[GS] Yet (Not) Another AI Thread

Pure24

Warlord
Joined
Jun 24, 2019
Messages
191
I know, I know... it has been talked about to death. But this time, I'd like to approach it from another angle. It is not a 'defence' of the AI per se, just a thought I had and would like you guys' opinion.

Two things. First, the issue of Deity AI being incompetent/unable to catch up to the player. Is this really fair considering that human players are constantly utilizing cheats and exploits? One would pick mods which suit them (not the AI), and even then... there is the issue of always chopping wonders and so on. That's not a cheat, and arguably not even an exploit. But still, the AI cannot compete because it wasn't designed for maximal efficiency in that way. If one were to try and maximise their efficiency WHILE prioritising roleplaying, don't you think they can find deity difficulty as challenging as intended. Imagine trying to compete for wonders without using 'cheap' tactics (sorry for calling them cheap, just can't figure out another word at the moment).

Do you think AI should be coded to do things like that, things like upgrading Moksha and Reyna and use them to insta-buy spacesports and go hard for a SV?

Second, human players usually reroll until they get a start they're happy with. And even when the game starts, they'll usually reload saves to get that wonder they missed by one turn or to save that unit who was killed. Again, not the AI at fault...
Yes, the AI has efficiency issues and doesn't seem to plan ahead right from the beginning - also they're not so good at combat. Not saying the AI doesn't need fixing. But... I just feel like the human player sometimes does things to maximise their advantage, and then feels let down that the AI appears crippled.
:thanx:Don't scream at me!
 
my personal grudge against the AI is the lack of use of naval and air units , and the absolute lack of any kind of strategy in general. the AI will start to move ( not attack just move ) to the other side of the continent with all its units only to scrap that idea and return a few turns later. also worth to mention
"pikes in front , archers behind" . use light cavalary to pillage if you will not get a kill , protect siege and not send them unescorted in huge numbers to attack. dont attack cities with archers only , no melee to capture. retreat if you cannot capture.
 
Is this really fair considering that human players are constantly utilizing cheats and exploits?

I don't play deity and maybe I'm the exception but I stay clear of using cheats and exploits. I'm rather surprised you just state it as a fact.
 
my personal grudge against the AI is the...
Exactly. Combat AI could use improvement. I acknowledge that

I don't play deity and maybe I'm the exception but I stay clear of using cheats and exploits. I'm rather surprised you just state it as a fact.
Most Deity players do. And anyways, I didn't state it as fact; my last sentence begins, and I quote, "I just feel like the human player sometimes does things to maximise their advantage...".
If you're not one of those, that's fine too
 
Most Deity players do.
Why do you think this is the case? If I'm guessing: you have watched many of the YouTube players that play CIV VI? Some of them are really in it for a fun to watch roll-from-the-start deity game, so they don't mind rerolling and restarting games. You would be right to say that a player who does that frequently shouldn't complain the AI is rather poor. I think you're mistaken though that that is something most deity players do, if I watch games e.g. on the GOTM and HOF here on Civ Fanatics there are just many good players out there who really know how to play this game well.

I would be surprised if there were many people playing with actual cheats or mods that basically make you superior to the AI. What would be the fun in that? If you want no challenge, just play a lower difficulty. For most exploits it's clear which ones are basically cheating: the pantheon one comes to mind ;-), but also e.g. gifting cities to win religious victory. There are also many parts where the AI isn't programmed very well, and some people might think something to be an exploit (e.g. trading diplo favor or works of art) and others may see it as part of the game. I think most people agree that the AI sucks at warfare, is it an exploit then to go to war? I generally go by the rule 'if it feels like cheating, don't do it'.

The discussion you started here is basically the question "is the AI really that poor, or do deity players just exploit the game too much?". My answer is the former: the AI in this game should really be cranked up a little (or a lot) for there are many many things the AI does and doesn't that are really not that hard to improve if the developers really wanted. Imho of course :).
 
Hmm... we have to wait and wait forever. Will this ever happen?

When I am dead?
 
Last edited:
as soon as they release the dll code i am sure the modding community will fix the AI that the AAA company firaxis could not bother with , so at most 1 more year.
 
If one were to try and maximise their efficiency WHILE prioritising roleplaying, don't you think they can find deity difficulty as challenging as intended.

Im one of those people allergic to chopping or removing resources in this game (let alone exploiting), and tries to stick to the theme of each Civ, roleplaying as you have said. And I completely agree.

Also, I think that a scientific victory, after stealing lots of AI cities and their districts and boosts, is no SV at all, so probably unpopular opinion :D

Playing on theme and no chops is "easy" with just a couple civs. Definitely not Canada. Damn I hit my head against the wall so many times last game with him. First time playing as him ok, but 520 turn diplo victory (epic speed). Just after a 295 turn cultural victory. With the Mongols :lol:. I started thematic with Gengis, but after taking my continent I didnt want to send all the horsies to finish the game in 20 turns. In any case, even they are much better at diplo than Canada! just because of their start bias, they always have useful resources closeby to megaboost the initial expansion phase.

The only thematic domination game Ive done is Alexander. I finished in a bit more than 300 turns as well, but I kept my armies on strike for about 75 turns because I wanted to breathe a bit from all that troop managing. This game was a breeze for example.
 
Rerolling a start for a strong start, and reloading an earlier saved game are cheating (the AI doesn't get to do either) technically (not that it matters in a single player game).

I'm sure there are deity players that do this and those that don't - though I'd almost suspect it's actually more lower-level 'builder' players (I really want that wonder!) that reload more.

It might be interesting to add an 'iron man' mode - like paradox games have - where you can't load an earlier saved game.
 
I know, I know... it has been talked about to death. But this time, I'd like to approach it from another angle. It is not a 'defence' of the AI per se, just a thought I had and would like you guys' opinion.

Two things. First, the issue of Deity AI being incompetent/unable to catch up to the player. Is this really fair considering that human players are constantly utilizing cheats and exploits? One would pick mods which suit them (not the AI), and even then... there is the issue of always chopping wonders and so on. That's not a cheat, and arguably not even an exploit. But still, the AI cannot compete because it wasn't designed for maximal efficiency in that way. If one were to try and maximise their efficiency WHILE prioritising roleplaying, don't you think they can find deity difficulty as challenging as intended. Imagine trying to compete for wonders without using 'cheap' tactics

I understand your argument, but no.

It worked partially in V, but not anymore in VI. I say this because I ALWAYS roleplay. I never decide before a game what kind of victory I'm pursuing, and I'll often get to the Industrial Era still with no idea what kind of victory I'm going for. I play fairly leisurely and I'll even play with handicaps, like not allowing myself the Petra or Halicarnassus (after the update especially), for instance. I tend to save over the same file and only reload in case of misclicks, or when the game bugs out.

Even with all these, Deity still doesn't feel that difficult. I do occasionally lose, but generally, it just makes for a slower victory.

Edit: I even made a point of completely avoiding those charts with optimal district placement like they were Game of Thrones spoilers.

In Civ V playing Deity without being optimal made for a much greater challenge than in Civ VI. I assume it's the addition of Districts which has caused this difference, and perhaps Governors? The AI seems unable to compete with Human micro-management.
 
Firaxis is probably still operating under the assumption that civ5 vanilla’s “play to win” AI was unpopular, and still would be if they brought back elements of it.

They wouldn’t code separate AI for each difficulty- a QA nightmare- and making the AI play to win means negating a lot of game systems. People complained about this nonstop in civ5.

Diplomacy exists for the pleasure of the player. The interactions with the AI are part of the game. This isn’t chess where we apply an AI to be our opponent; having AIs in civ to interact with is central to the design. And 98% of the players wouldn’t be thrilled if they turned back on “terminator mode.”

That said, while rerolling starts is perhaps the most common thing people do, we would hope that out of the other 5-7 AI players, some of them will get good starts too. I don’t think the AI is that bad at managing its economy. If you play a lot of games, even on lower difficulty like King, you’ll still see peaceful AIs snowballing. It’s the usual suspects in culture/science trees, but it just shows that they are capable of performing without the ridiculous bonuses. They overvalue certain things- instead of using policies like rationalism they slot every amenity and housing card they can- and just changing that would make them much more competent (think about when the powerful policies unlock and when the AI starts falling off hard.)

But ultimately, while I’m fairly convinced a modding team with the DLL could make a Deity AI that was brutal, I think even the deity crowd here only truly wants to have the “tension phase” of trying to overtake them drawn out further, rather than “the AI wins Science on turn 200-230 so beat that or lose.”
 
I don’t think the AI is that bad at managing its economy. If you play a lot of games, even on lower difficulty like King, you’ll still see peaceful AIs snowballing. It’s the usual suspects in culture/science trees, but it just shows that they are capable of performing without the ridiculous bonuses. They overvalue certain things- instead of using policies like rationalism they slot every amenity and housing card they can- and just changing that would make them much more competent (think about when the powerful policies unlock and when the AI starts falling off hard.)

True. I play on King for Civs Im not too familiar with, and usually its a quite chill affair. But sometimes stars align for a civ or 2, they get incredible start spots, useful stuff on the goody huts and grab the CS first. And suddenly its turn 100, there are no religions left, and theres a Civ doing 70-80 science/culture while the rest are still doing half, at best. Oddly enough, I find the domination focused civs first on those more often than other more peaceful AIs. Especially Lautaro. Once wars start boiling Gengis usually takes the biggest piece of the cake tho.
 
A few things from the thread so far:

- I'd LOVE an Ironman option! Since its optional, people can switch it off
- I don't know why the AI is the same for all difficulties. Then it really isn't 'difficulty' so much as it is 'handicap'
- I don't think AI (Civ VI AI, to be specific) will ever be as good as people expect because, even accounting for exploits, human players are just able to micromanage on another level.

It feels good when you time the stars to align such that you'll finish a Tech to boost that Civic which will finish and unlock the extra builder charge card when the five cities building builders are just about done.
No ways the AI is ever competing with that AND managing to juggle all the other quirks of the game.

Loving the feedback, even the critiques
 
- I don't know why the AI is the same for all difficulties. Then it really isn't 'difficulty' so much as it is 'handicap'
It's always been like this. It's just a lot easier to develop- you'd have to change many versions, debug them all, you'd need QA testers who could actually handle deity (most of the devs play prince-king) etc. From an engineering level, making one AI that does what it needs to well enough and then adding some numerical bonuses to make it more effective is just so much easier. They'd need to retain all these extra resources all the time, which is the years at a time between full game releases. It's a lot of cost for little benefit - because so few people play on high difficulty.

Currently difficulty only affects some stat bonuses they get and starting units+eurekas. I'm sure with some LUA script or even just more advanced modding we could make them do different stuff based on difficulty. (For example, just make them absolute campus fanatics on deity.)

I don't think AI (Civ VI AI, to be specific) will ever be as good as people expect because, even accounting for exploits, human players are just able to micromanage on another level.
We can take unlimited turn time with our fleshy processors, and all the AIs together can only take a reasonable amount of time that players will put up with. Maybe if they based the AI around "civ6-by-mail." I don;t even think the current AI is a million miles off, but there are ways to make the AI perform better without even touching the Ai code - the more balanced the game is, the better they will do. Humans can detect and leverage unbalanced aspects of the game to get an advantage that computers cannot. But Firaxis as a company would need to take on like, a monthly patch cycle and watch groups of multiplayer games to really get a good sense of balance. It's just not built for that kind of software product (not a bad thing, they focus on other things well!)
 
It's always been like this. It's just a lot easier to develop- you'd have to change many versions, debug them all, you'd need QA testers who could actually handle deity (most of the devs play prince-king) etc. From an engineering level, making one AI that does what it needs to well enough and then adding some numerical bonuses to make it more effective is just so much easier. They'd need to retain all these extra resources all the time, which is the years at a time between full game releases. It's a lot of cost for little benefit - because so few people play on high difficulty.

Currently difficulty only affects some stat bonuses they get and starting units+eurekas. I'm sure with some LUA script or even just more advanced modding we could make them do different stuff based on difficulty. (For example, just make them absolute campus fanatics on deity.)


We can take unlimited turn time with our fleshy processors, and all the AIs together can only take a reasonable amount of time that players will put up with. Maybe if they based the AI around "civ6-by-mail." I don;t even think the current AI is a million miles off, but there are ways to make the AI perform better without even touching the Ai code - the more balanced the game is, the better they will do. Humans can detect and leverage unbalanced aspects of the game to get an advantage that computers cannot. But Firaxis as a company would need to take on like, a monthly patch cycle and watch groups of multiplayer games to really get a good sense of balance. It's just not built for that kind of software product (not a bad thing, they focus on other things well!)
There are plenty of things they (devs) can (could) do to make AI perform better, though they should have put more effort there before release. I assume they (devs) atleast have a clear idea of how they'd like the game to be played. Humans do not only detect and leverage unbalanced aspects of the game, they (too) often call for them. If "watch groups of multiplayer games to really get a good sence of balance" is the future for 4X games, then I'm off sooner than later.

About Diety AI, I think it's alright if they would cheat - they are gods, not angels.
Remove Fog of War and other things that make their decision basis uncertain - Information is power.
 
Good thread. :)

I agree that humans have some natural advantages, such as being able to reload, reroll, and to varying degrees exploit the metagame. I don't think it would be much fun if the AI played for maximum efficiency with regards to the meta (and certainly not exploits). Having said that, I think there are some things which could be reasonably done to provide a more enjoyable challenge:
  • Provide an ironman mode. Also, do what Stellaris did, and disable achievements if you are not playing on ironman. That will effectively turn ironman into the real game, while non-ironman games can be considered practice mode.
  • Improve the tactical AI, at least a bit. I don't expect it to be fantastic, but it should be possible to make something which is at least semi-competent in the field. At the moment, going to war against the AI almost feels like an exploit.
  • Fix game mechanisms which create counterintuitive meta-play. For example, make chopping less powerful, and change the way tech/civic boosts work. Instead of pouring a lump sum of science/culture into the tech/civic, a boost should give a percentage modifier to science or culture while it is being researched. The current system creates a meta where you should almost always swap out what you are researching and wait for the boost if you don't have it already.
  • Provide greater asymmetric/environmental/internal challenges. Civ is a symmetrical game at its heart. A number of somewhat balanced civs start out at mostly equal footing, and generally compete for the same goals throughout the game. To create difficulty by throwing massive bonuses at AI players is going to look and feel like the AI is "cheating". However, if you could create other types of challenges, from the map itself, external events, or internal issues, you can increase difficulty, or at least provide the player with more goals to strive for, without it feeling cheap. This is maybe easier to do in sci-fi or fantasy settings, but I believe it should also be possible to do in regular civ.
 
If "watch groups of multiplayer games to really get a good sence of balance" is the future for 4X games, then I'm off sooner than later.
What I meant was, if a developer wants to know what is over/under tuned in their game, the most reliable way to do it is to see what players are doing in MP games. If horses are overpowered everyone will make them. If chopping is overpowered everyone and their mother will chop. Etc. It's about having a gauge to see what's going on, since humans are ruthless and will always find and exploit them.

Now, what they want to do with that knowledge is up to the devs. But even keeping your hand on the pulse of game balance isn't something most studios are set up to do, and civ especially doesn't have the kind of competitive scene to make that feasible.

My post was only truly aimed at the concept that develops can balance the game so that there are very few exploits for humans to leverage against the AI, thus making "how well the AI plays" the prime metric of performance vs "how well it exploits various things."
 
What I meant was, if a developer wants to know what is over/under tuned in their game, the most reliable way to do it is to see what players are doing in MP games. If horses are overpowered everyone will make them. If chopping is overpowered everyone and their mother will chop. Etc. It's about having a gauge to see what's going on, since humans are ruthless and will always find and exploit them.

Now, what they want to do with that knowledge is up to the devs. But even keeping your hand on the pulse of game balance isn't something most studios are set up to do, and civ especially doesn't have the kind of competitive scene to make that feasible.

My post was only truly aimed at the concept that develops can balance the game so that there are very few exploits for humans to leverage against the AI, thus making "how well the AI plays" the prime metric of performance vs "how well it exploits various things."
That's the great art of making the (next) flavour of the month.
I withstand, If "watch groups of multiplayer games to really get a good sence of balance" is the future for 4X games, then I'm off sooner than later.
 
They should consider removing the fog of war for the AI on higher difficulties. Or have the option to let the player decide its activity during game setup.

I have no problems with how AI manage their tiles/districts etc... but their incompetency in tegards to war tactics is just really annoying considering they keep declaring war every 5 minutes for stupid agenda thingies and/or declaring based on military points. It gets old quickly.
Although I have been defeated on turn 20, but thats because I had like 1 warrior.
The AI in Civ V played like Napoleon compared to this. They also hzd some kind of strategy (without mods); they would for example invade my territory from two sides at the same time.

Maybe in civ 6 its because of the way the game generates maps? Too many choke points maybe? No idea

Also no, i rarely reroll a map/start and play basically on ironman, unless I misclick or something
 
Last edited:
Maybe in civ 6 its because of the way the game generates maps? Too many choke points maybe? No idea
The change to the movement rules from civ5 I think makes the 1UPT system really unwieldy for the AI since they are extremely constrained. I understand the idea behind the change but tactically it makes things really easy for humans because there is a lot more "maneuver" through rough terrain- more ways for the AI army to get stuck, more safe zones for humans to hide in, etc. A swarm of AI knights on an open plain vs human crossbows is going to be much different than in the hills.
 
Top Bottom