. This wasn't a terribly impressive performance from him and it largely speaks to the weakness of the rest of the field that limping into half a dozen second place finishes graded out as one of the better results.


Genghis Khan Temujin of the Mongols
Wars Declared: 59
Wars Declared Upon: 27
Survival Percentage: 30%
Finishes: 1 First, 1 Second (7 points)
Kills: 14
Overall Score: 21 points

The Mongol leader wound up with the same 21 points as Washington but was much less consistent in terms of results. He had one really good match in Game #19 where Temujin was the outright winner and then a whole bunch of other performances where he flamed out in spectacular fashion. This was absolutely a high risk / high reward gameplan that typically failed to achieve much beyond a whole lot of fighting. Temujin was unsurprisingly involved in the most wars of any AI leader at 86 in total, well above second-place Willem's 70 total wars. (This is even more lopsided than the numbers suggest because the Mongols were eliminated at an early date in many games while Willem stuck around to the finish in practically every single match.) The Mongols fought, and fought, and then fought some more. The most common targets were Darius and Cyrus and it was rare to have a game where the first war didn't start with Temujin invading one of those two nations. If Temujin could find an ally to help out, he could have enough success to achieve an outright conquest and become one of the top contenders on the scoreboard. But it was more typical for the Mongols to get bogged down in fighting and stall out, not least because their warring usually began without having catapults on hand. Although Temujin was only First to Die twice, his poor teching and propensity to make enemies of everyone caused him to get eliminated in the midgame over and over again. This wasn't a recipe for success.

Temujin's gameplan worked out exactly one time in the atypical Game #19. This one resembled the actual Game Eight to a surprising degree as the Mongols managed to pull off a solo conquest of Cyrus in the early game, then successfully invaded and absorbed the lands of Darius. In the real Game Eight, the Mongols didn't start a fight with Willem until the Dutch had already snowballed into an unbeatable land and tech lead. However, in Game #19 Willem threw away a likely victory by invading the Mongols before he managed to achieve an edge in military tech, leading to a predicatable conquest at the hands of the great khan. This was an outlier result though and in most games the Mongols had far less success with their early attacks. Although Temujin scored a lot of kills from being involved in so many wars, he typically bit off more than he could chew and found himself getting smashed by more advanced rivals. The Mongols were unusually strong in the actual Game Eight and rarely enjoyed the same degree of success that we watched on Livestream.


Cyrus of Persia
Wars Declared: 31
Wars Declared Upon: 32
Survival Percentage: 25%
Finishes: 1 First, 2 Seconds (9 points)
Kills: 9
Overall Score: 18 points

Cyrus was similar to Genghis Khan in the sense of having one really good game and otherwise achieving little across the other map replays. His breakout performance was a victory in Game #15 where Willem was torn apart in a series of dogpiles and Cyrus eventually won by a late Spaceship. That game featured massive warring that dragged down everyone's economies and allowed Cyrus to sneak his way to the late victory. Game #15 was an extreme outlier though, not at all typical for how things played out on this map, and Cyrus struggled to accomplish much in his other appearances. He was eliminated in the first ten games that I ran before having better luck in the second group of ten matches. The biggest problem for Cyrus was being stuck next to the rabid dog Temujin as the Mongols launched invasion after invasion in these alternate histories. This stunted the development of Cyrus and tended to leave him with relatively few cities and territory. We also observed on Livestream how Cyrus often let his copper resource get stolen by Temujin and indeed that happened pretty frequently in the alternate histories. Cyrus generally found himself behind the rest of the field and struggled to catch up from a bad opening. Getting crushed by dominant Dutch culture in every game certainly didn't help either.

Many of the problems for Cyrus were also linked to his starting position. He had a seafood-heavy capital which would have been fantastic for someone like Hannibal but failed to synergize with Persia's starting techs and Cyrus' Imperialistic trait. I think Cyrus would have been very strong if he'd rolled the Dutch starting position but it was not to be in this game. For what it's worth, I noted that the Mongols were eliminated in every single game where Cyrus survived to the finish. He needed Temujin to be gone if he was going to hang around to the end. Overall though, Cyrus was the weak leader of a weak nation. He was conquered by Willem in a majority of these games and largely served as a tasty meal for stronger nations to consume.


Darius of Persia
Wars Declared: 5
Wars Declared Upon: 57
Survival Percentage: 10%
Finishes: 0 Firsts, 1 Second (2 points)
Kills: 3
Overall Score: 5 points

Darius was even worse off than Cyrus on this map, largely thanks to having an atrocious diplomatic environment. There was nothing wrong with the capital in this case, as Darius started out with a wet corn and an ivory resource that were excellent matches for Persia's Agriculture/Hunting starting techs. The problem for Darius was his high peace weight rating of 8, a number that painted a gigantic target on his back when bordering Willem (4) and Stalin (2) and Genghis Khan (0). These leaders were drawn like iron to a lodestone and charged off to invade Darius again and again. The eastern Persian leader was the First to Die 10 different times and only survived to reach the end of the game 2 times in 20 games. In my notes, I wrote some variation of "Darius dies in early dogpile" almost a dozen times. It didn't matter if Darius had amazing economic traits and strong local terrain if he was dead before Turn 150 in nearly every game! The fact that Darius founded his own religion a fair amount of the time to cause additional religious tensions certainly didn't help out here. Darius was the pregame favorite to be the First to Die and the community was certainly correct here. It was unusual for western Persia to be the first one to fall, with Cyrus only being First to Die 3 times in 20 games. We saw an atypical result for this category in the Livestream match.

There was exactly one good game for Darius, Game #18 where he finished in a strong second place behind Willem. This was a really bizarre match in which no one attacked Darius throughout the whole early game and allowed the Persians to develop their economy in peace. Darius could have won the game outright if he'd attacked Willem while the Dutch were slowly conquering their way through first Spain and then America. He also survived to the finish in Game #13 but was a last place nonentity in that match. Otherwise it was pretty much a long and painful series of eliminations for Darius across the replays of this map. He managed a piddling three kills (with two of the three taking place in Game #18) and launched 5 offensive wars against 57 defensive wars - yikes. Darius was a total pincushion and the other low peace weight leaders in his neighborhood delighted in stabbing him over and over again.


Isabella of Spain
Wars Declared: 8
Wars Declared Upon: 35
Survival Percentage: 5%
Finishes: 0 Firsts, 0 Seconds (0 points)
Kills: 0
Overall Score: 0 points

As bad as things may have been for Darius, they were even worse for Isabella. Darius could build a credible nation in the very rare situations where he was left alone but Isabella proved to be unable to achieve even that modest goal. She was just useless on this map and always seemed to be a weakling no matter what happened. This was not the result of early invasions, or at least not only the result of early invasions. While Isabella did get knocked out quickly as First to Die five times, it was more common for her to be the second or third AI leader to be eliminated from the board. She simply seemed to do a horrible job of developing her own civilization, chasing after additional religions and cultural stuff rather than expanding properly and constructing infrastructure. Isabella also suffered from something that I never intended while setting up the scenario: a barbarian city popped up with great frequency on top of her nearby copper resource. We watched this happen on Livestream and it took place roughly a third of the time in the alternate histories. Whenever this happened, Isabella was left with no metal resources (despite the copper being a mere four tiles away from her capital) and no way to defend herself from outside attacks. Most of Isabella's First to Die scenarios flowed as a consequence from that pesky barbarian city.

Nonetheless, Isabella was still able to settle that spot without issue in a majority of her games. Why was she so consistently terrible? There were a bunch of seafood resources at her capital which should have synergized nicely with Spain's starting Fishing tech. Isabella's poor performance largely seems to have stemmed from her AI personality as she simply did not prioritize the correct aspects of her civ. Her expansion was underwhelming in game after game and she struggled mightily with research despite having nearby gems resources. Isabella heavily overemphasized religion and often found herself unpopular with her neighbors when they founded their own faiths. But her biggest issue was having the strongest AI on the map next door, as Willem wound up conquering Isabella in almost every single game. They had a lot of border tension between Willem's Creative trait and Isabella's religion; this typically resulted in a war down the road and the Dutch basically always won that conflict. It turns out to be bad news when a historically powerful civ is your biggest rival, what a surprise. Isabella ended up with a goose egg for this scenario, zero top two finishes and zero kills across the twenty replays of the map. She was eliminated in 19 of 20 matches and survived to the finish only in Game #10. This was a map to forget for the Spanish queen.

Conclusions

We often talk about the degree of predictability in the AI Survivor games, whether it's possible to make intelligent picks in the contest or whether everything is basically dumb luck. There are certainly some scenarios that prove to be unpredictable when replayed, such as Game Five of this season having no clear favorite in the alternate histories. However, I've always maintained that it's not all RNG chance and that most of these maps do have AI leaders that are better and worse picks in each category. Game Eight offers powerful evidence in favor of this perspective: there was undeniably a "correct" answer on this map in the form of Willem. He won 19 times in 21 games that we ran under this setup and there's no way that that could be due to chance. The Dutch were unquestionably the strongest and either Stalin or Washington was going to come in second place in virtually every game, with Darius the heavy favorite in the First to Die category. Even though Cyrus may have been the first to get knocked out in the game we watched, Darius was still the right choice to pick and there was a clear favorite in each of the major scoring categories.

So the results prove that we had the most clear-cut, most-dominant AI leader that we've ever seen in terms of winning an AI Survivor game. In theory this should have been a slam dunk for the prediction contest. And yet, here were the pregame predictions for the winner:

Willem was the pick to win from only 22% of the contest entrants! (This graph was created from the first 107 submissions but the percentages remained essentially identical across the full 242 submitted entries. A random sample and the Law of Large Numbers tends to work out that way.) Stalin was far more popular than Willem among the community and even Cyrus was about half as popular as the Dutch leader. How could the community have misfired this badly given that Willem was in retrospect a gigantic favorite to win? I think that it speaks to two general factors. First, it's really hard to get the predictions correct! There are so many things that take place in each game of Civ4 and the AI leaders can act in baffling ways. Many of these games do swing based on the randomness of who declares war on whom and it can be hard to see how things will play out ahead of time. The other factor is that we've been working with limited data and tend to overvalue past AI Survivor results. Willem had poor showings in the first three seasons of AI Survivor so the community undervalued his performance. But that was based off a sample size of three whole games, too small to derive much of anything, and Willem followed that up with 2 firsts, 2 seconds, a third place finish, and an elimination in his next six games. We're up to 7-10 games worth of data for most of these AI leaders by now, definitely better but still a very small sample size. We shouldn't get too caught up in past finishes if it looks like an AI leader has a juicy starting position.

I'll wrap this up by patting myself on the back for predicting Willem to win, Stalin to take second, and Darius to be first eliminated. As the one who runs these AI Survivor games, I would feel pretty bad if I had failed to identify a 90% winrate leader ahead of time. I've had more than my share of terrible predictions but this was one game that I did have pegged correctly. Anyway, I hope this was interesting to read about and thanks as always for following along.