Status of Civ 6 AI

[mckinney156]

If you want to play against human like AI, why not just play a MP game? I don't think its within any companies budget to code a separate AI per difficulty level.

 

[KayAU]

I don't need the AI to be human-like, I just want it to be competent. I believe most people play civ single player only. For me, it is not only that I am not looking for a multiplayer type experience, I also do not have the possibility to schedule time for long multiplayer games. I prefer to play on epic game speed, and if I am to play a game like that with people online, we would all need to set aside a few hours of continuous play time each day. That might have been fine back when I was a student, but now that I have a job, a wife and a kid, I can't really play anything which can not be put on pause at any time. Turn based games against AI are ideal.

Also, I don't think they would need to code completely separate AI. Actually, I know they would not have to do this. They would code an AI which had several configurable parameters to decide how strong it is, like for example how deeply it would analyze the situation before making a move.

As for what is within the company's financial abilities, I don't think anyone here knows this very well. Civilization games sell a lot of copies though. Didn't Civ V alone move 7 million copies? I think they can afford to invest a bit in improving the most heavily criticized aspect of their games.

 

[Rupe]

Just played the beginnings of a game after leaving a several months ago.

Met Japan, had a disagreement, they declare war on me. I kill off their initial army, they proceed to build 3 settlers and 2 builders.All of witch are taken by barbs. They only got that many build because I got distracted with fighting off two barb camps that were harassing me.

The fact that they had no army and prioritized building civilian units when at war with me and my decent sized army was a game breaker. Things like that just should not happen. I'll shelf it again any try back after Christmas sometime. Something tells me it will never get fixed though. Finally hit the point where I have lost hope. Never thought it would happen with a Civ game.

 

[dexters]

Totally not buying this.

Sorry, but I think its exactly that the business model uses players as playtesters. Some of these issues could be fixed in ten minutes. Half the players here could fix them in ten minutes if they had access.

I think their in-house play-testing is pretty much auto games and cheating down situations to see if something specific is working as intended. Which means they are going to miss all kinds of things.

You're entitled to your own opinion of course, but raging about it now is kind of pointless. We're all just speculating.

Some of the bugs are easily repeatable and pretty obvious. And having been involved in beta process before, I know for a fact devs do ship with known bugs because they don't have time to get to them and the ship date is set.
This is why I spend most of my post postulating as to why they had to have a summer patch if a larger fall patch would have been ok for a lot of people. The Nubia DLC may be a good reason why

 

[redrum68]

The AI is still pretty poor. I think if it only built military units, it would probably be better than the current AI. Even on Deity, after the first 30-40 turns its very easy to have a larger military than the AI. This is even ignoring the fact that from a tactical standpoint the AI probably needs 3-4x the number of units as the human to have a chance currently.

 

[Archon_Wing]

If you want to play against human like AI, why not just play a MP game? I don't think its within any companies budget to code a separate AI per difficulty level.

Starcraft II and AOE2 HD have pretty decent bots that play smarter at higher difficulty levels. The former only cheats on the highest setting and the later doesn't cheat unless the user enters a code. These things have to control dozens if not hundreds of units in real time. While they're not enough to beat good players and are highly exploitable, both games are multiplayer centered so AI isn't even a priority.

And then we have modders in this franchise that have always done a better job than the default.

 

[illusiondrmr]

Old civ fanatic here (since civ1). I played on the release of civ6, clicked DIETY, won 2 out of 3 games and never touched it again.

Just clicked on civfanatics, only to find a thread like this, to check the current status of the AI.

Cya in a couple of months again, i will search a thread like this and see whats going on. Let's hope for a patch that delivers a competent AI in the future. Have fun my civ brothers ! (CIV4 <3).

 

[gettingfat]

The AI is still pretty poor. I think if it only built military units, it would probably be better than the current AI. Even on Deity, after the first 30-40 turns its very easy to have a larger military than the AI. This is even ignoring the fact that from a tactical standpoint the AI probably needs 3-4x the number of units as the human to have a chance currently.

That's why at emperor and higher, imho the biggest trouble is almost always the barbs. Why? Because they know only one thing --- keep spamming units and less likely retreat.

District is a great idea on paper, but like 1upt it is too complicated for the AI to properly execute. They rarely build commercial and industrial zones, so they have insufficient fund to support a large army and upgrade it, as well as insufficient production to spam units. That's why the AI civs in civ5 generally have better military. At least they carry more units because the game is simply more straightforward.

 

[sjudubbel]

For me the AI is too weak as well.
It it way too easy to win combat situations against the AI. It doesn't know how to operate a full scale attack.
But I do realize this can be quite hard to fix.

A much easier fix for Firaxis would be to make the AI focus more on its military production.
The AI really doesn't produce enough military units!
Perhaps a part of the problem is its extreme focus on religious units!
When I played deity the AI civs had literally 15-20 missionaries or whatever swarming all over my land.
This is a huge waste of production power!

 

[Victoria]

Looking through the logs, the AI goes through phases of play, not many but it could be tweaked better based on this.

I 100% agree that religion slows the AI a large amount. The race for 5 is just crazy.
The AI is now building more mines and has been for a patch now, this has increased their production for units but their gold strategy is still very poor..
The early pushing of science is also a bad move, all AI's should be doing their first expansion a lot better than it seems to.
I still say that the fact that the AI can move and shoot in the same turn is a huge improvement over civ 5, it was one thing I always could take advantage of.

 

[earlc]

That's interesting. And I was all for having a religion mechanic in the base game, but if it's that taxing then maybe it would be better off greatly simplified.

Starcraft II and AOE2 HD have pretty decent bots that play smarter at higher difficulty levels. The former only cheats on the highest setting and the later doesn't cheat unless the user enters a code. These things have to control dozens if not hundreds of units in real time. While they're not enough to beat good players and are highly exploitable, both games are multiplayer centered so AI isn't even a priority.

And then we have modders in this franchise that have always done a better job than the default.

Those games have far, far simpler movement, though, without turns or tiles to worry about, let alone districts.

I believe it was stated that the main improvements to the AI by modders were more about adjusting and streamlining production, which would allow the AI to have a better military.

Actual AI combat tactics probably have relatively little to do with the perceived difficulty. Like in that last game with Norway's aggression, Harry kept spamming knights at me after the initial attack, which appeared to reach me in groups of about 3, sometimes with a crossbow or catapult mixed in. If he had actually guarded the units and supplemented his attack with a navy, it would have been devastating.

 

[agonistes]

You're entitled to your own opinion of course, but raging about it now is kind of pointless. We're all just speculating.

Some of the bugs are easily repeatable and pretty obvious. And having been involved in beta process before, I know for a fact devs do ship with known bugs because they don't have time to get to them and the ship date is set.
This is why I spend most of my post postulating as to why they had to have a summer patch if a larger fall patch would have been ok for a lot of people. The Nubia DLC may be a good reason why

Raging?

 

[ashendashin]

Actual AI combat tactics probably have relatively little to do with the perceived difficulty. Like in that last game with Norway's aggression, Harry kept spamming knights at me after the initial attack, which appeared to reach me in groups of about 3, sometimes with a crossbow or catapult mixed in. If he had actually guarded the units and supplemented his attack with a navy, it would have been devastating.

No, this isn't right. With the tile mechanics it's very important to teach the AI to use it correctly. Otherwise it gives the player plenty of terrain and range related advantages. Advanced strategies to properly use flanking and getting past choke points is something that even Vox Populi has trouble with, and the tactical AI of any vanilla civ game is a joke compared VP.

 

[PhilBowles]

I mean if we're talking voice actors and personalities, that's a separate subject from what I was addressing, and I don't really see any issue with your comments there.

Not really. If your leader look and gestures are of a comedy caricature you're going to hear the dialogue and tone through that lens. Alexander in Civ V came across as a commanding warlord - Alexander in Civ VI is Prince Charming from the Shrek films, right down to the comic preening.

But I completely disagree regarding the leader graphics. I don't know anything about clash of clans, but the setup of leader interactions in VI is far more personal and engaging, making rival civs represented more by characters than color schemes and playing off the way our brains process information.

All I know about Clash of Clans is the graphical style, hence the comparison. Human brains are very well-designed to recognise human-looking figures and to isolate them from background details in much busier settings than the Civ V backdrops - I don't see what you're getting at with that assertion. We're also hard-wired to read body language and to distinguish appropriate from exaggerated gestures, as well as to read the latter in a negative light.

A degree of randomness isn't a bad thing. If the AI for certain civs acts exactly the same every single time, I'd count that kind of predictability as a drawback.

There always has been a degree of randomness - the way character coding has worked in all previous iterations is that leaders were given randomised scores for each aspect of their behaviour, with each having weighting towards preferred options for that leader. So they'd behave in a certain way more often than not, but not exclusively so.

Civ VI gets the worst of both worlds. AIs appear to differ little or not at all in favoured behaviours, all civs behaving similarly in most cases and with nearly all of the remaining variation fully random. But imposed upon that they have the agenda system, which means that, without fail, any given civ will act in exactly the same way in one specific context in every single playthrough. This produces a result where the civs manage to both lack meaningful personalities yet are very strongly predictable, moreso than Suleiman's backstabbing or Gandhi's love of nukes in past games.

On reflection, this may be a hallmark of Ed Beach design, in which subtlety is not a virtue. In BNW AI behaviour was the area that actively regressed following the expansion - the shiny new toys were culture and ideology, so every civ cared massively (and equally) about ideology, and save for a couple of hard-coded exceptions every civ always favoured culture over science, in both cases to the near-exclusion of other personality traits once these features became strongly relevant.

In Civ VI the shiny new thing is agendas, so those have to be the primary determinant of AI behaviour and any other sources of variation in personality are minimised in order to showcase them.

There should be some dread in settling near a civ you know has certain military bonuses

The dread shouldn't be based on their mechanics alone - and fear inspired by mechanics shouldn't be mitigated because you know that all you need to do is not backstab people and Tomyris will be friendly unless she randomly decides you're her enemy. The Ottomans and Mongols could be frightening in Civ V simply because of their personalities, even though in the early game they got no military bonuses (as they're uniques were medieval).

Kinda feels like we're falling into assuming everything with V is necessarily better just because of some of the failures of VI, which I don't think is fair.

That seems a weird interpretation. I'm going by long experience with Civ V and earlier games and experience with Civ VI. Civ VI simply is a step down in most respects, more than one in some, and the complete lack of AI personality is so grating that that alone has stopped me completing sessions of a game whose numerous other major flaws I'm able to overlook to some extent. I don't need to make any assumptions when experience with every iteration of Civ since the series started tells me that, were it not for having a main-series number attached, Civ VI would be better-compared in terms of its overall quality with spinoffs like Beyond Earth than with the main Civ games. It's better than Beyond Earth, to be sure, but if it were called Civilization: Earth or somesuch instead of Civilization VI it wouldn't warrant comparison with Civ IV or Civ V.

I genuinely struggle to see a single element in Civ VI that's a clear advance over equivalent mechanics in Civ V, with the sole exception of Great People. Touches touted as new, like districts, are just a classic Ed Beach approach - take something the game already encouraged and oversell it without really doing anything very new with it.

 

[PhilBowles]

How great would it be if at Deity, the AI got no bonuses, but played really well instead? This is not what I am expecting, though. At this point, I'd settle for it having basic competence at playing the game.

How about getting it to play correctly as a baseline before asking for it to vary by difficulty level? That's something it's hard to impossible to code outside games with very simple rules or stereotyped strategies. Someone raised the example of Starcraft, which has illusory differences between AI difficulties. The AI doesn't play any better (for instance it gets no better or worse at microing units) or according to any better rules, it's just given higher-level build orders and timings at higher difficulty levels.

This is relatively easy since there's a wealth of data from multiplayer as to how novices themselves level up, as well as on appropriate build orders at every level of play, so once the AI's presented them with a low-level challenge, like beating a Zerg rush, it's given a more sophisticated one to play. Civ games have nothing very closely equivalent, and far more complex variables to track.

 

[Archon_Wing]

Build orders are really the same thing as playing better. You'll probably end up worse if you open 4 scouts every game. That is just like a chess ai that has more openings available to suit people as they move up. It is not about being "smarter" but playing better while still playing relatively the same game as the player. I don't have any expectation of tactical improvements and do believe that ultimately ai needs bonuses-- but at least keep up a show. Clearly it doesn't work g ere.

Also, I am skeptical that these things are that mysterious in Civ vi. Sure, people have variations, but there are general benchmarks that people have regarding number if cities and tech levels that don't fluctuate thst much. In civ 4, we were able to set target research and city levels by a certain turn, and here we have people throwing out average win dates. It is not too far feteched to find out when you average player can build 4 archers given a certain amount of cities.

Sure, it will not work in all situations. Can't say it is is any worse than now. But sure, at this rate it would be nice to have ai that works at all.

 

[PhilBowles]

Build orders are really the same thing as playing better.

It's part of it, and certainly something that needs addressing, but Civ is much more context-dependent in that regard than Starcraft. A Starcraft AI can literally be fed a series of build orders and timings, and as long as it has limited flexibility to improvise, it will perform differently depending on the strategy its been given even if its actual execution of play mechanics is poor. RTSes have the advantage in that regard that they can leverage something computers naturally do better than human players - multitask. Turns instead of real-time remove that advantage.

You'll probably end up worse if you open 4 scouts every game.

True, but say you give the computer an approximate build order 'build military units at point X' in a city that meets certain conditions, until they reach a cap at point Y. That's the way human players might approach the game to maximise their economic efficiency - and it will work well for the AI if the player doesn't play especially aggressively, or if the production city isn't too far from the player's to reinforce in good time. And that's discounting any interactions with other AIs. But if the AI loses those units it's following the same instructions and doesn't know how to adapt it will struggle - not an issue in Starcraft because any given building is only capable of producing military units, so any resources are automatically fed into units, and each structure has its own build slots. Your barracks doesn't need to take a break because you're teching up or building SCVs (resources permitting).

Also, I am skeptical that these things are that mysterious in Civ vi.

It's less that Civ VI is 'mysterious' and more than you're massively overestimating the complexity of games like chess and Starcraft. Chess takes advantage of something AIs can do well - it has a finite series of moves available at any point and the AI is very quick at running through the list and identifying which one leads to the greatest value. Starcraft has intrinsically strongly restrictive mechanics that limit what the AI is allowed to do - most building slots can only produce units, for instance, set supply caps, and set spots for base expansions - and small unit rosters. Units in Starcraft have set counters and almost nothing in the game varies with context. AIs perform noticeably worse in RTSes that have less restrictive options or more complex rules and interactions, such as Age of Empires or Cossacks.

Sure, people have variations, but there are general benchmarks that people have regarding number if cities and tech levels that don't fluctuate thst much. In civ 4, we were able to set target research and city levels by a certain turn, and here we have people throwing out average win dates. It is not too far feteched to find out when you average player can build 4 archers given a certain amount of cities.

Sure, it will not work in all situations. Can't say it is is any worse than now. But sure, at this rate it would be nice to have ai that works at all.

That needs to be the priority, indeed. One option that occurs to me is either remove Holy Sites as a district type and just allow the associated buildings to be placed in the city centre, or at least don't count them towards the district limit (since their utility as districts for GP generation fades very rapidly). At the very least that minimises the economic penalty the AI suffers for spamming them inappropriately.

 

[Archon_Wing]

It's part of it, and certainly something that needs addressing, but Civ is much more context-dependent in that regard than Starcraft. A Starcraft AI can literally be fed a series of build orders and timings, and as long as it has limited flexibility to improvise, it will perform differently depending on the strategy its been given even if its actual execution of play mechanics is poor. RTSes have the advantage in that regard that they can leverage something computers naturally do better than human players - multitask. Turns instead of real-time remove that advantage.

That is correct at least in theory that turn based does remove the computer's advantage to do inhuman moves,

True, but say you give the computer an approximate build order 'build military units at point X' in a city that meets certain conditions, until they reach a cap at point Y. That's the way human players might approach the game to maximise their economic efficiency - and it will work well for the AI if the player doesn't play especially aggressively, or if the production city is too far from the player's. And that's discounting any interactions with other AIs. But if the AI loses those units it's following the same instructions and doesn't know how to adapt it will struggle - not an issue in Starcraft because any given building is only capable of producing military units, so any resources are automatically fed into units, and each structure has its own build slots. Your barracks doesn't need to take a break because you're teching up or building SCVs (resources permitting).

I think that's not really that big of a deal for a few reasons.

Starcraft (both)'s AIs are mostly one size fits all and there are many different maps so it still affects it there. Even on the same map depending on spawns, you could be much closer than usual and distance most certainly does matter. And while the AI might be sub-optimal in some cases, I think that's a sacrfice I'd be willing to make.

As for losing units, I don't think that is a big deal since SC's AI performs a unit count for each step. So build 16 marines, wait 1 minute, and gather 16 marines and attack, the AI would automatically rebuild those marines before proceeding. Now obviously that lead to variety of problems like being stuck in a loop, it's not impossible to get rid of. When I modded it, I set a certain time limit to either force the attack to be aborted or continue anyways. Of course, that isn't analogous to Civ at all, but I'm speaking in a more abstract sense, like maybe if it's the medieval era and we haven't attacked with archers yet, we would do something else entirely.

Civ 4's AI was capable of plotting wars and dropping them if it could not manage the proper attack force.

Also, for Zerg, they do interfere with unit/worker building though I suppose they don't do as well with them-- to be fair!

It's less that Civ VI is 'mysterious' and more than you're massively overestimating the complexity of games like chess and Starcraft. Chess takes advantage of something AIs can do well - it has a finite series of moves available at any point and the AI is very quick at running through the list and identifying which one leads to the greatest value. Starcraft has intrinsically strongly restrictive mechanics that limit what the AI is allowed to do - most building slots can only produce units, for instance, set supply caps, and set spots for base expansions - and small unit rosters. Units in Starcraft have set counters and almost nothing in the game varies with context. AIs perform noticeably worse in RTSes that have less restrictive options or more complex rules and interactions, such as Age of Empires or Cossacks.

Funny, I thought the HD version of AOE 2 had an amazingly competent AI (relatively speaking) at least against casual players. And excluding micro, it still pops up proper counter units which is arguably the most basic of things. And every map is different. You can do cheap tricks against it, but I don't see the point of that.

I'd also disagree that Starcraft isn't context dependent at all though there are probably more things that are fixed against an AI. Oh, and honestly, how big is Civ 6's unit roster? It's by and large rock paper scissors too. You have your melee, your ranged, your horses and your anti-horses. And sometimes siege and ships. And then there's the more advanced versions of these things until you reach planes.

All, and all, I don't really expect any AI to grasp all the nuances and be active like a human player in adapting. What I do want to see is an AI that can play its own game. And when I see an AI that cannot:

* Build a big enough army, even with bonuses
* Upgrade said army
* Properly place garrison units
* Decide when to war, even when compared against 5.

Mostly doesn't have to do with tactical AI or about adaptability. It's more about winning a war against an AI and feeling that you beat it as opposed to it beating itself.

And I see modders that change this, well....

My original point was against the idea that companies don't find it affordable to design varying levels of AI skill, when they have. It doesn't matter how much easier it is. And sure, I know RTS isn't analogous to TBS. My point was never that. My point was that these games are multiplayer focused, and they still put effort into single player AI of which most of their audience wouldn't really care about! And here we have a single player focused game and, well...

That needs to be the priority, indeed. One option that occurs to me is either remove Holy Sites as a district type and just allow the associated buildings to be placed in the city centre, or at least don't count them towards the district limit (since their utility as districts for GP generation fades very rapidly). At the very least that minimises the economic penalty the AI suffers for spamming them inappropriately.

They really need to stop the religion obsession.

 

[Ferocitus]

They rarely build commercial and industrial zones...

How many games have you played to reach that conclusion?
I have seen them build many of both districts. Some civs might give
preference to other things early in the game, but they definitely
can pump out comm and ind districts later.

But sure, at this rate it would be nice to have ai that works at all.

Ridiculous hyperbole.

 

[PhilBowles]

Starcraft (both)'s AIs are mostly one size fits all and there are many different maps so it still affects it there. Even on the same map depending on spawns, you could be much closer than usual and distance most certainly does matter. And while the AI might be sub-optimal in some cases, I think that's a sacrfice I'd be willing to make.

Starcraft still remains a very simple game mechanically. There are very few axes on which an AI can underperform: timings (not an issue, because they're programmable), build orders (both programmable and, thanks to linear tech trees and low numbers unit, upgrade and building options, tightly constrained), numbers of units (not an issue as you describe below), and micro (where it is generally poor other than timing to activate abilities).

Civ 4's AI was capable of plotting wars and dropping them if it could not manage the proper attack force.

As was Civ V's, though in both cases the AI then tended to stand around rather obviously with large stacks/carpets looking forlornly across the border. I don't think there's any dispute that Civ VI AI is poor by the standards of at least the last two, somewhat comparably complex (Civ V moreso than Civ IV from a mechanical perspective) Civ games. My point was simply that Civ gameplay isn't sufficiently stereotyped for the AI to be made variable by difficulty level. That, and it's more important to fix its baseline performance anyway - bonus-based AI difficulty isn't a problem, the problem is when those bonuses don't help the AI to win because flaws in its coding prevent it exploiting them effectively.

Funny, I thought the HD version of AOE 2 had an amazingly competent AI (relatively speaking) at least against casual players.

You could be right - I've only played campaigns. But I don't recall AoE AI being anything special back in its day (I always preferred the first game to AoE 2).