Civ VI is SO close to greatness - A call to the developers

[Archon_Wing]

Vanilla did have a mechanic where the AI would see if you're winning and proceed to hate you, shutting off diplomacy. That proved so unpopular it got removed, so I don't think we'll ever have a cutthroat AI... it's also probably why they didn't want diplo victory either, since well, that's the antithesis of trying to win.

Being able to share a victory condition is probably the only viable path.

I may be a minority on this issue, but the diplomatic game is pretty good. World Congress needs fixing though.

 

[oSiyeza]

I think if they wanted to improve the game some more, they could nerf coastal civs, maybe reduce the utility of navies, and make global warming happen earlier and more gamey.

You sir play with fire. One day a FXS employee will take you seriously and grant your wishes. And This forum will hunt you down xDDDD

Part of the big issue from a developer perspective is that we actually disagree in what we want and which areas should get priorities, as we disagree on the definition of a "good AI".

Some people want an AI that is able to win the game at all cost, following the game's rules the best it can. Some people want an AI that is able to provide a challenge without causing frustration by abusing the game's rules. Some people want an AI that is able to RP a Civilization with more logical diplomacy. Not even talking of the game's mechanisms that would have to be changed to fit each type of AI (we can't have a good RP AI with the current diplomatic rules and victory conditions IMO)

Which means for an AI to please everyone, it would require a lot of options (and as many different code sections)

I don't agree. An improved AI with more reliable path-finding, better management of units and specially groups of units, more effective in air and naval operations, better in managing and obtaining resources and more focused in the victory conditions that suit the game state and the leader agendas will fit the definition of a better AI of everyone here (I can be proved wrong but that would be a surprise).

That being the case, it is possible to greatly improve the AI without making any compromise to any type of game play. An AI more capable of using the current game systems, more focused and better at using units and resources will be good for challenge and role play and will not be frustrating to casual players due to the big flexibility in difficulty options the game has.

PS: I will name this a "It can't be agreed upon extreme extrapolation" fallacy xDD

 

[Bibor]

will fit the definition of a better AI of everyone here

I'm all for better AI and dropping difficulty to Prince. Apparently everyone can play Deity these days, as if its a challenge. The consequence of Deity is that you end up with more cities sooner than you usually would.
At least Civ4 had the common decency to completely crash your economy whenever you were rushing a deity AI.

 

[tab]

I agree with Bibor that the AI needs to treat humans as the biggest threat to victory especially on higher difficulty levels. Even if some people are not willing to lower difficulty settings, I think a simple checkbox would suffice that would enable/disable "Cutthroat AI" for people who dont want/like this challenge just like some people dont like barbarians and can disable them. Then they can have that sim city experience that they want whereas for us warmongers it would satisfy the need to stomp on an AI that really went for your gut.

 

[Unconquered Sun]

I agree with Bibor that the AI needs to treat humans as the biggest threat to victory especially on higher difficulty levels. Even if some people are not willing to lower difficulty settings, I think a simple checkbox would suffice that would enable/disable "Cutthroat AI" for people who dont want/like this challenge just like some people dont like barbarians and can disable them. Then they can have that sim city experience that they want whereas for us warmongers it would satisfy the need to stomp on an AI that really went for your gut.

A zillion years ago I proposed a "Gamist AI" checkbox to differentiate from the regular "Roleplay AI". Back then the series even had an "Aggressive AI" checkbox but it actively made the game easier because the AIs trashed their research while attempting to warmonger.

That being said, Civ 6 problem isn't just the combat AI. Part is mechanics that break immersion (like terribad trades), part the AIs are simply too slow with development and victory dates (and heavily 'frontloaded' with bonuses in the very early game). The latter need not be addressed with AI tweaks at all: recalculating their bonuses to research, yield, and upgrades will create a marked improvement with very little effort from Firaxis.

 

[ashendashin]

I don't agree. An improved AI with more reliable path-finding, better management of units and specially groups of units, more effective in air and naval operations, better in managing and obtaining resources and more focused in the victory conditions that suit the game state and the leader agendas will fit the definition of a better AI of everyone here (I can be proved wrong but that would be a surprise).

That being the case, it is possible to greatly improve the AI without making any compromise to any type of game play. An AI more capable of using the current game systems, more focused and better at using units and resources will be good for challenge and role play and will not be frustrating to casual players due to the big flexibility in difficulty options the game has.

PS: I will name this a "It can't be agreed upon extreme extrapolation" fallacy xDD

I think you put it better than I ever managed. Probably because I 'tainted' my arguments a tad much with VP, which is largely discussed and played by experienced players who Gazebo pleases by actively upping competition/balance while making general AI improvements. The AI is a buggy mess with poor foundations(safe observation/assumption from all examples, especially civ5 as that's been proven), so it's not as if FXS even really tried for a middle ground in fear of scaring off anyone. I don't understand why such extreme assumptions are made when the basis of what we're arguing about revolves around something that was never built properly in the first place.

 

[Gedemon]

I don't agree. An improved AI with more reliable path-finding, better management of units and specially groups of units, more effective in air and naval operations, better in managing and obtaining resources and more focused in the victory conditions that suit the game state and the leader agendas will fit the definition of a better AI of everyone here (I can be proved wrong but that would be a surprise).

That being the case, it is possible to greatly improve the AI without making any compromise to any type of game play. An AI more capable of using the current game systems, more focused and better at using units and resources will be good for challenge and role play and will not be frustrating to casual players due to the big flexibility in difficulty options the game has.

PS: I will name this a "It can't be agreed upon extreme extrapolation" fallacy xDD

Ho, globally, we agree , yes.

but when we go into specifics:

"An improved AI with more reliable path-finding, better management of units and specially groups of units" :

IMO just change the 1 combat UPT rule to n combat unit per tile where n is a finite number, n > 1 and n < stack of doom. Then you can improve your tactical AI on those rules (groups compositions per tile), but it will perform better already (pathfinding and formations partially solved) and you can spend more development time on other areas.

But a lot of people don't wan't to change the tactical aspect of 1UPT combined to 2 tiles ranged units, and while, yes, you can make a better AI for those rules, you may have to spend a lot of development time on it (surely less than what the VP's team had spend on it, but still some time)

And I'm pretty sure that if we can make a good tactical AI for 1UPT in x hours, then we can make a better one for nUPT in less hours.

The question about which way to go is then "how much better do you want the AI to be and when do you want it ?"

"more effective in air and naval operations" :

That one I think is the point everyone agrees on.

"better in managing and obtaining resources" :

IMO, again, more a problem of design than pure AI, strategic resources are highly RNG dependent, without a lot of alternate ways to get them (while the AI may require the strategic before being able to conquest it for example)

So first add other possibilities for acquisition than trade or conquest for a player that has none in his territory: small income in strategic from each mines even without deposits, a new smuggler unit similar in usage than the trader, etc...

Then see if the AI still need improvement here.

"more focused in the victory conditions that suit the game state and the leader agendas"

I disagree on that one and I'd change it to "build an Empire to stand the test of time" or I would accept it as a priority if we first change the rules to create different "games states" than the current ones. OFC, it's just an opinion, part of the RP crowd, I don't mean it should be that way, I'm just pointing again that, yes, opinions may differ on what/how.

A zillion years ago I proposed a "Gamist AI" checkbox to differentiate from the regular "Roleplay AI". Back then the series even had an "Aggressive AI" checkbox but it actively made the game easier because the AIs trashed their research while attempting to warmonger.

Remember that one, used it as the AI was not agressive enough otherwise. And it's a good point about difficulties in balancing AI priorities.

 

[Bibor]

I take back what I said about AI not using battering rams. There was one in my latest playthrough. Against a city-state.

So yes, apparently clustering units is not the issue (I was wrong on this one), it's the human petering out the cluster too fast (and the AI not anticipating/overbuilding for it).

 

[RohirrimElf]

The AI is getting better with each update, small steps at a time time. But what civ 6 needs is 2 solid years of modding by the community.
Every time i wait for an update i hope this would be the time where i would play civ nonstop. But this moment has not happened yet. Only like 700 hours on my steam page. When i start a new save after a long break i am hopefull if the AI challenges me. But half of the time it just does something irratic no human would ever do. chokepoints are just beyond firaxis developers to solve.

Have not given up on civ 6 do. We have reached a point where i do enjoy the game for the time i play. There is much to like about civ 6 i dont find on europa universalis 4

 

[oSiyeza]

I'm all for better AI and dropping difficulty to Prince. Apparently everyone can play Deity these days, as if its a challenge. The consequence of Deity is that you end up with more cities sooner than you usually would.
At least Civ4 had the common decency to completely crash your economy whenever you were rushing a deity AI.

Personally i dont find too much fun playing in deity. As I have to optimice every decission instead of enjoying different options.

I have a friend all about deity, and his style is that he follows a very specific build order and playstyle in order to maximize stats, and rerrolls for a more or less speciffic map to succeed. If he does not achieve certain conditions in his first 50 turns he just restarts. He also claims the game is too easy, but i think cracking the game AI and systems to such a degree, when he always uses the same rigid optimal roadmap is a kind of a trap that gets in the way of fun.
Im not saying is not a valid way to play, but expecting challenge with this approach strikes me as a bit odd.

 

[cvb]

Have not given up on civ 6 do.

I had my best civ4 (modding) time 2010-2012. civ5 is in its zenith now. civ6's bloom is still way ahead ...

 

[TheMeInTeam]

Once again, the fallacy in disguise (or am I wrong?). The same can be said of civ 5, yet two fans, in their spare time, destroyed the fallacy. Why would I assume the same cannot be done with civ 6, and thus, void the fallacy fallacious...?

I stand my ground; 1 UPT, nor other mechanics, are to blame here for mediocre AI.

Lots of things contribute to poor AI performance. Some of it is raw lack of skill/investment. Some of it is, in fact, the mechanics. A large part of it is that the devs don't care and/or make it not try intentionally (don't know how much of Soren's philosophy is with Ed Beach, probably a lot since the AI throws at strategic levels hard, not just tactical levels).

I have a friend all about deity, and his style is that he follows a very specific build order and playstyle in order to maximize stats, and rerrolls for a more or less speciffic map to succeed. If he does not achieve certain conditions in his first 50 turns he just restarts. He also claims the game is too easy, but i think cracking the game AI and systems to such a degree, when he always uses the same rigid optimal roadmap is a kind of a trap that gets in the way of fun.

It is possible to win consistently on deity w/o any cookie cutter openings or rolling for good starts.

 

[seanos08]

The AI can be programmed to be a bit better at war. Like not attacking a city with 3 catapults and no melee or cavalry. Having doctrine such as garrisoning a unit in every city and building walls asap. Also checking military levels of human and other AI and building military units when less than 80% of their strongest neighbor.

I had a AI Roman city flip to me. The Romans were allies and decided to betray me. They had 2 tank armies, a cavalry army and an artillery army with a balloon. There was a Roman encampment 3 tiles from their lost city. I had a machine gun in the city and a modern armor army next to it.

All the Romans had to do was put the artillery in the encampment, bombard the MA on the first turn and then kill or redline it with 2 tank armies. Then they could bombard the city from their encampment with their artillery immune from attack and take the city.

Instead they moved their artillery to within 2 tiles of the city and raced their 3 armies deep into French territory. I was able to bombard their artillery then destroy it with the MA. Their cavalry and 1 tank army got destroyed and the other redlined so it made a hasty retreat to Roman territory.

 

[tab]

The AI can be programmed to be a bit better at war. Like not attacking a city with 3 catapults and no melee or cavalry. Having doctrine such as garrisoning a unit in every city and building walls asap. Also checking military levels of human and other AI and building military units when less than 80% of their strongest neighbor.

I mean how hard is that?

Instead they moved their artillery to within 2 tiles of the city and raced their 3 armies deep into French territory. I was able to bombard their artillery then destroy it with the MA. Their cavalry and 1 tank army got destroyed and the other redlined so it made a hasty retreat to Roman territory.

This to all the people saying that the AI is improving incrementally with each update. I mean how do you keep on playing when the the AI is hell bent on killing itself.

Edit: Im gonna resume my playthrough just to get pics and gifs myself as a picture is a thousand words.

 

[glider1]

Lets hope the next patch works on the tactical AI. War is declared. I had three AI field cannons enter into my territory with no support. First cannon breaks off and decides to catch the train using my rail and finds itself deep in my territory picked off in a single turn by city and encampment fire. Next turn the other two cannons do nothing not even pillage and get picked off similarly. This is after nearly three years of development on Civ 6.

 

[tab]

First cannon breaks off and decides to catch the train using my rail

Haha. The only incremental improvement i see here is that instead of sending every type of unit into the shredder, it is now making more and more of the seige unit and sending em to the shredder. "Improvement".

 

[darko82]

Lets hope the next patch works on the tactical AI.

Dream on!

 

[oSiyeza]

I mean how hard is that?

This to all the people saying that the AI is improving incrementally with each update. I mean how do you keep on playing when the the AI is hell bent on killing itself.

Edit: Im gonna resume my playthrough just to get pics and gifs myself as a picture is a thousand words.

Haha. The only incremental improvement i see here is that instead of sending every type of unit into the shredder, it is now making more and more of the seige unit and sending em to the shredder. "Improvement".

Dream on!

Lets not pointlessly complain about how bad the game it is. That is not constructive at all.

Also to use the "I saw the AI do a stupid thing ergo the AI has not improved and it will never be and I hate it" attitude only harms the community. Lets not use biased observations to support a preconceived conclusion.

The AI being currently lacking does not support at all your claim that has not improved at all or that FXS does not care about the AI. A lot of complains from the initial state of the AI have been addressed.

The game has received 13 patches and 2 expansions and all of them or almost all of them had a section for AI changes. Read the forum and read the patch notes. Neglecting the work that has been done to improve the game will not help in any way to improve the game in the future. That will only contribute to see the players as irrational children that cannot be satisfied and disregard any attempt to address our complains.

You noticed an odd behavior of the AI, cool.

When did it happen? is this a consistent or frequent bugged behavior? can be replicated? do you have a saved game to submit to the bug reports?

 

[oSiyeza]

I mean how hard is that?

It is in fact extremely hard. I recommend to take a look on the discussions of the real strategy mod for those of you that are interested in how AI Civ works (BTW it is an awesome mod that improves the AI in a significant way).

The point is, AI behaviors are not algorithmically implemented. It is a tree with layers of strategies and sub-strategies. Bugs are hard to predict and odd behaviors will happen even with a perfect code. And is not a bad thing, cause those odd behaviors are also a side effect of flexibility. However, there is a lot of room for improvement and more immersive and challenging strategies can be implemented while a lot of the odd behaviors the AI does can be identified and corrected.

Also a lot of odd behaviors from human perspective, have sense from an AI perspective. Trowing catapults to an attack is a behavior that has been talked about in some other post for example. It may happen when AI suffered an attack and most standard AI army units have been killed. Since catapults are not usually used in defensive tasks, the AI may end having a bunch of catapults and no regular units. In an odd way the AI may prefer to launch an attack with those units instead of disband most of them and build replacement army units. But that behavior does not mean the AI is dumb.

We should instead point to those specific instances in the bug report section. And help the developers to iron those behaviors.

 

[oSiyeza]

Ho, globally, we agree , yes.

but when we go into specifics:

"An improved AI with more reliable path-finding, better management of units and specially groups of units" :

IMO just change the 1 combat UPT rule to n combat unit per tile where n is a finite number, n > 1 and n < stack of doom. Then you can improve your tactical AI on those rules (groups compositions per tile), but it will perform better already (pathfinding and formations partially solved) and you can spend more development time on other areas.

But a lot of people don't wan't to change the tactical aspect of 1UPT combined to 2 tiles ranged units, and while, yes, you can make a better AI for those rules, you may have to spend a lot of development time on it (surely less than what the VP's team had spend on it, but still some time)

And I'm pretty sure that if we can make a good tactical AI for 1UPT in x hours, then we can make a better one for nUPT in less hours.

I think you are wrong here. The entire game as it is now has been designed with 1UPT in mind. Yes there is some nuances that are implemented here and can be expanded upon. But if the key point is resources and development time, you are mistaken.

Consider that to use your approach you should first implement a new UPT system and then rebuild the AI including many systems that already work fine in order to just start to implement war strategies. Also this affect the balance of the entire war system, from production costs to unit strengths.

The alternative of just improving and iron the current system is no doubt much less time consuming and much less likely to need debugging, testing, QA and patching. Those will require alone more resources than the necessary to just improve and deepen the AI.

If you are aware of how software development works, you can easily see that changing in this stage of the game a core mechanic would lend to months of developing and insane costs, in addition to the actual AI development.

This is just not a viable solution and even more probably not a solution most players want since it removes the fundamental strategic elements of combat. Yes making it more simple for the AI to handle but also more boring for the player to use.

"better in managing and obtaining resources" :

IMO, again, more a problem of design than pure AI, strategic resources are highly RNG dependent, without a lot of alternate ways to get them (while the AI may require the strategic before being able to conquest it for example)

So first add other possibilities for acquisition than trade or conquest for a player that has none in his territory: small income in strategic from each mines even without deposits, a new smuggler unit similar in usage than the trader, etc...

This is a key point. I don't think the game has to be substantially changed here. Just tweaking trade, resource distribution and resource requirements will help a lot no doubt. But if resources are too easy to get, the core game mechanic would be lost and I think it is one of the best ideas in the game and should not be simplified too much. In my opinion the path to follow is to implement AI strategies to judge the resources it needs and effective ways to pursue them.

"more focused in the victory conditions that suit the game state and the leader agendas"

I disagree on that one and I'd change it to "build an Empire to stand the test of time" or I would accept it as a priority if we first change the rules to create different "games states" than the current ones. OFC, it's just an opinion, part of the RP crowd, I don't mean it should be that way, I'm just pointing again that, yes, opinions may differ on what/how.

Maybe i expressed myself badly. When i said "game state" i was merely expressing the ability of the AI to judge the situation of the game. IE, how is my map, how are other civs doing and what opportunities do i have according to my current situation. The AI already does this. And this is pretty much the approach taken by the Real Strategy mod. In that mod the AI commits to one or other strategy depending on civ and leader preference, but also on the current situation of the civ and its known rivals in that victory path. Also commits longer to a strategy instead of flipping constantly and takes into account only the available victory conditions of the game. This is building on the AI systems that the game already has. This mod has shown how good this approach is in making a AI more competitive and FXS should take note.

As an even more cool side note. Recently somebody talked in the Real Strategy mod page about the possibility of introducing counter strategies that the AI would use in order to mess with its rivals (I'm unsure about if that was implemented). But how cool would be to expand on that idea?