Civ VI AI not THAT bad

[Bizrock]

AI is stupid, you can beat a larger army just with positioning and the AI will keep killing his own units. AI will only be significant and competitive with a cheat. Impressive how AI can be bad fighting in a "chess style" game.

 

[Disgustipated]

I seen Rome conquer a ton of cities my previous game, but that was with the AI+ mod. I was getting some weird crashing issue, so not able to finish that game. I probably wouldn't play with that mod anyways, I'm not always a big fan of how other people want Civ to be. Yes a powerful Rome is kind of awesome, but that's at the expense of all the other civs that got crushed by the Roman onslaught. It's weird to give one specific civ this massive aggression and not the others. Anyways, that's with a mod. My most recent game on the same map with no mods, Rome did fairly well, but not nearly as well as with the AI + mod. I had barbarians off, so Poland was able to expand quite a bit and was in 2nd place. Still the problem of the AI focusing too much on holy districts. A lot of good those holy districts do when I'm going into space.

I think in the future I may just try to limit the amount of cities I have. I could go up 2 levels, but that style of play isn't fun for me, I like being able to build wonders. Self limiting myself to under 10 cities may be the way to go, and keeping barbs turned off.

 

[conorbebe]

The AI isn't that bad because if it was, Firaxis wouldn't have a game. But I think most people agree that the AI could be vastly improved without having to completely overhaul the system. Simple adjustments like making AI cities prioritise unit production whilst under attack would make a big difference.

 

[ThERat]

The AI is really bad in this game. It is magnified in R&F as the computer has got no clue how to deal with the new features at all. Has anyone seen the AI making good use of the most obvious feature, the Magnus-chop? It doesn't even know what is that. I can play on Immortal and get bored after 100 turns as the AI is just not posing any serious challenge.
In fact the only challenge you get in terms of warfare are barbarians as they get units out of the blue sky and often, 1 unit a turn if a scout manages to run back home. I would disable huts and barbarians but the Age scores are tied to them and if you want an early GA, those are necessary elements of the game.

I wish, we can get some competent people who are willing to spend time to improve the AI. Currently, I can only play PBEM to have some fun.

 

[darko82]

This game is designed to be played against yourself. The AI pretends to be there. It's just an illusion.

On high difficulty, it should at least pose a big threat to a human player, some challenge. And if it cannot play properly, if it's buggy etc. it cannot pose any threat. And this is the point when the fun ends.

 

[sonicmyst]

How do you know the AI is not using Magnus-chop?

I have seen the AI chop everytime. They even use it to defeat me in wonder building. I have also seen the AI utilize the emergencies well (though not all the time) and use Amani to snatch suzerainty.

Having said this, I still believe the AI has still more room for improvement. Especially with use of air units and nuke.

 

[seankhan]

the ai is bad all round.its resorted to 1 turn building units and still fails.they might aswell go back to stacks because having a unit on each tile doesnt cut it.its weakness is it doesnt play a domination game and looses because of this.this is true on all levels of gameplay.infact there are no harder levels.

 

[Sostratus]

Both times it was a city state ally that took the cities while my army was fighting the major civ.

I think the AI itself is pretty decent at fighting- if it has the right units together in the right spot. City states all build a decent military and keep them close to their city- so when they go to war, they are ready to deploy. It's quite interesting to watch a CS move its soldiers and take cities. I suspect that however the AI is programmed, a major civ often has it's units not in the right spots to leverage the tactical programming it could execute if they were. Perhaps some kind of adjustments to make the major civ ai function more like city states wrt military- perhaps literally encamping groups of units around a few of its cities so it can direct them as strike forces in war time.

 

[orasis]

The only strategy game I have ever played that had a pretty good A.I was Galatic Civilizations II. So... I mean it's not like this is the only franchise with a so-so default AI.

 

[Disgustipated]

I do sometimes get impressed when one AI unit sidesteps to allow another AI unit in range to attack, and so both units can ultimately attack. This actually seems more recent, but maybe I never noticed before.

 

[Trav'ling Canuck]

I do sometimes get impressed when one AI unit sidesteps to allow another AI unit in range to attack, and so both units can ultimately attack. This actually seems more recent, but maybe I never noticed before.

I, too, have been impressed by some of the small things the AI does well, including coordinating ranged and melee attacks, bringing up units to get flanking bonuses, etc.

It's the bigger picture things that remain the issue. For me, personally, I'd prioritize:

• Air units: making them, using them. Whatever they need to do to bring air combat into the game, I'd like to see them do. Not sure if that's getting rid of the complexity introduced by the "patrol" command, or lowering the insane production costs of air units, or letting cities build air units without building an aerodome first. Something. Oh, making the game more challenging so you don't have to sandbag to avoid winning before the AI can build planes would be nice, too
• More flexible targeting system, or at least a more finely tuned system. So that sometimes they'll pull back when injured, sometimes they'll focus on your units, sometimes they'll focus on your city, etc., ideally with some parameters that help them to make choices that are more likely to be the right ones in the circumstances.
• Greater clarity on what the AI is trying to do. Some of the frustration is not understanding what the AI is trying to achieve at any given time. So we extrapolate to the AI what we think their objective should be, which may in some circumstances make their behaviour seem even odder. I'm not sure how best to implement this, but perhaps some linkage to the diplomacy system could give you some level of visibility on their strategic objectives during a war?

 

[Lily_Lancer]

Agreed. In Civ6 the time when AI have the biggest advantage is at the beginning. If AI start attacking you at T0, 5 warriors can easily overrun 1 warrior, however you play.

In fact in PVP games when I have knight at for example, T55 standard speed, my enemy cannot defend, too. Despite you are human brain or not, you can't use 28-strength chariots to defend 48-strength knights. My human opponents are actually losing cities quicker than AIs since they don't have walls.

 

[abandag]

I think it's getting better, a little bit at a time. In a recent game I took a small force to attack a city. It was comparable to a military size I would have easily taken the city with a few months ago, but they made some smart decisions like building (maybe buying) some chariots to give my archers some trouble. I retreated and came back a little later with a slightly bigger force and took their city. This time they didn't build any more units in the city despite the fact they had plenty of cash (they offered it to me in peace deal. No, I wanted the city). So it's a little better, then bad again.

 

[Eliminator_Sr]

Simultaneously, the people who act like it's egregiously game-breaking and endlessly whinge about it are typically exaggerating to reach their preordained conclusions or innermost desire to hate the game no matter what. You can see this same pattern here on the forums with other supposed pitfalls of the game's design.

I actually really like the game though I did stop playing vanilla solely because of the AI. I picked it back up with R&F and I'm enjoying the expansion but I still don't see much significant improvement with the AI.

It kind of boggles my mind that people are actually defending it. I mean come on - I've finished at least a few games with kill counts on my side in the hundreds where I have not lost a SINGLE UNIT.

 

[darko82]

I actually really like the game though I did stop playing vanilla solely because of the AI. I picked it back up with R&F and I'm enjoying the expansion but I still don't see much significant improvement with the AI.

It kind of boggles my mind that people are actually defending it. I mean come on - I've finished at least a few games with kill counts on my side in the hundreds where I have not lost a SINGLE UNIT.

Exactly. I like the new features and design. But the stupid AI kills all the fun. Simple as that.

 

[earlc]

I actually really like the game though I did stop playing vanilla solely because of the AI. I picked it back up with R&F and I'm enjoying the expansion but I still don't see much significant improvement with the AI.

It kind of boggles my mind that people are actually defending it. I mean come on - I've finished at least a few games with kill counts on my side in the hundreds where I have not lost a SINGLE UNIT.

Sounds like a self-fulfilling prophecy.

Like I said, there are things that can probably be fixed. But if you go into the game fixated on the AI and what it can or cannot do, you're setting up your game to revolve around that issue, and likely, precisely how to exploit it.

The AI has surprised me a number of times - and if it matters, I am a long-time civ player who regularly plays on immortal or deity, but also kicks it down to emperor and king for certain games. I recall one pre-R&F, pre-fall patch game where Norway stormed my borders with probably 20 knights - from across the ocean. Just recently, Ghandi unexpectedly had an absolutely killer navy and nearly took a city before I sought peace. In a separate game Korea had early caravels and did take one, and another with her land force. In a fourth I tried to attack Kongo with a larger army than I typically build and actually got knocked back without managing to take a single city, which was the biggest surprise yet (granted, the Kongo had one civ between as well as some mountains that prevented me quite getting into position).

The clearest parallel is the barbarians issue (which is still a real issue if too many people are having problems with it) - people allege the game is impossible to play because barbarians overrun your borders in every game, but I regularly mop them up with just a scout, slinger and warrior, maybe buying another unit or two if I get in a pinch. And that's across hundreds of games.

So my 'defending' it is less saying there are zero things wrong with it than casting doubt on the sheer game-breaking terrible inadequacy of it all. Personally, I could see some improvements to the way the AI builds and manages units, and if possible (people have raised issues of computing power and coding labor) combat tactics too. I'd count air combat as the biggest failure; even with the small improvements lately, it still has a way to go in actually building (which it does some of now) and using air units. But even that I'm not too upset with. Late-game domination is boring anyway.

 

[Siesta Guru]

Greater clarity on what the AI is trying to do. Some of the frustration is not understanding what the AI is trying to achieve at any given time. So we extrapolate to the AI what we think their objective should be, which may in some circumstances make their behaviour seem even odder. I'm not sure how best to implement this, but perhaps some linkage to the diplomacy system could give you some level of visibility on their strategic objectives during a war?

It's not really trying to do anything usually. There's only one real thing it has in terms of 'strategic objectives' and that is that it can send a group of units to a city.
Anything beyond that is just trying to patterns where there aren't any. It sends units to an enemy city under these three circumstances:
- it has available units
- it sees a city with high enough odds of capturing the city (afaik, thats just a simple combat str based formula)
- it's at war with the defender or the diplomacy system allows a declaration of war

That's all there is to it. It doesn't like evaluate the worth of a city capture, it doesn't try to find good terrain, it doesn't set up a plan of attack involving multiple conquests or something.

As for defending, it has a 'group of units' thing too, but it's usually fired at inappropriate times/at inappropriate cities so you rarely see it.
Almost all defending you see is random movements causing units to bump into yours, or sometimes you end up meeting an attacking group that was going for your cities.

 

[Trav'ling Canuck]

It's not really trying to do anything usually. There's only one real thing it has in terms of 'strategic objectives' and that is that it can send a group of units to a city.
Anything beyond that is just trying to patterns where there aren't any.

Then no wonder the AI fails on big picture objectives, if it hasn't been coded to have one. I would have thought the AI would be set up to have, at the first level, one of a number of different game objectives, set either at the beginning of each era or every X number of turns or after major events (is declared war upon, for example). Things along the lines of the following:

• Survive: build decisions, diplomacy, research, unit behaviour, is all geared towards survival, as the civ is under immediate military danger from a superior foe.
• Grow/Expand Peacefully: the above are devoted to general expansion of the empire, with military being brought up only to a specified amount of defensive forces. More likely if they have peaceful relations with their neighbours.
• Grow/Expand Militarily: the same, but they will create a larger military and attack a militarily inferior neighbour. More likely if they have poor relations with their neighbours.
• Pursue Victory: as per the above, but focused now on a single victory objective, instead of overall expansion. More likely if they are leading in a particular victory type.
• Delay Victory: decisions are tilted towards slowing down rivals who are viewed as winning. More likely if they have poor relations with rivals who are leading in a victory type.

I'm shocked that in a game as long and complex as Civ 6 the developers would set up the AI to follow a single logic tree for all decisions without setting some overriding parameters to coordinate those individual decisions.

 

[Siesta Guru]

I'm shocked that in a game as long and complex as Civ 6 the developers would set up the AI to follow a single logic tree for all decisions without setting some overriding parameters to coordinate those individual decisions.

There's a tiny bit of that in the xml. Like a slightly higher preference for science if pursuing a science victory. But it's all very minimal and based on iffy conditions (it doesn't even have a higher desire to build units while invaded etc.). It's nowhere close to the level of intricacy that you describe, while the complexity you describe should honestly be some kind of minimal baseline. It's really no different on the small scale either, it just doesn't really have sophisticated behavior for any aspect of the game.

There's a reason that I'm one of the most vocal people on AI shortcomings while also probably being the most knowledgeable about the civ 6 AI outside of firaxis employees thanks to my work on AI+. From how much simple XML based changes do, and how predictable the bots are, I can tell just how little else there can be inside the internal codebase.
It's like they tried at all costs to move their entire AI system into xml for whatever reason (modability?), without stepping back and realizing that a couple of variables and connections is just never going to allow making an actually powerful AI. I imagine they kind of just gave up at some point because of how draining and time intensive it is to try to use xml to control a bot.

The only reason civ actually offers some challenge is because the game itself is set up so that it just kind of inevitably causes you to win through building random horsehocky. Add bonuses, a bit of luck, and that players feel ideally challenged when winning 90% of the time against 8 opponents, and you've got yourself just enough illusion of AI competence to make some say it's all fine.

 

[isau]

There's a tiny bit of that in the xml. Like a slightly higher preference for science if pursuing a science victory. But it's all very minimal and based on iffy conditions (it doesn't even have a higher desire to build units while invaded etc.). It's nowhere close to the level of intricacy that you describe, while the complexity you describe should honestly be some kind of minimal baseline. It's really no different on the small scale either, it just doesn't really have sophisticated behavior for any aspect of the game.

There's a reason that I'm one of the most vocal people on AI shortcomings while also probably being the most knowledgeable about the civ 6 AI outside of firaxis employees thanks to my work on AI+. From how much simple XML based changes do, and how predictable the bots are, I can tell just how little else there can be inside the internal codebase.
It's like they tried at all costs to move their entire AI system into xml for whatever reason (modability?), without stepping back and realizing that a couple of variables and connections is just never going to allow making an actually powerful AI. I imagine they kind of just gave up at some point because of how draining and time intensive it is to try to use xml to control a bot.

The only reason civ actually offers some challenge is because the game itself is set up so that it just kind of inevitably causes you to win through building random ****. Add bonuses, a bit of luck, and that players feel ideally challenged when winning 90% of the time against 8 opponents, and you've got yourself just enough illusion of AI competence to make some say it's all fine.

In fairness to you, Siesta Guru, Arabia and India did stomp me in my recent game as the Khmer when they dual-declared as part of an emergency and showed up with (buffed by Quo's Combined Tweaks mod) Catapults and Archers, which they actually used fairly well. But I'm in total agreement with you about shortcomings of the AI's master strategy. Just wanted to put it out there that you've done some nice work, to the degree it is possible with the systems Civ 6 gives us.


EDIT: Also might be worth mentioning that Civ's 4's AI didn't pursue Culture Victory at all until the Beyond the Sword expansion. That's back when to win a Culture Victory the only requirement was "3 cities at max Culture level," nothing close to what Civ 6 evolved into via expounding on Civ 5's second expansion pack.