Civ4 AI Survivor Season 4: Game Two Writeup


The second game of Season Four rolled a group of leaders with an uncommonly economic bent: four different leaders with the Financial trait! There are only ten such leaders in the whole game and this particular map ended up with four of them. The group started with Pacal from the pool of protected leaders, one of the rare Financial leaders with a low peace weight. He's more comfortable hanging out with the likes of Montezuma than with Lincoln. Joining Pacal were a pair of Financial leaders with great traits (Willem and Victoria) who had experienced essentially no success to date in past seasons of AI Survivor. Was this the game for them to break through finally? Then there was also Wang Kon, the Troll King, winner of perhaps the most preposterous victory in Civ4 AI Survivor history last season. His trolling had advanced to the point of metagaming the humans watching the AI competition, and I genuinely feared for what he would do this year. We also drew Gandhi into this game, probably the best non-Financial AI leader from an economic perspective. Gandhi either runs away with a peaceful victory or gets crushed, one or the other. Finally we added two more aggressive, low peace weight civs in the last two slots, Mehmed and Louis. They had the potential to shake things up in what otherwise could be a big research lovefest.

The picking community had centered on Mehmed as the favorite, largely due to a ridiculously strong capital with triple gems resources combined with the wide expanse of open space to his east. Pacal, Gandhi, and Wang Kon were roughly similar choices after Mehmed to top the field. On the other end of the spectrum, Willem was the top choice for the First to Die category along with Louis as a more distant second choice. There was immediate consternation from the viewers when Mehmed chose to ignore researching Mining tech (which he didn't start with) in favor of chasing after a religion. Gandhi would end up establishing the first religion via Polytheism (taking the expected Hinduism), followed by a mad scramble for the second religion. Pacal would by the first to Medidation by virtue of a turn order tie with Wang Kon, and Mehmed was only a single turn behind the two of them. Pacal chose to establish Confucianism and both Wang Kon and Medhmed were left out in the cold. The Ottoman leader ended up researching *FIVE* different techs before he finally went after Mining, leaving those gems unimproved for the first 35 turns of the game. It's truly incredible how the AI can self-sabotage itself sometimes.

By way of contrast, Willem was executing one of the better openings that we've seen from an AI leader. He used his starting Agriculture tech to improve the corn resource first, followed by a quick Animal Husbandry for his pigs and horses, and then into Mining to connect his gold resource. This was the impressive result, a capital with four improved resources by Turn 30 and a full nine commerce coming from the gold tile. The Dutch leader also picked a great spot for his second city, sending it due north into a juicy floodplains area where another gold resource was located. With the Financial trait and double gold resources, Willem was easily leading the field in early game research. He then went on to build Stonehenge as well, locking down total border control between the free monuments and his Creative trait. A little bit later, Willem also reached Monotheism tech first and established Christianity to provide the game with a third major religion. Was this going to be the game where he finally achieved something?

The other noteworthy story from the early turns was a failure to expand properly from Victoria. She was sitting on three cities for the longest time and not making much of an effort to grab the territory to the west of her capital. Given that Victoria has the Imperialistic trait, this was a truly puzzling development. As far as we could tell, Victoria was trying to build a series of different wonders and failing to complete them as other rivals finished them first. The marble resource at her capital was perhaps a bit of a poison pill in this respect, repeatedly encouraging the English to chase after wonders in ill-conceived fashion. The net result was that there was a wide open gap of space in the northeast corner of the continent, and Louis was doing a good job of filling that up with cities. I had been expecting the French to be squeezed by their high peace weight neighbors on each side, and it simply wasn't happening because of Victoria's inexplicable struggles.

The first war broke out in this game at a very early date, with Wang Kon invading Pacal on Turn 60. Much of the map remained unsettled at this point, and typically the AIs don't get too aggressive until they run out of land. This would end up helping Louis settle more of that northeastern gap on the continent with Korea focused on its southern neighbor. But unlike so many other early game wars that tend to be self-sabotaging, Wang Kong's invasion had lucked its way into a perfect setup. Korea had a copper resource near the capital and the presence of that copper along with an early Bronze Working tech had led Wang Kon to build a bunch of axes. Notably Korea did not have any horses and therefore couldn't build any chariots. Meanwhile, Pacal did not have a copper resource and therefore couldn't build axes. He instead built lots of holkans, his resourceless unique unit spearman, and actually skipped out on researching Archery for a long time. (Yes, there's a Mayan archer in the above screnshot but that's one of the free Deity starting units.) As a result, Wang Kon opted into the perfect tactical coup: he had an army full of axes attacking a defender who was massing nothing but spears. Uh oh. Even with cultural defensive bonuses, that wasn't going to be enough to stop the attacking Koreans. The city of Lakhamba fell first followed by Chichen Itza and Wang Kon's forces began marching deeper into Mayan territory.

The religious gameplay was beginning to take form by this point and we were seeing the emergence of the first alliances. Gandhi's Hinduism held sway on the eastern side of the map, where he converted Victoria to his cause. This was very much expected given their shared high peace weight scores along with Gandhi's religious obsession. Willem's Christianity had spread north to Mehmed, no surprise there, but somewhat shockingly it also popped up in France and was prevalent enough for Louis to convert. This was bad news for Gandhi and Victoria, as Hinduism would have been the more logical choice of religion for Louis, and instead he was aligned with Willem's more distant faith. Pacal's Confucianism had spread to Wang Kon, which unfortunately wasn't enough to help the Mayans since it didn't take place until after their war broke out. Wang Kon quickly captured the Confucian Holy City and therefore took over as the leader of that faith, making him unlikely to switch. This was the basic religious configuration that would last for the rest of the game, at least until leaders switched to Free Religion much later.

We thought that Mehmed would go after Wang Kon with his first attack, given their peace weight and religious differences. Instead he chose to strike out at Willem on Turn 75, a war that didn't have much effect in terms of territory changing hands, but which did slow down Willem and disrupt his early frontrunner status. By Turn 90 Louis had surprisingly taken over as the score leader, and he made a wise move to continuing snowballing that advantage by declaring war on Victoria. The English had failed to expand beyond four cities in totally inexplicable fashion, and Vicky's territory was ripe for the plucking. She was able to hide behind city wall defenses briefly, but it didn't take long before Louis researched Construction tech and brought catapults into the mix, with the siege units sounding the death knell for Victoria. She needed someone else to intervene into the war or it was all over.

There was no help coming for Pacal in the other major war taking place. He had an iron resource at his capital yet stubbornly refused to research Iron Working, going all the way up to Construction without ever finishing the tech that would save him. It was incredible to watch Korean axes and swords effortlessly hack down the defending holkans, archers, and chariots without ever having the Mayan version of those same units make an appearance. One city after another was taken by Wang Kon in ruthless succession. The capital city of Mutal fell on Turn 98, and Pacal made his exit on Turn 104:

We had a question on the stream about whether this was the earliest exit ever in Civ4 AI Survivor, and after going back to check the previous seasons, I can confirm that it was indeed the earliest. Pacal's defeat actually shattered this particular record, which was previously held by Augustus in a playoff game from Season One with a Turn 106 exit. It was hard to believe that one of the best performing AI leaders from previous seasons could be eliminated so quickly, but this truly had been an exceptional series of circumstances. Wang Kon had exactly the correct unit mix on attack and Pacal's resource situation left him with absolutely the worst unit mix possible for defending. And it's not as though the resource situation from the map was unfair or anything, as Pacal literally had iron at his capital and refused to research the tech needed to connect it. Between the early demise of Pacal and Mehmed sitting in fifth place, the picking contest entries were not off to a great start.

We were now curious to see if Wang Kon would intervene in the war between Louis and Victoria in time to save the English from destruction. Willem and Mehmed also signed a treaty in their own pointless war at this time, freeing up the two of them for action. Gandhi was Victoria's best friend and the most likely candidate to help her out, except that he was founding a whole bunch of minority religions and building the Apostolic Palace. The Indians seemed happy to focus on infrastructure forever until they won a peaceful victory condition. Thus no one came to save Victoria, and she was eliminated as well on Turn 123:

Note that this was the third-earliest elimination in Civ4 AI Survivor history, which would have been second-fastest Pacal hadn't been booted even sooner in the same game. What a weird situation. Louis and Wang Kon were now the two leaders topping the board, and between them Louis looked to have the better situation. His next move was almost certainly an invasion of Gandhi, and a conquest of the Indians would make France the undisputed runaway AI. What Gandhi needed was for Wang Kon to attack Louis, tying up the two strongest AI leaders in their own conflict and clearing a path for India to chase after a cultural or space win. Wang Kon and Gandhi effectively needed one another to survive since they were the only two high peace weight leaders remaining. If one of them were defeated, the other one would almost certainly be eliminated as well.

Fortunately for Gandhi, Wang Kon launched the hoped-for invasion of France on Turn 128. If Wang Kon had chosen to attack Mehmed or Willem instead, this could have been a very different game. Wang Kon was able to capture the French border city of Orleans just as Willem initiated his own puzzling war:

The Dutch were attacking the Indians of all people. This was one of those conflicts that makes sense from a numerical perspective (big peace weight difference, religious tension, etc.) but none at all from a strategic outlook. It was hard to see Willem having much success given the massive supply lines that he'd have to be working with and the ease of Gandhi's units fighting within their own territory. With that said though, Gandhi struggles to build enough units due to his pacifistic tendencies, and any fighting would pull Gandhi away from his relentless cultural pursuits. Meanwhile, Wang Kon's early success in his invasion of Louis was turned around when Mehmed was pulled into the conflict on the side of the French. Mehmed had not been plotting war and took a few turns to gather his forces, but when they were grouped together it was enough to capture a border city from Korea. Now Wang Kon was stuck in a two front war and things didn't look so good for the high peace weight leaders. Louis took his border city of Orleans back again along with the overall score lead, and French units began pushing into Korean territory.

For his part, Gandhi was struggling against Willem in defiance of all tactical logic. The Dutch managed to capture Vijayanagara away from Gandhi and take control of one of Gandhi's four Holy Cities, then began pushing on Gandhi's capital city of Delhi next. The Indians appeared to be on their way out, Gandhi done in by the same flaw that always cripples his games: not building enough units. However, the diplomacy began shifting over the following turns in Gandhi and Wang Kon's favor. First the Koreans were able to secure a badly-needed peace treaty with Mehmed. The Ottomans had picked up one city for their troubles and likely should have continued the war against a reeling Korea. This treaty allowed Wang Kon to stabilize against Louis and turn that conflict into an extended draw. Then Gandhi went to work diplomatically and induced Mehmed to declare war on Willem! This was the critical turning point of the game, relieving the pressure on India and forcing Willem to face off against his closer northern neighbor instead. Gandhi broke the siege of his capital, recaptured Vijayanagara from the Dutch, and even used the Apostolic Palace to cheese a city away from Louis:

Gandhi was absolutely not the "rightful owner" of Chartres, a city that he had never owned at any point in time, and yet he was able to use his cultural influence over the city to flip it to his control. That wonder is really stupid sometimes. In any case, we were looking at a close four-way race for the lead between Louis, Wang Kon, Willem, and Gandhi. The other three leaders were tied up in warring at the moment, allowing Gandhi to go back to spamming out endless infrastructure and religious buildings across his empire. His culture was already starting to become a concern, with three cities approaching the 5k mark. In the ongoing wars, Louis and Wang Kon appeared to be an even match for one another, continuing to clash with big armies over the city of Wonsan without either side gaining an advantage. Off in the west, Mehmed successfully captured a core city from the Dutch with the initial invasion, and nearly took a second one before falling just short outside Utrecht. This would be the high water mark for the Ottomans, who had a smaller empire and were weaker economically than Willem. Eventually the tech edge of the Dutch would make its presence felt, and the results would not be pretty.

By Turn 200 the stalemate between Wang Kon and Louis was beginning to tip towards the Korean side. This was so gradual that it was almost imperceptible at first, starting with Wang Kon decisively breaking the siege of Wonsan, and then continuing with the capture of first one, then two French border cities. The long score lead that Louis had enjoyed began to slip, as he fell back into the pack with the loss of these border fortresses. It was all the more impressive that Wang Kon was able to pull this off despite crushing war weariness in his cities:

That was the Statue of Zeus at work there, doubling the war weariness that Wang Kon had accumulated to truly absurd levels. Only the presence of a bunch of resources that India was trading kept Korea from a total collapse here. Things were pretty bad in France as well, and both civs couldn't grow their populations due to the massive amounts of unhappiness from their seemingly endless war. This was allowing Gandhi and Willem to surge ahead on the scoreboard, and the two of them increasingly seemed to be the top leaders in the bunch. Gandhi was building the Statue of Liberty and all of the Modern era wonders, racing out to a clear tech lead. Any fears that he would be overrun militarily were gone now, with the presence of first Rifling and then Assembly Line techs. Willem was a little bit further behind on the tech tree yet far ahead of the other leaders. He was using some pretty lousy tactics and repeatedly failing to capture Ottoman cities, but eventually the presence of rifles and cavs against janissaries had to be enough to swing the odds. By about Turn 250, the Ottomans were in the process of disintegrating:

With their fall of the Ottoman capital, those infamous triple gem resources passed under Dutch control. Both Gandhi and Willem were about 1000 points ahead on the scoreboard as compared to the rest of the field and looked to have the two playoff spots locked down. Mehmed was a dead man walking unless he could get a peace treaty of some kind, and Willem wasn't in the mood to show mercy. After the war sat in a static position for long turns on end, the final city captures came quickly. It was all over for Mehmed on Turn 263, an ignominious end for the pregame favorite. We keep thinking that Mehmed is a good leader season after season and it never seems to materialize for him. Meanwhile Gandhi was accelerating into the endgame of a cultural victory by now. He was already looking at a victory date before Turn 300, and then decided to speed things up further by turning on the cultural slider:

886 culture/turn in the weakest of his three Legendary cities had the finish date coming even sooner. Just look at all those juicy watermill tiles and the use of the Build Culture option to convert production into culture, sheesh. This could have been a human playing the game. Now the question was whether Wang Kon could finish off Louis before time ran out on the game via Gandhi's impending cultural victory. It didn't look like there would be enough turns remaining, not with Gandhi running the culture slider, however Wang Kon ended up surprising us. He used Open Borders through Gandhi's bloated cultural territory to move quickly to the remaining French cities, and Korean infantry proved to be sufficient to cut through defending French rifles. Louis was eliminated on Turn 284, and Gandhi wrapped up his victory three turns later when his third city hit 50k culture:

For a game with four civs eliminated, we had a shockingly low number of war declarations: only seven in the whole contest! These AI leaders had chosen to keep their wars going for long turns on end rather than signing peace, with the best example being the Korea vs France war that started on Turn 128 and didn't end until Louis was eliminated. All three surviving leaders had reason to take pride, with Gandhi capturing yet another first place finish and Willem grabbing his own first ticket to the playoffs. This was one of those unusual games where the score leader didn't win the victory as Willem had a significant score lead at the finish. As for Wang Kon, he successfully defeated his gamelong rival of Louis and scored two total kills. We'll get a chance to see him troll the rest of the field again in the Wildcard game, where he looks to be one of the more dangerous competitors.

The overarching story of this game was Gandhi successfully getting to run his religious-heavy pacifistic strategy. We have pretty clear evidence of this from four seasons of AI Survivor: when Gandhi gets to avoid wars and build in his corner of the map, he always wins the game. When he gets roughed up, he falls apart like a paper tiger and quickly exits the game. Gandhi came extremely close to defeat in this game, and when Louis took out Victoria at an early date I thought that Gandhi was very likely to be next. Instead, Wang Kon saved Gandhi by attacking France and kept Louis busy for literally the entirety of the game. Gandhi only faced a single war all game, and he was losing that one to Willem until he managed to get Mehmed to intervene. That attack sealed Gandhi's cultural win and simultaneously led to Mehmed getting eliminated by Willem. It was a bit of a tightrope that Gandhi had to walk in this match, and he was able to pull it off successfully.

Congrats to MathBandit for winning the picking contest in sole first place with 19 points. This was a bit of a disaster for the picking contest field, as the average score was a mere 4.62 points across about 150 entries. Better luck to all of us next time!

* * * * * * * * * * *

Over a year after Season Four concluded, Wyatan decided to rerun the Season 4 games, 20 times each, and publish the results. The objective 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.

These were the results of Wyatan's investigation of Game Two from Season Four. This game had a few remarkable features. First, Wang kon set a record for the fastest elimination of another leader (Pacal, Turn 104). Second, it saw the emergence of Willem as a strong leader, by finishing second in this game. He would later reach the Championship game. When Sullla started AI Survivor, people thought he'd be a strong contender. But he was eliminated in the opening round of each of the three previous seasons (actually even first to die every time iirc), and at that point people had given up on him: he was the favourite for first to die in this game. Third, it saw yet another confirmation of how strong a leader Gandhi is. With the hindsight of future games, I fully expected that this game would be a competition between Willem and Gandhi, with very few games not going either way. Well, here's how it actually went:



(note : "A" column tracks the number of war declarations initiated by the AI, "D" the number of times the AI is declared upon, "K" the number of kills)

Pacal: He was the biggest disappointment of this game, being the arguably the worst performing AI here, in spite of his two wins. His survival rate is by far the lowest, and he essentially remained passive until someone killed him. The wars he launched mainly consisted in joining a dogpile, and pulling out before he made any substantial gains. His first win was extremely lucky: if Mehmed could plot at pleased, he'd have easily won. His second victory was more deserved, with an aggressively initial settling pattern that he was able to capitalize upon. But even there, his victory was late, he benefited from Mehmed's extremely unusual lack of early settlers, and from the two strongest eastern civs (Louis and Gandhi) being unusually weak in that game. Now, a lot can be explained by his very poor land: to start with, a coastal start in the middle of the map means no backlines. Then, the surrounding land had a lot of tundra otherwise poor terrain (desert, mountains). The only somewhat decent land was towards Korea... and Korea usually settled most of it. But that's not the whole story. Look at his A/D stats, and compare them with Roosevelt in the previous game: Roosevelt was in an even worse spot, but at least he tried to breakout. Pacal just sat back in his corner. Now, we've seen Pacal be a lot more aggressive in other games. But that was when he was in a position of strength. I think he's just a coward.

Gandhi: He was also a disappointment. He performed decently, winning four games in a deserving fashion. But he fell far short of expectations. When Willem did well, he simply lost to Willem. And when Willem faltered, he failed to rise up to the occasion more often than not. The reason of his losing to Willem seems easy to explain: pursuing a Cultural victory plan requires religions for the multipliers. And the better teching Willem would most often get them, not Gandhi. As for when Willem was out of the equation, Gandhi's overall passivity usually allowed Louis to become too strong for him to handle. In the rare occasions when he took a more proactive approach, he fared much better. So his peaceweight might not be the only reason he never made it to the Championship. Is this the case of a somewhat overrated leader?

Louis XIV: Louis was a really strong performer on this map. Second only to Willem, and far ahead of Gandhi and Wang Kon. He usually expanded very well, and even extremely well when Victoria struggled. His position between three high peace weight leaders could have made him a strong candidate for first to die, but: Gandhi was essentially passive, Victoria struggled with poor land, Wang Kon was equally surrounded by enemies. So while he did face a few dogpiles (and folded in those cases), more often than not he was able to pick one of his neighbours off and become one of the game leaders from there. His problem as an AI, and it showed here as well, is that he's torn in multiple directions, and that leads him to throw good positions away, like when he plays for a domination run but techs as if going for culture, or the opposite. And then, obviously, there was the particular issue in this game of Willem who'd win too fast for Louis to cash in on his dominant position.

Mehmed: Mehmed is a warmonger who can't plot at Pleased. And that's a very bad combination. It's come to the point I believe that as a rule of thumb, you shouldn't pick a leader who "can't plot at Pleased" for the winner (although conversely, they might make good picks for runner-up). They're essentially playing with a hand tied behind their back. I mean, we've had some of those win, sure. And it's less of an issue if they're pursuing a cultural or spaceship victory. But a warmonger? Now, in this game, it rarely came into play (with the glaring exception of the second game he'd have won otherwise). Because this game was all about the Mehmed-Willem conflict. And that's a conflict Mehmed rarely won. He usually either lost, or both got dragged down. It's actually a bit hard to explain. A warmonger with good land should win a 1v1 vs an eco leader more often than not. One explanation is Wang Kon, Mehmed's other punching bag, who punched back hard. Another could be that a lot of attacks were directed at Willem's second city which happened to be placed on a hill. Another might be that Willem's better teching allowed him to get to Guilds a lot sooner than Mehmed. Getting Mining right out of the gate (it did happen quite a few times ) usually backfired for him, ironically: he would then get a super early Bronze Working and Iron Working (pre-turn 40, when for instance, Victoria often got it around turn 100), start plotting, and attack Willem very early. Which usually ending badly for both of them.

Victoria: Another bad performer here. She's a bit of a puzzle, because she's an AI with great traits, and whom many a player report as doing well in their games. And yet, apart from the wildest card game, she's always performed extremely poorly in AI Survivor. In this particular game, land can explain a lot: at first glance, she's got lots of room to expand, with some rich (green) terrain. But, the food. Her land simply has no food. There's the banana/rice spot where she would always found her second city: but that's useless before Iron Working, a tech she'd never prioritize. And by the way, that second city placement would thus hamper her developement. And then, the only food available around her was located in the northern tundra peninsula. You should always settle the food, towards the center of the map: two opposing directives in her case. She still managed in some cases, but she struggled more often than not: founding slow-developping cities towards the center of the map first, then having barbarians occupying the peninsula... that she had no copper to deal with. Then of course there was Louis. A struggling Victoria was set up as Gandhi's shield against a strong Louis. Bad recipe. But there might also be an explanation in her A/D scores: she was weak... and she declared 37 wars while being declared upon 15 times only? Is she a suicidal AI?

Wang Kon: While the story of this game was the story of the Mehmed-Willem conflict, the real star of the game was Wang Kon. By all rights, he should have been first to die: a high peace weight AI surrounded by four(!!) low peace weight AIs. And sure, the dogpiles kept coming. And yes, he succombed at time. But more often than not, he doggedly fought on, survived, and thrived. It sure helped that Louis was also facing threats on his eastern front, that Mehmed was often more busy with Willem than him, that Pacal was weak, and that Willem was more actually more ally than foe. But still.

Willem: If there was a Willem vs Gandhi competition... Willem crushed it. At first glance, they had extremely similar positions: Both in a corner of the map, with great land available. Gandhi's land was the richer, but Willem had earlier access to gold tiles, and some of Gandhi's land needed jungle removal to be exploited. Both had an arch-enemy: Mehmed for Willem, Louis for Gandhi. Both had an ally: Wang Kon for Willem, Victoria for Gandhi. Both faced a minor secondary threat: Pacal for Willem, across-the-map offensives for Gandhi. But in truth, Willem faced the tougher situation: there were a lot of games where Gandhi was left alone for ages, when Willem had to deal with Mehmed in every single game. And yet he was by far the best performing AI on this map. When he was in a race with Gandhi, he just won. He was faster to the cultural win, and had a strong spaceship plan B. But he had a major Mehmed issue, and that conflict would basically shape up every single game. If Willem could delay it sufficiently, or win it in a timely fashion, he'd go on to win the game. Otherwise, he was out of the running.

We all know now the major weakness Willem has about skipping Rifling for way too long. But I was also surprised at how inefficiently he fought at times: he would have twice Mehmed's power and units, and yet Mehmed would have a stack capturing cities, while Willem couldn't manage to form his own stack. That was... painful to witness. And that caused him to stay at war way too long in some games. He doesn't have the single-minded focus on the cultural victory that Gandhi has (meaning for instance his industrial era research won't be a hard beeline to Mass Media), but if he can get enough early religions (which his teching power certainly helps him with) and spread them around (which requires monasteries that he won't build when at war), he'll pull the trigger much sooner than Gandhi ever does. Willem is thus confirmed as a very strong AI, and at least from this game's results, stronger than Gandhi. But is he a weaker version of Huyna Capac, or in the same league?