My Yearly "Is the AI Any Good Yet" Thread
- Thread starter Peng Qi
- Start date
Pfeffersack
Deity
I'm not sure if lacking gold is the (only) reason for this - I see outdated units even for civs with plenty of cash. Ressource shortages might explain some additional cases, but I'm leaning towards that there is probably another issue - similar to what prevents the AI from being more determined, when it comes to improving ressource tiles.
Game AI has to be able to make decisions in between turns (so hopefully not much more than a few seconds) on a wide variety of different machines, some very old and weak. Suggesting that game AI should use cutting-edge algorithms that require supercomputers to run is... absurd.
There is room for state of the art machine Learning algorithms in AI. The total war series used them with different levels of succes since Shogun I. A game like civ, however, I don't think would benefit from it much, except maybe in pathmaking, and low level management. Why I think so would be quite technical, but it is besides the point.
The secret to program a good AI is most often than not, to trick the player into thinking the AI is playing the same game while it is indeed responding to a completely different set of goals and beding the game rules in a lot of ways.
The best game AIs use a lot of different techniques combined to solve different problems, and also cheat in a lot of ways to be more effective and also more fair end entertaining. Fxs tried to make the AI play with the same rules as the player, assuming they could create a roleplaying competitive non-cheating AI. Which is a totally unrealistic goal and big mistake for the kind of game civ is. And then, they also clearly under-funded and under-supported the AI during developement. They should know better after so many civ games.
Last edited:
I agree: the goal of game AI is not to actually be intelligent, but to behave it a manner that the player perceives as intelligent. It doesn't matter how complex and advanced your algorithms are if they don't result in behavior that the player notices.
Yesterday, I set up an AI only match for which I was the spectator. Deity-level AIs, using handpicked civs that I figured are some of the better AI ones in the game—Kongo, Pericles Greece, Rome, Korea, Russia, and Macedon. To me, the results showed that the AI is competent at early expansion, has gotten significantly better about city placement and districting (with the exception of Korea, who always destroys her Seowon adjacency by clustering a bunch of other districts next to them), and competes hard for most wonders (except Pyramids, Petra, and the Colosseum, which were all built late). However, it completely fails to target any victory conditions or wage offensive wars with any efficacy after walls come up (even if one AI has an overwhelming scientific and military advantage over the other). The result is that winning in single player is basically a game against yourself and the map, with AIs being closer to map flavor and sources of gold than real competition.
In my observed game, the Kongo eventually launched the exoplanet expedition and won via science victory around turn 335. All of the AIs expanded alright during the early game, taking out a fair amount of city states, building some good wonders and districts, and teching up to between 60 and 120 science per turn at turn 100. Their main issues early on involved struggling with barbs, never building a government plaza, and never removing an improvement after strategic resources spawned underneath. Mid-game, they started to lose focus and built a bunch of entertainment complexes, military buildings, and units to replace the ones they lost. Several civs had some gold difficulties that led them to disband units while still building others, and they never seemed to build enough traders and builders. The pointless wonders started getting built. Greece built a zero tile Petra, not adjacent to an Acropolis, and then later on spent another 15 turns of production on seawalls to protect it. Still though, all of the AIs managed to build up enough of an empire by turn 150 that a decent human player could have probably won the game by turn 250 at the latest if they’d taken control of the AI empires over at that point.
Late game is where the AI really fell apart, and in ways that seemed like it was working to actively avoid victory and squander military advantages. The religious civs gave up on converting others. Culturally, Russia came kind of close to winning for a brief period...but then stopped building rock bands. Diplomatically, Greece was 3 points short of victory, but never even started to build the Statue of Liberty. Guess they thought that a sixth spaceport and another nuclear submarine were higher priorities. The tech leader, Kongo, basically only achieved a science victory after doing everything else possible, including building a massive late game army that took a total of 1 city. Every AI built a spaceport in every single city once they could, but they researched the space project techs last and Kongo didn’t build any laser projects to speed up the finish.
The AIs civic and government choices also hindered them. They always chose fascism as their tier 3 government (ironically making them more peaceful because of positive shared government modifiers) and ran civil prestige (extra housing and amenities for cities with governors) as their main economic policy card. They always ran raid, but never pillaged a single tile or district. They dumped a bunch of production mid and late game into units, but either sent the units out in bunches of 4 to pointlessly die or kept repositioning a massive army inside their territory (despite having defenseless neighbors next door). Perhaps most importantly, their late game military targeting was absolutely pathetic. I don’t think a single city with medieval walls or higher fell, regardless of military superiority. I watched Kongo fail to take a single, ungarrisoned 100 strength city while having 7 Giant Death Robots and 3 bombers—the GDRs just wandered around, taking hits from multiple cities and encampments, never significantly damaging any one city.
I honestly don’t think these issues are impossible to fix and I think it’s reasonable to expect a Diety AI to pose a much bigger late game threat. An AI researching flight at turn 150 should be able to push for a ~t250 science victory by using some relatively simple strategies (build 2 spaceports, run campus projects, maximize science with policy cards, and prioritize researching techs with space projects). In terms of an offensive military, I don’t see why it can’t, once at war, just target whichever city is closest to its borders, instead of aimlessly spreading units out across two or three cities. And in terms of production and policy choices, it seems like the AI should be able to make better choices by calculating the various yields of those decisions (science, culture, gold, faith, production, tourism) and applying formulas to determine what has the overall highest yield.
In my observed game, the Kongo eventually launched the exoplanet expedition and won via science victory around turn 335. All of the AIs expanded alright during the early game, taking out a fair amount of city states, building some good wonders and districts, and teching up to between 60 and 120 science per turn at turn 100. Their main issues early on involved struggling with barbs, never building a government plaza, and never removing an improvement after strategic resources spawned underneath. Mid-game, they started to lose focus and built a bunch of entertainment complexes, military buildings, and units to replace the ones they lost. Several civs had some gold difficulties that led them to disband units while still building others, and they never seemed to build enough traders and builders. The pointless wonders started getting built. Greece built a zero tile Petra, not adjacent to an Acropolis, and then later on spent another 15 turns of production on seawalls to protect it. Still though, all of the AIs managed to build up enough of an empire by turn 150 that a decent human player could have probably won the game by turn 250 at the latest if they’d taken control of the AI empires over at that point.
Late game is where the AI really fell apart, and in ways that seemed like it was working to actively avoid victory and squander military advantages. The religious civs gave up on converting others. Culturally, Russia came kind of close to winning for a brief period...but then stopped building rock bands. Diplomatically, Greece was 3 points short of victory, but never even started to build the Statue of Liberty. Guess they thought that a sixth spaceport and another nuclear submarine were higher priorities. The tech leader, Kongo, basically only achieved a science victory after doing everything else possible, including building a massive late game army that took a total of 1 city. Every AI built a spaceport in every single city once they could, but they researched the space project techs last and Kongo didn’t build any laser projects to speed up the finish.
The AIs civic and government choices also hindered them. They always chose fascism as their tier 3 government (ironically making them more peaceful because of positive shared government modifiers) and ran civil prestige (extra housing and amenities for cities with governors) as their main economic policy card. They always ran raid, but never pillaged a single tile or district. They dumped a bunch of production mid and late game into units, but either sent the units out in bunches of 4 to pointlessly die or kept repositioning a massive army inside their territory (despite having defenseless neighbors next door). Perhaps most importantly, their late game military targeting was absolutely pathetic. I don’t think a single city with medieval walls or higher fell, regardless of military superiority. I watched Kongo fail to take a single, ungarrisoned 100 strength city while having 7 Giant Death Robots and 3 bombers—the GDRs just wandered around, taking hits from multiple cities and encampments, never significantly damaging any one city.
I honestly don’t think these issues are impossible to fix and I think it’s reasonable to expect a Diety AI to pose a much bigger late game threat. An AI researching flight at turn 150 should be able to push for a ~t250 science victory by using some relatively simple strategies (build 2 spaceports, run campus projects, maximize science with policy cards, and prioritize researching techs with space projects). In terms of an offensive military, I don’t see why it can’t, once at war, just target whichever city is closest to its borders, instead of aimlessly spreading units out across two or three cities. And in terms of production and policy choices, it seems like the AI should be able to make better choices by calculating the various yields of those decisions (science, culture, gold, faith, production, tourism) and applying formulas to determine what has the overall highest yield.
Last edited:
tedhebert
Emperor
I am personally very hopeful for Humankind in all regards ( https://forums.civfanatics.com/forums/humankind-by-amplitude.599/ ), AI included, but not because they have said the game will have an amazing AI, but because Humankind will not have 1UPT but instead limited - stacks - fighting - on - separate - tactical - battlefield. Like, it will have stacks of IIRC up to eight units which, when meeting enemy stack, fight on a small tactical layer.
I am absolutely convinced at this point that it is technologically impossible, in the modern era, to program not awful AI for the 1UPT system of civ6 WHICH IS ALSO not extremely AI turn time and performance draining. Civ6 1UPT system, with a great quantity of units and their types, very restrictive terrain movement, strategical and tactical layer in the same time, a lot of players and an enormous amount of variables is absolute nightmare to program AI for. Just too many variables, horrifying amount of variables, too much creative micro and macro management on too many layers in the same time to code with our still poor AI entertainment technology, working on the average gaming PC quickly.
You can design a tactical game focusing on turn based "1upt" combat entirely, with less players and other interfering factors, and barely designe somewhat decent AI. You can design a grand strategy game with abstracted combat with somewhat decent AI (I consider most of Paradox games to be in this category, though AI is neverending struggle to patch and update). Civ6 tries to have enormously complicated strategy game on a grand scale and enormously complicated tactical combat on the same layer, and quick AI turns. This is just damn impossible, and if Firaxis can't do that despite being among few biggest strategy game studios on the markey, then I don't believe some random studio can do a magical breakthrough in this regard.
Don't get me wrong, I don't think Humankind's AI will be particularly clever. In fact, I am quite ready for it being quite dumb on release. But civ6 AI is IMO far below the "acceptable level of dumb strategic AI" - it has an entire CRUCIAL layer of the game, "can AI opponents directly endanger the human player past very early game", crippled. It is one thing to have a strategy game where the opponent is not very bright but systems work in such way that you still have a challenge roughly equal in many areas. It's another thing to have the combat system which effectively makes human and AI wars akin to adult man kicking down a mentally challenged child. Or to have a game where the best way to to do everything is through military, because of meta - reason of AI being so bad in this thing in particular.
Humankind, and other games with any other system of limited stacks, tactical battles or abstracted military, still won't cause cosmic breakthrough in AI technology, but their fundamental systems will more easily allow for AI to breathe down your neck and rise your heartbeat. Civ4 is generally harder than civ6 for a very simple reason - it is within a capacity of AI to build a somewhat decent stack of many, high quality units, and send it against you in the not-dumbest-moment-possible, many strong units in one place. You can't cheat or exploit yourself out of that hole, you need decent strength of army against that. Well, in 1UPT system what happens is that poor overwhelmed AI manages to create a big, high quality units army, which then dissolves into chaos caused by the permanent traffic jam that is the horror of movement in this game, units shuffle chaotically and each one of them presents a metric ton of choices, they block each other, get stuck on bad terrain, are unable to take cities because of the horrible idea to make it depend on melee units only (thereby you essentially need to kill only half of AI army and you are 100% safe), and you systematically massacre them with a vastly smaller number of your micromanaged troops.
At least in Humankind AI will have far less weight on its shoulders and far less opportunities to screw up.
Last edited:
Good thing that authors of the Vox Populi didn’t know that what they had done was impossible.
As a fan of Endless Space 2, I don't have high hopes for Humankind. ES2 was left in a poor state after the last expansion with no communication from the devs. They finally released a teeny tiny bug patch after far too many months and have all but said they don't plan on addressing the major balance/gameplay issues that remain. Only some more minor bugfixes.
It is a beautiful and interesting game with fantastic music...
Ok, end rant.
kaspergm
Deity
- Joined
- Aug 19, 2012
- Messages
- 5,829
Great write-up indeed.
Going slightly off topic (and very off game), this reminds me of what is easily one of the funniest things I ever read online. This is not about Civ, but anyone who ever played Heroes Of Might And Magic should go to section XI of this page and read about an AI experiment someone did back in the days of Heroes IV. Caution: You may just die from laughter. (The rest of the page is funny as well, but section XI tops it all.)
Abaxial
Emperor
- Joined
- Sep 14, 2017
- Messages
- 1,216
These observer experiments are always fun. There was a very interesting test conducted by Marbozir: what happens if you give an AI civ a GDR on turn 1 (and enough uranium to power it)? Answer: very little. With a GDR from the start of the game, the civ so gifted should have made mincemeat of the rest, but it just pottered about, killed a few barbs, and never attacked an enemy city..
Yep, we would be surprised on what competent people could do with time and resources. Or even without resources, but with passion and dedication as proved by vox populi. Shame that the AI developers in civ 6 (though probaly only one guy worked full time on the AI) were probably never given the proper time or the proper resources to make the AI the game needed. And I think it is not the only mechanic that was underdevelopped.
On a possitive note, Im surprissed on how much support the AI actually received during the game cycle. Im sure the "lead" AI designer has done a monumental job to add support for the constant flux of new mechanics.
I can only wonder what were the constrains, when started to design and code the AI. To me It is also obvious that Fxs does not have a good AI culture, or that they didn't have enough time plan for a very good design to begin with.
Hope that they will realize soon that if they don't put the effort on creating what the players really want and deserve. Sooner or later someone elses will, and if Fxs does also refuse to give proper modder support, it will be in someone's elses roof.
If someone really thinks it is not posible to make a good AI to play civ. I recomend you to research on the AI of the total war saga. That is a much more complex problem to solve, as they need to manage thousands of units in real time with individual distinct behaviors, coordinations, pathmaking, dynamic goals and subgoals, morality realistic terrain weather and cover modifiers... and all in real time. Meanwhile in civ VI, they have been tried to make the AI use planes properly for 4 years.
Last edited:
I ran the same map and configuration again with @Infixo's Real Strategy AI mod. I found some noticeable, though modest, improvements in AI performance. This time, Korea ended up winning the space race around turn 325, with Kongo and Russia also being a few turns from victory. Kongo was the first to launch the exoplanet expedition, but didn't launch any laser projects, while Korea and Russia both launched a few. All of the AIs nearing science victory still made the mistake of avoiding techs that unlock space projects and building a spaceport in every city. The AI didn't prioritize holy sites or most of the wonders quite as early (though Pyramids went at ~t80, which is one of the earlier finishes I've seen). The third and fourth religions went much slower, which actually created some difficulties for Russia in converting civs, as Korea and Rome popped religions after being fully converted. There were some sporadic religious wars thereafter, but no civ managed to convert more than 2 for long.
Culturally, Greece was the main contender this round as Russia opted to develop science more heavily and became the world's bully. Still, despite a pretty impressive early culture per turn (271 by turn 150), Greece never got particularly close to winning and any chance they had of a culture victory vaporized after the first Moon Landing. There was still plenty of dumb wonder-building. Peter built the Panama Canal from his capitol to an ice-locked harbor. Alex built St. Basil's (rather than Petra) in the desert, beating Peter's actually good St. Basil's by a couple turns. Surprisingly, no one built Petra, even though a few cities would have been decent locations for it.
The Real Strategy AI was genuinely good about clearing out barb camps--like, better than most human players are, if the complaints on this forum about overly-aggressive barbs are any indication. Settlement locations were further improved and the AI claimed good land faster. I only saw a few off-water settlements, mostly in locations with no water available. Districting and policy choices were more or less the same as the base AI (decent and bad, respectively). The RS AI dealt with loyalty much better, with only one city flipping during peace over the course of the game.
The RS AI's military efficacy was not noticeably better than the base AI. Kongo and Russia conquered almost all of the city states between them but no war between civs resulted in anything more than one border city changing hands. Kongo and Russia ended up being the main two military powerhouses but didn't manage to do much of anything. At different points, each marched four GDRs into the others territory, but the GDRs aimlessly attacked different cities and ended up either dying or retreating. Russia nuked Greece but it was more like a spite nuke than anything strategic--the city was on the other side of the world from their border, they had no units in the area, and there would have been no way for them to hold the city loyalty-wise.
The RS AI was definitely better about sensibly building and actually using navies, though still not great. Unlike the base AI game, there were no lake-locked navies for most of the game and I saw the AI semi-successfully deploy submarines against opponents. However, they seem to still have trouble keeping a good balance of melee and ranged ships. At turn 325, Korea has bombarded a Macedonian city to 0 health with a big fleet of nuclear submarines, but has no melee ships or units to finish the job.
And finally, here's the science and culture per turn at each 50 turn increment:
Overall, science and culture per turn was pretty comparable for most of the game, though the RS Ais were able to ramp up late game science much more effectively than the base AIs. Note that, though I used the same configuration and civs for each test, in the Real Strategy play-through, Korea spawned where Greece had spawned in the original one (I guess they have the same start bias). So Korea had the rough start next to horse barbs on the base AI run and the sheltered, mountainous start on the RS AI run.
Culturally, Greece was the main contender this round as Russia opted to develop science more heavily and became the world's bully. Still, despite a pretty impressive early culture per turn (271 by turn 150), Greece never got particularly close to winning and any chance they had of a culture victory vaporized after the first Moon Landing. There was still plenty of dumb wonder-building. Peter built the Panama Canal from his capitol to an ice-locked harbor. Alex built St. Basil's (rather than Petra) in the desert, beating Peter's actually good St. Basil's by a couple turns. Surprisingly, no one built Petra, even though a few cities would have been decent locations for it.
The Real Strategy AI was genuinely good about clearing out barb camps--like, better than most human players are, if the complaints on this forum about overly-aggressive barbs are any indication. Settlement locations were further improved and the AI claimed good land faster. I only saw a few off-water settlements, mostly in locations with no water available. Districting and policy choices were more or less the same as the base AI (decent and bad, respectively). The RS AI dealt with loyalty much better, with only one city flipping during peace over the course of the game.
The RS AI's military efficacy was not noticeably better than the base AI. Kongo and Russia conquered almost all of the city states between them but no war between civs resulted in anything more than one border city changing hands. Kongo and Russia ended up being the main two military powerhouses but didn't manage to do much of anything. At different points, each marched four GDRs into the others territory, but the GDRs aimlessly attacked different cities and ended up either dying or retreating. Russia nuked Greece but it was more like a spite nuke than anything strategic--the city was on the other side of the world from their border, they had no units in the area, and there would have been no way for them to hold the city loyalty-wise.
The RS AI was definitely better about sensibly building and actually using navies, though still not great. Unlike the base AI game, there were no lake-locked navies for most of the game and I saw the AI semi-successfully deploy submarines against opponents. However, they seem to still have trouble keeping a good balance of melee and ranged ships. At turn 325, Korea has bombarded a Macedonian city to 0 health with a big fleet of nuclear submarines, but has no melee ships or units to finish the job.
And finally, here's the science and culture per turn at each 50 turn increment:
Spoiler Show table :
Overall, science and culture per turn was pretty comparable for most of the game, though the RS Ais were able to ramp up late game science much more effectively than the base AIs. Note that, though I used the same configuration and civs for each test, in the Real Strategy play-through, Korea spawned where Greece had spawned in the original one (I guess they have the same start bias). So Korea had the rough start next to horse barbs on the base AI run and the sheltered, mountainous start on the RS AI run.
glider1
Deity
I wonder if the way Firaxis will go in the future is to redesign civ7 for much shorter game durations setup for multiplayer first and with enough cut down options that a neural net AI can be trained to play it so that Firaxis doesn't have to spend any time at all on AI development? They would need to rent deep learning time though for each update cycle to retrain the AI. I doubt they will go that way though. Maybe some kind of a hybrid solution where the strategy is hand coded and the tactical moves are machine learned?
I doubt they will go that way.
Any way given the nature of the game, a NN may not be the best a approach. The good (and bad) thing about civ, is that there are a hundred of subsystems that do no rely in each other that much.
City building, research, pathfinding, war, trade,and many other systems can be almost independent of each other. You need a general strategy director system that will send and receive information, and many small subsystems that may range from a Monte Carlo optimization algorithm to a decission tree or an expert system. NN are not a good solution for a game with such complicated always changing dev cycle, but sure modern NN can be used too.
If I recall correctly, in Shogun 1 NN were used to control individual units. While diplomacy and empire management were handled by other systems. Now they use far more complex solutions.
But anyway, the way to go would be to me (if they havent done it yet, which they probably have to some extent) to divide the AI turn in separate stages. Research-Diplomacy-Build-Move. Each stage a far simpler problem to solve with each problem handled by a system / algorithm suited for it and with a general strategic system obove all coded with some specific personality traits for each leader.
Last edited:
superpowerfixer
Chieftain
- Joined
- May 17, 2020
- Messages
- 20
So what was firaxis ai dev doing all the time?Was he watching paint dry or something?
He is paid money to do his job,what time and resources does he need more?
One guy is the norm in making ai.You don't need machine learning or a team,only one guy being competent at playing the game.
And as the ai choices in wonders/government/districts prove,he is not.
Same problem with civ v. Ai went for piety/honor and failed miserably,even with a big lead.
Since the last patch, probably the dev was working his ass off to make an acceptable AI for RD season 2. Before that, he had plenty of work to do adding new mechanics, balancing tuning and testing. The AI was changed in every single patch the game got.
Also, a game like this should have a team of 3 to 5 people working in the AI full time in the main developping stage. With one person dedicated to testing and balancing, and at least 2-3 working in different layers of the subsystem. In the postdev active stage, a minimun of 2 people are required. You never put a major game component in the hands of one guy if you are not an indie studio.
A game this size, has a minimun of 50 people in the peak of active development (Conservative number) including in music, programing, asset design, multi platform, multiplayer systems, modeling, animation, ui, qa, and who knows what else... And you think one guy making all the AI is enough? That mentality is the mentality that ends with the kind of AI we got.
Also one guy is absolutely not the norm in games with a budget higher than 1M. I would say Civ is in the 20M budget range. Using one guy for AI is a warranted disaster.
Also the AI has many other problems to solve besides euristic rule systems that you can code by simply knowing how to play. Pathfinding, movement, building research, dynamic goals and subgoals... all those problems require flexible optimization techniques. Most of the suitable ones are in the machine learning field. Your appreciation is totally incorrect.
Last edited:
Kriegsspieler
Chieftain
- Joined
- Jan 25, 2017
- Messages
- 11
I rarely post here, although I follow the Civ Fanatics forum fairly closely, and have done so for years.
Unlike some others here, I have seen only minor and very incremental improvements in Civ VI's AI. Yes, its combat/military behavior has improved a little, but as JesseS's interesting post above observes, in many other ways the AI does little to give the player much of a challenge. So for me, Civ V remains a more satisfying game, largely because one can include mods that make the AI better.
In my opinion, one of the things that really works against significant improvements to any game's AI is the extent to which many games are now built for multi-player play. I have no principled argument for saying this, only my years of experience playing computer games, which goes back to the original Civ -- back when it was Sid Meier's game in something more than name only -- and the days of 256k processors. A good example of what I am talking about is the evolution of Paradox's Hearts of Iron. HOI2 was an interesting and challenging game, and it enjoyed the benefit of a number of fantastic mods that improved both the flavor and the AI's performance. Although HOI2 could be played multi-player, both the hosting capacities and the design of the game itself made that something of a chore. By the time we get to the current edition, HOI4, the game has been designed from the start as both an MP and an SP game, yet its AI is pretty miserable and has improved little over the game's lifetime. I believe the challenge of creating a game that is balanced for MP play has eviscerated the AI. I also believe other Paradox games show the same effect.
Are there good games out there with challenging AI's? Yes, there are, and I'll cite three of them. Although not designed as a 4X game, AGEOD's Field of Glory:Empires is Mediterranean world-spanning game that gives you the opportunity to play at least 3 of the X's in the 4X game. It is complex, immersive, and very challenging to play. The AI-controlled countries will not make your life easy, and if you manage to finish in first place, you will have earned it.
Second up is Old World, which was previewed here a couple of months back by Soren Johnson before it was released. It is a 4X game with a great AI. Set the difficulty on one of the higher levels, put the Barbarians on "strong" and see how well you fare. I'll add something else here about Old World, which is that its imaginative use of events and chains of events to provide unexpected story lines makes the game incredibly replayable. It also shows how far behind the curve Firaxis has fallen in creating unexpected twists and turns in gameplay. There's a reason why many people find Civ pretty boring in the middle and late game. There's too much mindless grinding.
The third example, Matrix's Shadow Empire is more niche, a post-apocalyptic 4X game that creates a planet for you, plops you down on it, gives you a few undistinguished military units, and tells you to go. It is a complex and -- let's be honest -- not an entirely transparent game. But it too has an excellent and challenging AI, as well as interesting and unexpected events that upend even an experienced player's best-laid plans!
So yes, there are games with good AI out there. Is it an accident that all three of these games are made by companies far more modest in means than either Firaxis or Paradox? Will Humankind also be one of them? I am hopeful, but also skeptical. If Humankind is designed out of the box as an MP game, I fear we will be disappointed.
Unlike some others here, I have seen only minor and very incremental improvements in Civ VI's AI. Yes, its combat/military behavior has improved a little, but as JesseS's interesting post above observes, in many other ways the AI does little to give the player much of a challenge. So for me, Civ V remains a more satisfying game, largely because one can include mods that make the AI better.
In my opinion, one of the things that really works against significant improvements to any game's AI is the extent to which many games are now built for multi-player play. I have no principled argument for saying this, only my years of experience playing computer games, which goes back to the original Civ -- back when it was Sid Meier's game in something more than name only -- and the days of 256k processors. A good example of what I am talking about is the evolution of Paradox's Hearts of Iron. HOI2 was an interesting and challenging game, and it enjoyed the benefit of a number of fantastic mods that improved both the flavor and the AI's performance. Although HOI2 could be played multi-player, both the hosting capacities and the design of the game itself made that something of a chore. By the time we get to the current edition, HOI4, the game has been designed from the start as both an MP and an SP game, yet its AI is pretty miserable and has improved little over the game's lifetime. I believe the challenge of creating a game that is balanced for MP play has eviscerated the AI. I also believe other Paradox games show the same effect.
Are there good games out there with challenging AI's? Yes, there are, and I'll cite three of them. Although not designed as a 4X game, AGEOD's Field of Glory:Empires is Mediterranean world-spanning game that gives you the opportunity to play at least 3 of the X's in the 4X game. It is complex, immersive, and very challenging to play. The AI-controlled countries will not make your life easy, and if you manage to finish in first place, you will have earned it.
Second up is Old World, which was previewed here a couple of months back by Soren Johnson before it was released. It is a 4X game with a great AI. Set the difficulty on one of the higher levels, put the Barbarians on "strong" and see how well you fare. I'll add something else here about Old World, which is that its imaginative use of events and chains of events to provide unexpected story lines makes the game incredibly replayable. It also shows how far behind the curve Firaxis has fallen in creating unexpected twists and turns in gameplay. There's a reason why many people find Civ pretty boring in the middle and late game. There's too much mindless grinding.
The third example, Matrix's Shadow Empire is more niche, a post-apocalyptic 4X game that creates a planet for you, plops you down on it, gives you a few undistinguished military units, and tells you to go. It is a complex and -- let's be honest -- not an entirely transparent game. But it too has an excellent and challenging AI, as well as interesting and unexpected events that upend even an experienced player's best-laid plans!
So yes, there are games with good AI out there. Is it an accident that all three of these games are made by companies far more modest in means than either Firaxis or Paradox? Will Humankind also be one of them? I am hopeful, but also skeptical. If Humankind is designed out of the box as an MP game, I fear we will be disappointed.
Last edited:
kaspergm
Deity
- Joined
- Aug 19, 2012
- Messages
- 5,829
Maybe a dumb question, but for someone who hasn't played Civ5 for ages, and who might want to go back, which mod is it I should use for AI in Civ5?
Similar threads
