Why Civ 6 AI is terrible - Part 1: Wonders

[Stringer1313]

That is a valid point, but if you haven't already, check out the Civ5 Vox Populi combat AI. If Civ6 combat AI could even be half as good, I would be happy.

What was the secret to programming the Vox Populi combat AI in Civ 5?

 

[Acken]

A (or more) dedicated coder.

From what I remember the tactical AI was the main focus of the coder of artificial unintelligence which is the basis of VP tactical AI IIRC. I dont know how much the people from vox populi have added on his work.

Quality AI just takes time and dedication. Its no secret.

 

[Acken]

Obviously stupid moves are frustrating to watch, and in some cases reduce the challenge (Oh no, my city's at risk! Oh well, the AI lost interest), but ultimately the AI needs to be a challenge for peaceful victories. The way domination works in Civ V and VI means that the AI can literally not win a domination victory (another AI failing - vacillating repeatedly over preferred vicvory types and then choosing ones it has no ability or takes no action to pursue - "X is aiming for Domination Victory" followed by no wars or attacks being especially obvious), so as far as actually winning the game is concerned, there's an upper limit to how far it can get with units however much the combat AI is improved.

Thats mostly because it takes a lot of time to fine tune weights and logic behind the AI moves for something like AI domination to be a thing.

Here is an example of @IronfighterXXX losing to a deity AI on my mod in which I spent a lot of time tuning the AI warmongering part. But you know what ? Ive seen complaints of the AI being overly agressive as a result in the mod. And its true that Deity is really hard. The problem with civ is that there is a lot of different ways to enjoy it and people will like/dislike whatever direction you take. Make a really mean AI and you'll frustrate more casual and peaceful players. Some players just like to think diplomacy should purely be based on modifiers acquired through gifts and allow you to manipulate it. They dislike backstabs with a passion even though its a lot closer to human behavior ! On the other hand make something too easy and veteran will get bored. I think that Firaxis has just decided who their target audience is when it comes to the numbers.
https://forums.civfanatics.com/thre...-ai-improvements.550671/page-93#post-14911033

The part where all players aggree is to reduce AI nonsense. Like random DoW and not defending. But outside of these clear scenarios that break immersion you'll have to come up with a complexe difficulty level structures that could accomodate many different player profiles. And to allow that customization of AI behavior requires work. Although, if at least the nonsense wasn't there and it could simply be tweaked, mods would be enough to satisfy everybody between the peace loving and world wars fans.

I also think that a part of the problem also comes from the balance and design choices. Complicate things and making a good AI becomes harder. Have imbalances and unless hardcoded the AI will often fall into the weak options while human players quickly learn what is good. This is a tremendous advantage.

 

[Magma_Dragoon]

Every time this issue is raised, some individuals suggest that denying a wonder to hurt other players is a potentially beneficial move. If the AI made such a decision we could argue about that, but that's not what it does. It throws a dart at a board when it decides what to build, and if it lands on wonder it picks a wonder. It doesn't stop and evaluate "hey, none of these wonders help me" and change its mind to building a bank instead. Even if it decided to build a wonder to sabotage somebody else, it's rolling the dice to maybe set back somebody else and definitely set itself back, production down the tubes. An AI is cutting off its nose to spite the face of at most one of 7 opponents, while 6 others gain. It is a myopic argument that rustles me greatly. If another player has a city in position to become a super Petra city, I'm happy. He's going to build a nice thing, and when I decide to take that city off his hands, it will be my nice thing. And I don't buy "nothing better to build". If it is so far ahead it can build wonders willy nilly and win, it clearly would also have an advantage at building units and take more land from other players.

 

[PhilBowles]

Thats mostly because it takes a lot of time to fine tune weights and logic behind the AI moves for something like AI domination to be a thing.

In which case domination victory should not be an option the AI is able to choose. As I point out, as a simple function of game mechanics the AI cannot win a domination victory - the conditions to do so (player survives with no capital, while every other civ has theirs taken before the player can regain any of them) are in theory possible but not in practice.

An aggressive AI can force a player out of the game - that isn't the same as a domination victory, however, and AIs can do that whether or not coded to aim for a domination victory.

Here is an example of @IronfighterXXX losing to a deity AI on my mod in which I spent a lot of time tuning the AI warmongering part. But you know what ? Ive seen complaints of the AI being overly agressive as a result in the mod. And its true that Deity is really hard. The problem with civ is that there is a lot of different ways to enjoy it and people will like/dislike whatever direction you take. Make a really mean AI and you'll frustrate more casual and peaceful players.

Previous Civ games addressed this in part with AI personalities - certain AIs were more primed to be aggressive than others, and players who disliked them could simply remove them from their game. But how likely a civ is to attack is a separate issue from making the AI somewhat competent militarily - Brave New World is an example of a case where the AI's combat behaviour was more sophisticated and capable than Civ VI's, but AIs were mostly nonaggressive.

Some players just like to think diplomacy should purely be based on modifiers acquired through gifts and allow you to manipulate it. They dislike backstabs with a passion even though its a lot closer to human behavior !

Civ is ultimately a strategy game. Player behaviour should always be capable of manipulating the outcome - but gifts are a crude way of doing it. Backstabs should happen (never seen them in Civ VI), but there need to be ways of at least mitigating the chances of that happening - as in Civ V, with the importance of being a friend of a rival civ's friend/enemy of their enemy.

 

[ashendashin]

There's something I'm curious about regarding Vox Populi. Did anyone actually truly face difficulties fighting the AI? Or was it just a matter of making more units than you normally would to deal with them?

Here's a thread I started over a year ago when the mongols could've ended my game https://forums.civfanatics.com/threads/overwhelming-force-of-mongolians-potential-defenses.607370/
The AI has had massive improvements since then and a new unit supply system was made to lessen the amount of units you have to deal with.

On an even field the human is plain better at fighting, but the AI isn't entirely without bonuses so it takes a lot of planning to take them down if you're playing at the right difficulty; doubly so if you're dealing with warmongers because those guys are genuinely difficult to deal with. The AI will attempt to flank you, perform amphibious assaults, bomb you, blockade your cities, move their units in formation, etc. while deal with other wars at the same time. Efficiently, mind you. Hell of a lot more efficient than almost all strategy games out there anyhow. While the tactical AI did get a dedicated coder, the rest of the AI and balancing VP does also contributes heavily to the overall difficulty of warring with the AI. Diplomacy is a major part as the AI WILL make sensible or even unreasonably hateful bribes for over civs to take you down if you piss them off. They can surprise you with joint wars and even the occasional well timed backstab. It's very easy to end up in a downward spiral if you go conquering without cooperating with the AI. The fact that VP overhauls just about everything in the game obviously includes all things combat related, so keep in mind it isn't so easy to game everything.

All of this after dealing with the many constraints they faced years ago when the project started. Now that many such constraints were dealt with, they could rewrite things to be a good deal more efficient. But it's been years of work revamping a AAA game and there's no more major overhauls in sight. Not like I've seen a better AI for such a complex game anywhere else.

 

[ashendashin]

In which case domination victory should not be an option the AI is able to choose. As I point out, as a simple function of game mechanics the AI cannot win a domination victory - the conditions to do so (player survives with no capital, while every other civ has theirs taken before the player can regain any of them) are in theory possible but not in practice.

An aggressive AI can force a player out of the game - that isn't the same as a domination victory, however, and AIs can do that whether or not coded to aim for a domination victory.

Check this out https://forums.civfanatics.com/threads/monty-prince-ai-domination-victory.629024/

 

[isau]

I've tried Civ 6 with AI+ and while it doesn't fix everything it does make fighting the AI much harder. I'm not sure what the mod changes specifically but they build a LOT more units and are much more aggressive with them. I struggled very much in some of my AI+ games. (Note I also play with my Combined Tweaks mod, which contains some "rule smoothing" to make the AI tend to make smarter decisions as well, or at least be more likely to luck into making good ones).

I remember the AI in Vox Populi being very challenging. Harald declared war on me once in that mod and nearly wiped me out. I'm hoping we can get something like that for Civ 6 eventually It's a real shame Firaxis won't even do a limited release DLL release for the hardcore modders. But I think they might be protecting their business interests in not letting us access the full game until they complete their expansion pack plans.

 

[Ryika]

Every time this issue is raised, some individuals suggest that denying a wonder to hurt other players is a potentially beneficial move. If the AI made such a decision we could argue about that, but that's not what it does. It throws a dart at a board when it decides what to build, and if it lands on wonder it picks a wonder. It doesn't stop and evaluate "hey, none of these wonders help me" and change its mind to building a bank instead. Even if it decided to build a wonder to sabotage somebody else, it's rolling the dice to maybe set back somebody else and definitely set itself back, production down the tubes. An AI is cutting off its nose to spite the face of at most one of 7 opponents, while 6 others gain. It is a myopic argument that rustles me greatly. If another player has a city in position to become a super Petra city, I'm happy. He's going to build a nice thing, and when I decide to take that city off his hands, it will be my nice thing. And I don't buy "nothing better to build". If it is so far ahead it can build wonders willy nilly and win, it clearly would also have an advantage at building units and take more land from other players.

If that's a response to me then no, that's not the argument. The argument is that the wonder system NEEDS competition to work properly and that's why it's good that AI players compete for wonders, even if they're not of much use to them. The system does not need good reasoning and intention on the side of the AI to achieve that, it just needs the AI to do it.

If you can pretty much always get the wonder you want because it's rather specialized and pretty much all other AIs skip it, then the system isn't working as intended. So the AIs are doing the right thing by competing for the wonder, if there's something wrong, then it's that the wonders are too situational*, so the AI doesn't get much out of the thing they've produced. (*Which isn't to say that Wonders should not have situational effects, I think they're very much a step up for Civ V's wonders, but adding some non-situational baseline yields would for example be a potential improvement) That's what the wonder race is for.

You'll probably still disagree with that, but that's because you think of AIs as individuals, as "opponents", while I think of them as an obstacle for the player to overcome. As such, you value the production efficiency of AIs over the integrity of the wonder race, while I think the wonder race is more important. In my opinion, the only primary task for AIs when it comes to wonders is to make sure that you have a somewhat hard time getting them, or at least feel like you have. There are secondary tasks, and they boil down to immersion: Don't construct wonders while you're being attacked if you need more military, don't build wonders that are obviously useless (imho best fixed by changing wonders to be worth racing for, as I explained above), etc.

 

[Siesta Guru]

I've tried Civ 6 with AI+ and while it doesn't fix everything it does make fighting the AI much harder. I'm not sure what the mod changes specifically but they build a LOT more units and are much more aggressive with them. I struggled very much in some of my AI+ games. (Note I also play with my Combined Tweaks mod, which contains some "rule smoothing" to make the AI tend to make smarter decisions as well, or at least be more likely to luck into making good ones).

I remember the AI in Vox Populi being very challenging. Harald declared war on me once in that mod and nearly wiped me out. I'm hoping we can get something like that for Civ 6 eventually It's a real shame Firaxis won't even do a limited release DLL release for the hardcore modders. But I think they might be protecting their business interests in not letting us access the full game until they complete their expansion pack plans.

Ai+ author here. It does quite a few changes, but imo the most important is the faster and more reliable settling. The vanilla AI runs around with its settlers all the time. Getting city X 30 turns earlier allows a lot of extra time to build infrastructure and units. The second thing is that through some changes, it makes it less likely that some units will remain unused. If you watch AI games in vanilla, it regularly has 80% of its units just stand around while a war is going on. AI+ improves that to about 50% (ugh, give the dll please).
Your mod definitely also makes it a lot better. The AI can use the +1 movement really well and escaping settlers are a lifesaver for it.

 

[Magma_Dragoon]

If that's a response to me then no, that's not the argument. The argument is that the wonder system NEEDS competition to work properly and that's why it's good that AI players compete for wonders, even if they're not of much use to them. The system does not need good reasoning and intention on the side of the AI to achieve that, it just needs the AI to do it.

If you can pretty much always get the wonder you want because it's rather specialized and pretty much all other AIs skip it, then the system isn't working as intended. So the AIs are doing the right thing by competing for the wonder, if there's something wrong, then it's that the wonders are too situational*, so the AI doesn't get much out of the thing they've produced. (*Which isn't to say that Wonders should not have situational effects, I think they're very much a step up for Civ V's wonders, but adding some non-situational baseline yields would for example be a potential improvement) That's what the wonder race is for.

You'll probably still disagree with that, but that's because you think of AIs as individuals, as "opponents", while I think of them as an obstacle for the player to overcome. As such, you value the production efficiency of AIs over the integrity of the wonder race, while I think the wonder race is more important. In my opinion, the only primary task for AIs when it comes to wonders is to make sure that you have a somewhat hard time getting them, or at least feel like you have. There are secondary tasks, and they boil down to immersion: Don't construct wonders while you're being attacked if you need more military, don't build wonders that are obviously useless (imho best fixed by changing wonders to be worth racing for, as I explained above), etc.

I think a better solution to keeping wonder races hot would be for the AI to prioritize settling locations where situational wonders can offer good returns. Good Petra or CI cities are not bad without their respective wonder they're just not as good, and always worth settling anyway. The AI already has the capacity to judge whether a city location is desirable (to some degree), I don't think it would be difficult or cpu intensive to pass that function a collection of wonder tile yield modifiers of all wonders not yet built. Changing what wonders do is also an option, placing effects with narrow application later in the tech/civic trees so that more than one player will have a chance to place or capture a city in a location where it would be beneficial.

 

[isau]

The AI issue is the reason I think Civ 6 needs to bring back the Domination rules from Civ 4, where Domination was won by controlling a percentage of the map, not taking capitals. Those rules created extreme high pressure situations, because an enemy AI was very capable of steamrolling the world and winning. It wasn't so much an issue of strong AI as much as the fact that the rule was easy to implement. More tiles = progress toward Domination victory. I lost several Civ 4 games by failing to stop an enemy AI that went berzerk. I miss that high pressure in Civ 5 and 6.

 

[jihe]

I can't understand how a game released 10 years later has vastly inferior AI than its predecessor two generations ago.

 

[LukaSlovenia29]

I'm an avid player of CiV with Vox Populi, where playing the game on Deity/Immortal is extremely difficult. The AI is superb and the game mechanics have been tremendously balanced!

One of the things I most like about Vox Populi is that it changes the game from "Higher difficulties=Bigger advantages to AI" to "Higher difficulties=Better decisions by the AI".

What that means is that on lower difficulties in VP, the AI selects (for example) between the top 3 choices it "deems" to be best, with a small RNG factor to determine which of the three it will choose (but the "optimal" choice has the largest chance to be selected). On Immortal, it selects between the top 2 choices, so it never chooses the 3rd best option, which means its decisions will be better. And on Deity it (almost) always makes the optimal choice, so its decisions are by and large much better on Deity than on King or Emperor difficulty.

Also in VP, the AI is great at using ranged units, citadels, diplomacy, withdrawing, risking units to capture cities, not accepting peace when it has your city nearly conquered, naval warfare, the military aspect has been tweaked to remove the need to set up siege units before firing, thus making it much easier for the AI to use, the religious game by the AI has been greatly improved...

VP AI is the best AI of all the turn-based strategy games I've ever played! I strongly recommend it to every Civ fan.

 

[Acken]

What that means is that on lower difficulties in VP, the AI selects (for example) between the top 3 choices it "deems" to be best, with a small RNG factor to determine which of the three it will choose (but the "optimal" choice has the largest chance to be selected). On Immortal, it selects between the top 2 choices, so it never chooses the 3rd best option, which means its decisions will be better. And on Deity it (almost) always makes the optimal choice, so its decisions are by and large much better on Deity than on King or Emperor difficulty.

Well that already exists in the base AI of civ5. I'm sure the VP team has refined it for the mod though.

 

[ashendashin]

Well that already exists in the base AI of civ5. I'm sure the VP team has refined it for the mod though.

Yeah but the key difference is that the base AI ran on the flavor system and was generally limited in scope, so optimal choices were next none. More of an overhaul than refinement (like most things in the game). Went digging for the exact post from Gazebo, who I won't directly quote 'cause he's done enough for Civ already.

For example, when choosing beliefs, the AI weighs all available beliefs and scores them based on hundreds of variables (yields, synergies, grandstrategy links, policies, relationships, angry neighbors, etc. etc.). Once totaled, the AI ranks them based on weight and chooses from them. At Deity, they always choose the highest-weighted entity. At lower difficulties, they'll choose randomly from the top x (2, 3, 4).

For the record, this has been a huge task on my part. Formerly these were all decided by flavors. If the AI was FLAVOR_NAVAL 10, they would choose all naval things they could regardless of whatever they were doing/needed/stuck in. Land-locked v. Shaka and behind in science? Let me continue to obsess with naval techs and naval-related policies. Stupid AI. Not anymore. Flavors are about 10% of AI logic now (used to be 100%).

They do this for techs, policies, beliefs, buildings, units, wonders, opinions, war desirability, and ideologies.

G

 

[Balkans]

My take on why there's no progress with the A.I., and probably won't be any soon is simply because Firaxis is OK with what they got. On the large scale of things the sales are OK and there is absolutely no reason for them to improve the A.I. in the game. To be more precise, what minority of players want, let's be realistic, is better A.I. That means - get new human resources of developers or reassign 10 of 120 graphic designers to a dedicated A.I. development team to work constantly on A.I. That would bring in the progress and necessary changes. But once again - why would they do that? The game is perfect as it is if you look at it from business perspective and that is all that matters.
Since i was paying close attention from the start until i realized just how bad the game is and there is no chance in hell it will get better any time soon what i concluded is that they simply have too much manpower on eye candies and absolutely no brutal/agile development when it comes to the meat and potatoes of the game. But also i realized that the problem is me, 99% of community is ok with the A.I. and the problems with the game, the 99% of the community think the game is perfect. If i may add 90% of "fanatics" think that the game is fine.
And the guy makes a beautiful analysis and points and documents obvious issues, and Firaxis has people on salaries that probably don't even know about this stuff going on in the game. Quite funny actually. I can see them in their little politically correct stream playing the game completely oblivious to facts like this while presenting their AAA title.

 

[steveg700]

You'll probably still disagree with that, but that's because you think of AIs as individuals, as "opponents", while I think of them as an obstacle for the player to overcome.

The logic you express is based on incorrect assumptions, so the output is flawed. So, there's a lot of reasons to disagree with it.

A big reason though is the presumption that there's always supposed to be this "wonder race"--that wonders serve the purpose of creating a race, rather than the race being incidental to the face that only one can be built.

 

[Ryika]

A big reason though is the presumption that there's always supposed to be this "wonder race"--that wonders serve the purpose of creating a race, rather than the race being incidental to the face that only one can be built.

To claim that it is "incidental" is to claim that Sid (and Bruce?) had no idea what he was doing when he first created wonders in Civ. The wonder race is a natural, and direct result of the way wonders are implemented, as unique buildings of which only one can exist in the world, unlocked by only requiring technologies that are easily accessible to all players. It does not have to be "the purpose" of wonders to create a wonder race (and I did not claim it is), but it clearly is a direct result of the design of wonders. "Wonders" that are not meant to be competed for, are designed as National Wonders and exist since Civ 3. (And no, it's not because World Wonders only existed once in the real world, after all, Oxford University, the East India Trading Company, Forbidden Palace, the Hermitage and Circus Maximus - just to name a few examples - are National Wonders in Civ, too.)

You might be able to make a reasonable case that the wonder race is obsolete in the current iteration of very situational wonders (which I would disagree with, but then we'd be talking about personal preferences), but the fact that there's competition for wonders has never been incidental, it was an obvious part of the design.

 

[ThunderLizard2]

That's a nice idea, but it's not really practical to implement. Better strategy means longer turn times. It also means spending a lot of development resources to tune a more strategic AI that only a small minority of players will benefit from. Besides all of that, making a good, strategic AI is actually quite a bit harder than just piling on bonuses. Just about every 4X game cheats like this. It's simply a practical, non-ideal solution to a very difficult problem.

I don't agree with letting the devs off the hook for creating a terrible product. The Civ IV AI played well and that was with computer technology that was 1/10 of what people have today. It's just poor planning and development by Firaxis and of course the 1 upt system.