Any mod that proves whether players like/dislike AI playing for victory?

[Gori the Grey]

To clarify, there is nothing inherantly wrong with the AI DoWing you to slow you down or because it sees you as weak but it is the obvious, false and poorly implemented way it happens.

I have had strategy games (not just civ) where the "we have to kill the player" check box is activated and everyone just DoWs me even if they are my friend or they are the civ i left alive with one city so I didn't get the 'wiped out a civilisation' massive debuff (even though they probably attacked me [repeatedly] ) like it has a final suicide wish.

A couple of big AIs ganging up on me to beat up on me seems fair enough. The whole world DoWing me because we are X turns from me winning, including a load of small AI who I can roll over in 1 turn is obvious, easy to deal with, boring, unchallenging and often just makes the victory more assured as I can now wipe out all my main competitors as everyone hates me anyway.

It is the lazy implementation that is the problem, not the concept.

Well, yes, I've had one-city civs for whom joining in on the attack was essentially suicidal, and therefore implausible. But the problem there is upstream, in the mechanic you mentioned where if you entirely eliminate a civ, you take a major reputation hit, so you leave all these one-city civs all over the place. Maybe those civs could be programmed to be quietly giving resources behind the scenes to the viable attackers.

Anyway, on @aelf's main point, I don't want to sell modders short, but I think "AI playing to win" would probably have to be built into the game from the ground up. As I mentioned in a post a while back, I think it should be, for the players who do want a real challenge. Then, for the lower difficulty levels, just make the AI less effective at doing so.

Post in thread 'Civ VII single player suffers from the same problems Civ VI did'

Mar 1, 2025

If I were lead designer on a Civ game, my very first brainstorming sessions with the team would be "how can we design the game so that AI can play it well?" I would take each sub-system--city-building, economics, culture, religion, warfare--and ask "how can this be a thing computers can do well?" The big thing computers can do well is calculate, so I think the answer to my questions would often be to make the choices in the game ones where a very precise calculation can ferret out a marginally preferable result (and the sum total of those would keep the AI players competitive).

I think...

• Gori the Grey

• 

 

[Thormodr]

Here is an excellent video on how, IMHO, AI should play the game. AI should play to facilitate the player's experience and trying to win should be secondary.

Playing to Lose: AI and Civilization (GDC 2008)

I gave a talk at GDC 2008 on developing the AI for the Civilization series. I highlighted the difference between “good” AI and “fun” AI and how writing AI for Civ is tricky …

www.designer-notes.com

 

[uppi]

That might be another thing worth polling, because if you are correct, then I am in the minority.

In my ideal civ game, each AI would at some point pick a victory condition. This would come about 100 turns into the game and the decision would be based on the victory for which the civ was best suited in general and their accomplishments up to that point in the game. They would play out the rest of the game min-maxing toward that victory.

There would be ways of my knowing which one they picked and how far along they were. Not with absolute accuracy, but with a certain degree of reliability.

That would make the late game interesting to me, because I would feel the race (that Civ games fundamentally are) as a race. I'd have a sense of what I needed to do to reach my chosen victory condition more quickly than any of the AI reached theirs.

In short, it would feel a lot like playing a boardgame against human players.

But min-maxing their own victory condition is something different from playing to win. (Good) boardgame players will not only min-max their own victory, but also try to prevent others from winning. This is why alliances on boardgames are always assumed to be valid only until the sudden but inevitable betrayal when one of the parties to the alliance makes their bid for the victory. In civ terms, the most concrete consequence of an AI playing to win is that an AI should always backstab you when this would increase the chances of victory. No matter whether you have been friends since the beginning of the game and the relations are maxed out: If the player is in contention for the victory, a backstab is often what would be required to win.

You could even take this one step further and ask yourself, whether the AI should consider the metagame when trying to win? Ganging up on the player or doing Kingmaking in favor of another AI might not increase the chance for the victory in this game, but it improves the chance of the AI to win more games in total. An AI that was really trying to play to win, would organize an anti-player alliance on turn 1

 

[Gori the Grey]

In civ terms, the most concrete consequence of an AI playing to win is that an AI should always backstab you when this would increase the chances of victory. No matter whether you have been friends since the beginning of the game and the relations are maxed out: If the player is in contention for the victory, a backstab is often what would be required to win.

I think I could live with this.

The game could put some brakes on it by having backstabs be detrimental to AI1 with all of the other AIs, and having AI1 make its own calculations as to whether it is in its own interest to suffer those detriments.

 

[ProMeTheus112]

I think I could live with this.

The game could put some brakes on it by having backstabs be detrimental to AI1 with all of the other AIs, and having AI1 make its own calculations as to whether it is in its own interest to suffer those detriments.

Yeah I think I wouldn't mind that either it would make a more chaotic end of game and harder to take the win for the player which isn't a bad thing, in return you could tweak the AI bonuses down a little if necessary to bring overall difficulty to the best place per difficulty mode.. Anyway in Civ7 right now I feel the AI is in a good place in trying to go for a win generally, although it might lack a bit that backstab intent as you approach the win. Even though, I'm not actually sure yet that the main reason it didn't happen in my games, is because in these they didn't have the comparative military to gain any advantage from declaring war. Because in other scenarios I have seen AIs I think doing some backstabby things if your military is lower than theirs.

Come to think of it I also haven't seen them try and sabotage railways when going for economic win for example. Counterplay in general is cool in a strategy game.

 

[aelf]

Civ4 AI does go for the culture victory in a planned manner. (Which is not to say they are good at it.) The AI culture victory strategy has several phases. They can enter the first preparatory phase VERY early in the game. In these preparatory phases they found and spread religions, construct culture buildings (in particular the religious culture multiplier buildings like Cathedrals). Then in the last phase they use the culture slide, directing all of their empires commerce into culture for the last sprint to victory. Civ4 AI will not go after domination or spaceship victories in such a cohesive manner. They can win either victory when they snowball hard enough. In particular, they build spaceship parts with a very high priority when they become available. So when they do reach the end of the tech tree they will launch a spaceship in very short order (2-4 turns or so on high difficulties). They do not prioritise spaceship techs over others.

Oh boy, I had to go back and read up on culture victory in Civ4 because I recall nothing about it. I'd never seen it happen. Never did it and I don't think I ever saw the AI actually get close. I have some vague memory that this is a victory type that, if pursued by the AI, can creep up on you all of a sudden without much of a warning. But it seems so difficult to pull off that it's almost never a serious factor.

Anyway, on @aelf's main point, I don't want to sell modders short, but I think "AI playing to win" would probably have to be built into the game from the ground up. As I mentioned in a post a while back, I think it should be, for the players who do want a real challenge. Then, for the lower difficulty levels, just make the AI less effective at doing so.

You're probably right. The closest example in the Civ series is probably what others have said about the AI ganging up on the player when the player is going to win, whether or not it makes sense.

 

[Noble Zarkon]

In civ terms, the most concrete consequence of an AI playing to win is that an AI should always backstab you when this would increase the chances of victory. No matter whether you have been friends since the beginning of the game and the relations are maxed out: If the player is in contention for the victory, a backstab is often what would be required to win.

That sounds like Vanilla Civ V and players hated it as it totally destroyed the immersion.

I don't know of any Mods for this type of question but if we look at Civ VI Steam achievements for victory levels:-
Settler 37%
Chieftain 33%
Warlord 31%
Prince 29%
King 14%
Emperor 10%
Immortal 7%
Deity 6%

I know on CFC the stats would look very different but the vast majority of players play at the easier levels for what that's worth.

 

[Gori the Grey]

The closest example in the Civ series is probably what others have said about the AI ganging up on the player when the player is going to win, whether or not it makes sense.

I thought Civ V handled it well. It usually comes after you've conquered your third capital, so you've made it clear to the entire world that you're a menace. In the diplo screens, you've been getting the "they have concerns about your warmongering" message. It's a plausible response on the part of the world to your actions.

If it comes again just to stop you from winning the game, that has built into it the historical implausibility that "how do these civs know we are reaching the 'end' of history?" But that implausibility is built into there being victory conditions at all, so it's one I've already swallowed, just to play the game, as a game, at all.

 

[itix]

You could even take this one step further and ask yourself, whether the AI should consider the metagame when trying to win? Ganging up on the player or doing Kingmaking in favor of another AI might not increase the chance for the victory in this game, but it improves the chance of the AI to win more games in total. An AI that was really trying to play to win, would organize an anti-player alliance on turn 1

Not necessarily the anti-player alliance. MicroProse / Sid Meyer used to make games where AI was doing that and it was annoying.

A competitive AI would eliminate the weakest civs or capture/raze weak cities. In Civ 1, AI was doing that, probably by luck than due to smart programming, but it led to stronger civs. This made games more interesting because instead of 6 lousy/mediocre civs against you, there were 1 or 2 (semi-)strong civs.

 

[Uberfrog]

I don't know of any Mods for this type of question but if we look at Civ VI Steam achievements for victory levels:-
Settler 37%
Chieftain 33%
Warlord 31%
Prince 29%
King 14%
Emperor 10%
Immortal 7%
Deity 6%

I know on CFC the stats would look very different but the vast majority of players play at the easier levels for what that's worth.

Care must be taken when interpreting the civ 6 achievement stats, as these achievements are actually “Win a regular game at X difficult or harder.” So the % winning at each level also includes the % achieving anything above that level. On top of that, the stats show (as is commonly known) that the majority of players don’t win any games at all.

So adjusting for these issues, the actual breakdown of the highest difficulty level achieved by players who win their games is*:

Settler 11%
Chieftain 5%
Warlord 5%
Prince 41%
King 11%
Emperor 8%
Immortal 3%
Deity 16%

So actually most players who can win at Emperor or above also win at least one game at Deity. The achievements don’t tell us what level people usually play at, but this is the best we can do.

Of course you can win Deity by cheesing it, but I don’t think enough players care about cheating their way to 100% achievements to really skew the stats. Indeed, you would expect such a completionist to at least have beaten the game with the least popular vanilla leader (Gandhi), and only 3.3% of players have done that…


* subtract from each percentage the percentage achieving the level above, and divide by the percentage achieving victory on Settler or above.

e.g. for Settler difficulty we have 37 - 33 = 4% of players winning on Settler and no higher. Which is 4 / 37 = 10.8% of those who have won on any difficulty.

 

[Melkus]

The 14% for King (or harder) is the most surprising to me here - I'd say that's a solid measure for searching any kind of challenge and seriously engaging with a game until the end?
Not that much.

I mean - if a number cruncher at 2k would have to allocate resources for AI development - what would he conclude from this? (That's not quite how it works in the real world, luckily, but they still have to justify the dev hours.)

 

[Melkus]

I'd still like to see meaningful penalties for the human player on higher difficulties. If this is part city builder, part competitive strategy - make both parts harder.

Look at this (badly formatted, but you'll manage) table from Civ IV:

Difficulty level (Civ4)

Back to Civilization IV The difficulty level is a game setting in Civilization IV and its expansions that allows players to compete against AI opponents matched to their skills and desire for challenge. In order from lowest to highest, the difficulty levels are as follows: Settler Chieftain...

civilization.fandom.com

iFreeUnits, iUnitCostPercent, iResearchPercent, iDistanceMaintenancePercent, iInflationPercent etc. etc.

Yes, these are serious difficulty modifiers for the player. That's a major reason why it's so hard - even when you're playing on your own, you run into problems you wouldn't have on lower difficulties - early and often.
And it doesn't throw the balancing and pacing of the whole game off. That's why the game setup in IV tells you that the AI plays on the normalized difficulty - Noble (You can't change that though and they still get bonuses, of course. I'm just pointing out that AI bonuses are not the only way to make a game harder.).

They have completely gone away from that in VI - that's why its difficulty curve is the silliest yet.
VII - which is comparatively well-designed for AI in general, just not "there" yet - reintroduces slowing tech rates at high difficulties at least - but ironically science is not that important anymore in this game. With a more Soren-Johnson-like design philosophy, the player would have lowered settlement limits on Immortal and Deity - just as an example.

 

[Noble Zarkon]

Prince 41%

Thanks for doing this, this makes sense - most people playing on the "even" level.

 

[Mark the Bold]

Not necessarily the anti-player alliance. MicroProse / Sid Meyer used to make games where AI was doing that and it was annoying.

A competitive AI would eliminate the weakest civs or capture/raze weak cities. In Civ 1, AI was doing that, probably by luck than due to smart programming, but it led to stronger civs. This made games more interesting because instead of 6 lousy/mediocre civs against you, there were 1 or 2 (semi-)strong civs.

I could not scientifically reproduce it, but sometimes your map conditions would create fantastically competitive AI's. One Civ 5 game I recall fondly, a Gaia / Large / Immortal map had two civs Persia and Iroquois just steam roll all the other civs so it was just me and them two by mid game (almost Orwellian). Both AI's had enormous armies and (shockingly to me) effective Navies working together. They ganged up on me, and it was all I could do to stay afloat. Then a miracle happened, they both turned on each other and the Iroquois got the upper hand between the two. This allowed me to recover and win the game (scientifically).

I never once had a similar game experience either before and after. I don't know what 0's and 1's combined to make this situation happen, but it was 100% pure kino and fun. I coudn't believe my eyes when Persia surprise attacked me by crossing the border, in 2 turns landing considerable forces behind my peninsular defense force AND bombarded said force with substantial Naval ships. It felt like being attacked by a human being who actually was trying to win, in the Clausewitzian sense. They totally annihilated my large, but in hindsight, POORLY positioned main veteran army in 8-10 turns.

It would be great if Firaxis actually leveraged the power of AI to make more challenging opponents rather than newer map texture assets. I'd pay for THAT DLC.

t. blog post

 

[Marla_Singer]

What is the expected challenge with AI players is certainly one of the things that lost me with Civ successive games over time.

Civ1 and Civ2 could be considered somehow as "battle royale" games in which AI players were regularly conquering one another, pressuring the Human player to do the same in order to stay in the race for victory. Less than 50% of the initial civs on the map survived at the end of the game. And among those surviving, some were considerably stronger than others (like 10 times stronger).

Civ3 and Civ4 had more of an "AI vs Human" system in which you were still threatened by the AIs, but AIs were not conquering one another as often. The challenge was still there because if you wouldn't be careful the AI could still throw at you an invading army that you couldn't counter (some people enjoyed that as a challenge whereas it made rage quit others).

Civ5 and Civ6 is where I'm getting lost (I think I don't understand those games). On one hand, combat is supposed to become more advanced with 1UPT bringing tactic, but on the other playing peacefully is more encouraged and military conquest isn't as central. The AI could still try invading you but 1UPT makes it easier to counter it. Most, if not all AI players that were there at the start are still there in the end. So I assume the expected challenge of the game is about achieving victory conditions like culture, science or religion that fundamentally relies in snowballing mechanics.

If my description is correct, with no disruption such as an AI conquering other ones to suddenly grow more powerful and threatening, then it only makes sense that the final outcome is already known mid-game. I would even say that is true by design. But then, Civ5 and Civ6 were, and are still, successful games nonetheless, so there is undoubtedly something else that motivates players. Without knowing well those games, my first instinct would be to say that, if victory conditions mostly relies on snowballing mechanics, then a more challenging AI would be one that would snowball as fast as the Human player, or that could catch it back.

Am I totally wrong or does this help?

 

[Taefin]

What is the expected challenge with AI players is certainly one of the things that lost me with Civ successive games over time.

Civ1 and Civ2 could be considered somehow as "battle royale" games in which AI players were regularly conquering one another, pressuring the Human player to do the same in order to stay in the race for victory. Less than 50% of the initial civs on the map survived at the end of the game. And among those surviving, some were considerably stronger than others (like 10 times stronger).

Civ3 and Civ4 had more of an "AI vs Human" system in which you were still threatened by the AIs, but AIs were not conquering one another as often. The challenge was still there because if you wouldn't be careful the AI could still throw at you an invading army that you couldn't counter (some people enjoyed that as a challenge whereas it made rage quit others).

Civ5 and Civ6 is where I'm getting lost (I think I don't understand those games). On one hand, combat is supposed to become more advanced with 1UPT bringing tactic, but on the other playing peacefully is more encouraged and military conquest isn't as central. The AI could still try invading you but 1UPT makes it easier to counter it. Most, if not all AI players that were there at the start are still there in the end. So I assume the expected challenge of the game is about achieving victory conditions like culture, science or religion that fundamentally relies in snowballing mechanics.

If my description is correct, with no disruption such as an AI conquering other ones to suddenly grow more powerful and threatening, then it only makes sense that the final outcome is already known mid-game. I would even say that is true by design. But then, Civ5 and Civ6 were, and are still, successful games nonetheless, so there is undoubtedly something else that motivates players. Without knowing well those games, my first instinct would be to say that, if victory conditions mostly relies on snowballing mechanics, then a more challenging AI would be one that would snowball as fast as the Human player, or that could catch it back.

Am I totally wrong or does this help?

I joined civ in VI, and recall discussions of how AI used to conquer each other back in the day. I tuned some game parameters to dramatically increase AI damage against cities, and increase their risk taking in attacking them, and AI did begin to conquer each other, with powerful AI civs emerging.

I am getting ready to start slowly working on how to do this in VII, since I agree that the game is enhanced when this can and will happen.

In my view, this makes games more distinct and expansionist AI civs more of an unpredictable threat that you need to keep your eye on and play around.

 

[Noble Zarkon]

Civ3 and Civ4 had more of an "AI vs Human" system in which you were still threatened by the AIs, but AIs were not conquering one another as often

AI conquering was in Civ IV and is back in Civ VII (in my experience - e.g. in my current game Hatty eliminated Catherine in the Antiquity Age) having been missing in V and VI.

 

[aelf]

I remember conquering cities in Civ1 and 2 is easier, especially in late game. Howitzers are really strong, and you can ride on the enemy's railroads to get to the target in one turn. It's possible to conquer an entire railroaded empire within one turn.

That's probably a big reason why AIs could conquer each other more often.

 

[Noble Zarkon]

It's possible to conquer an entire railroaded empire within one turn.

You are bringing back some memories lol

 

[Siptah]

In 5 and 6, with cities having a rather powerful ranged attack and substantial HP even without walls, conquering was made much more difficult for the AI. Then cities also gained auto-heal each round that can only be avoided by sieging the city, with requires 3 correctly positioned units. This proved to be too difficult for the AI in 6, especially when the defending AI can put a new unit on the field each turn, while reinforcements for the attack were slow to trickle in.

I think 7 has improved a lot in that regard. No more ranged attack by the city itself, and unwalled settlements have low HP. Especially unwalled towns are easy to conquer for humans and AI, because they are often not garrisoned, and can be taken by an army in a turn. And yet, the game offers an option to defend, if the AI or player invests: cities with walls are per se harder to take over, and in most cases require multiple turns. If a city has 5+ fortified districts, the battle can easily go on for a while, even with some artillery, and reinforcements and logistics start to play a role. But this is then also the time when the AI stops conquering others in my experience: in antiquity, a civ might be wiped out, but I haven't seen it later in the game, when they usually have at least one well-fortified city.

Still, the biggest threat in civ 7 for your cities isn't the other players, but the coastal independent powers. If they arrive at your freshly founded town with 3 galleys... argh!