Whatever they do I hope the AI is substantially upgraded

[SupremacyKing2]

@SupremacyKing2 The behavior you describe in the post is a result of how Operations and partially Tactics work in Civ6. It has nothing to do with "role playing" (there is no such thing in Civ6) nor "not to crush human player" (I wish it was that simple).

Every city attack the AI performs is a separate operation that recruits its own units. The target for the attack is chosen based on city valuation and distance. Issues:
1. No bigger picture when running a war - this one is true. Each operation is independent and runs as long AI calculates it can win. Once it determines it cannot win, it either retreats or looks for another target.
2. The vanilla game allows for operations as far as 30 tiles away from you. That explains why AI makes stupid attacks deep into your territory. It didn't matter in vanilla because you could keep that city buy with R&F it changes drastically as such a city flips over anyway.
3. City valuation is quite complex, but in general Original Capitals are valued 200-300 pts. more to incetivize Dom victory. It so happens that each Minor is also an Original Capital and that is why AI values them so much (at least early and mid game, later this value is less important). There is no easy fix for that as there is no parameter to distinguish Majors from Minors when running an operation.
4. The actual behavior on the map is controlled by Behavior Trees. In one of latest patches Firaxis added a shortcut to faster take cities called "CanAlphaCity". I suspect that there is a bug in the implementation here that AI only executes it correctly when a melee unit is adjacent to the city. It is how the attack was performed before that change - melee units are surrounding a city, AI takes its health to 0 and then captures the city.

Thanks. That is very informative. I think it does support the main point I was trying to make. The problem with the AI has more to do with the higher level strategy than basic capability. For example, the AI is capable of capturing cities but the logic it follows to decide which city to attack or when to attack is flawed and predictable. I also think it is this predictability that makes it weak because once a player figures what makes the AI tick then the player can figure out the best counter and the AI will not be as difficult to beat anymore.

 

[Victoria]

I also think it is this predictability that makes it weak because once a player figures what makes the AI tick then the player can figure out the best counter and the AI will not be as difficult to beat anymore.

I love it when the AI decides to attack a city and their troops have to march past another walled city I have to get to it. I have even planted a settler and chopped in a wall to take advantage of it. I did feel bad afterward, but in a good way.

 

[Trav'ling Canuck]

@SupremacyKing2 The behavior you describe in the post is a result of how Operations and partially Tactics work in Civ6. It has nothing to do with "role playing" (there is no such thing in Civ6) nor "not to crush human player" (I wish it was that simple).

Every city attack the AI performs is a separate operation that recruits its own units. The target for the attack is chosen based on city valuation and distance. Issues:
1. No bigger picture when running a war - this one is true. Each operation is independent and runs as long AI calculates it can win. Once it determines it cannot win, it either retreats or looks for another target.
2. The vanilla game allows for operations as far as 30 tiles away from you. That explains why AI makes stupid attacks deep into your territory. It didn't matter in vanilla because you could keep that city buy with R&F it changes drastically as such a city flips over anyway.
3. City valuation is quite complex, but in general Original Capitals are valued 200-300 pts. more to incetivize Dom victory. It so happens that each Minor is also an Original Capital and that is why AI values them so much (at least early and mid game, later this value is less important). There is no easy fix for that as there is no parameter to distinguish Majors from Minors when running an operation.
4. The actual behavior on the map is controlled by Behavior Trees. In one of latest patches Firaxis added a shortcut to faster take cities called "CanAlphaCity". I suspect that there is a bug in the implementation here that AI only executes it correctly when a melee unit is adjacent to the city. It is how the attack was performed before that change - melee units are surrounding a city, AI takes its health to 0 and then captures the city.

@Infixo I appreciate you taking the time to provide these explanations. Thanks!

 

[SupremacyKing2]

@SupremacyKing2 The behavior you describe in the post is a result of how Operations and partially Tactics work in Civ6. It has nothing to do with "role playing" (there is no such thing in Civ6) nor "not to crush human player" (I wish it was that simple).

Every city attack the AI performs is a separate operation that recruits its own units. The target for the attack is chosen based on city valuation and distance. Issues:
1. No bigger picture when running a war - this one is true. Each operation is independent and runs as long AI calculates it can win. Once it determines it cannot win, it either retreats or looks for another target.
2. The vanilla game allows for operations as far as 30 tiles away from you. That explains why AI makes stupid attacks deep into your territory. It didn't matter in vanilla because you could keep that city buy with R&F it changes drastically as such a city flips over anyway.
3. City valuation is quite complex, but in general Original Capitals are valued 200-300 pts. more to incetivize Dom victory. It so happens that each Minor is also an Original Capital and that is why AI values them so much (at least early and mid game, later this value is less important). There is no easy fix for that as there is no parameter to distinguish Majors from Minors when running an operation.
4. The actual behavior on the map is controlled by Behavior Trees. In one of latest patches Firaxis added a shortcut to faster take cities called "CanAlphaCity". I suspect that there is a bug in the implementation here that AI only executes it correctly when a melee unit is adjacent to the city. It is how the attack was performed before that change - melee units are surrounding a city, AI takes its health to 0 and then captures the city.

I wanted to add that I think it would be much better if the AI was programmed to attack cities based on distance from the front line and loyalty. That's basically what I do. I attack the closest city to the front line that will be easy to keep from flipping and attack it as quickly as possible. When I capture it, I keep my units in place and heal up and then move quickly to the next closest city. I keep going like this, shifting the front line with every city I capture. And this way, I get a continuous territory that is easy to defend and keep. Capturing cities based on an objective valuation score divorced from any strategic context is a bad way for the AI to wage war. So I would suggest changing the value score for cities to factor in distance and post capture loyalty.

 

[Trav'ling Canuck]

I wanted to add that I think it would be much better if the AI was programmed to attack cities based on distance from the front line and loyalty. That's basically what I do. I attack the closest city to the front line that will be easy to keep from flipping and attack it as quickly as possible. When I capture it, I keep my units in place and heal up and then move quickly to the next closest city. I keep going like this, shifting the front line with every city I capture. And this way, I get a continuous territory that is easy to defend and keep. Capturing cities based on an objective valuation score divorced from any strategic context is a bad way for the AI to wage war. So I would suggest changing the value score for cities to factor in distance and post capture loyalty.

It's a tough gig for the AI developer to guess at what the best approach should be when the game is still in development. The dev team will tell you how they think it should work, but that's not always going to be the best advice. Plus, as the rules change, some of the AI programming should change, too, which isn't always going to happen (see conversation about re targeting cities and new impact of Loyalty).

In the end, we need:
(a) a final set of rules (which is a big reason I'm personally hoping there is no 3rd expansion)
(b) time in the field to identify what the AI does well and where it's approach is completely at odds with what better players will do
(c) the tools for motivated modders to make adjustments based on (b).

In the interim, what might help most at this time is better balance between options, particularly with respect to production costs. Wonders being more valuable, on average, relative to their production costs (including the risk of losing 50% of those costs), would really help the AI, for example.

 

[Infixo]

@SupremacyKing2
I experimented a bit with distance params for operations in the latest version of Real Strategy mod. I achieved the situation where AI mostly goes for border cities or cities close to the border.
The 1st problem here is that it is unknown how exactly this distance is calculated (e.g. from city to city, or from city to the border, etc.) Also, you cannot go down too much because early game cities still are a bit far away from each other and there still needs to be an option to attack such a city.
The 2nd problem is yet again related to city valuation. Each city has some residual value related to its infrastructure (districts, buildings, etc.) Mid and late game it becomes the most important factor (can reach thousands pts). Border cities, as usually settled later, tend to be less developed so they are naturally scored lower. I had such a situation with Shaka in Africa - he wanted to go for a city in Middle East. However, there was a city along the way, that he should take first. But he simply din't want to because it was a small one, will little infra, and the scoring was 0 for it.
As for loyalty, I haven't seen any evidence suggesting that it is considered during target selection. For me proof enough that is isn't is the fact that AI attacks cities deep in enemy's territory.

(c) the tools for motivated modders to make adjustments based on (b).

Except for the ultimate tool which is the source code, all tools are already available, imho. Or do you have anything specific in mind?

 

[TheMeInTeam]

New audience explains everything. What @TheMeInTeam likes to call "doormat market". I prefer to see it as the instant gratification generations market (which also explains everything, and not only in gaming)...

On the other hand, plain incompetence would also explain a lot, that's why I cannot disregard either... hard to say, really.

It's a lot easier to conclude plain incompetence when a game does poorly.

For successful games that are in many ways shoddy, such as Civ 6, they have clearly allocated resources to pull in players specifically, at the expense of things like basic UI or designing a game the AI can play. They had a posting for a UI person *after* the game dropped, right? Would they do that with the people doing animations for leaders? I suspect not, but it's the same practice in both cases.

Apparently the doormat market doesn't care as much for UI as it does for the leaderheads (compare threads about how attractive leaders are vs threads about UI). It COULD be that Firaxis doesn't realize this and blundered into emphasizing the things to sell an instant gratification market. That's possible in principle, but two games in a row with such practices makes me suspect otherwise.

Meanwhile Pdox has their shovel out too with recent patches to EU 4 and their approach to HOI in general, trying to see who can lower the bar further I guess.

If I am right then the reason the AI is "weak" has more to do with the devs' design philosophy rather than lack of effort into the AI. In other words, it's not that the devs are incapable of making a good AI, they just choose to make an AI that is good in terms of their design goals for the game but that unfortunately, is also easily taken advantage of by competent human players.

This shifts blame from AI programmers to designers, and I strongly suspect that is where blame belongs in most strategy games. Civ included.

The AI in Civ games consistently acts dishonestly compared to the in-game incentives. The devs can't handle the design of these incentives...so they leave that to MP where people talk about "gamey" practices and deliberately make the AI game-throw. They have done this in every single Civ game since at least Civ 4, likely earlier too.

It must be hard to be an AI programmer when the design of the game dictates your AI isn't allowed to try.

 

[isau]

One thing the AI does very poorly IMO is keep stock of how many friends it is actually advantageous to keep. Basically, it views each civilization completely individually and never as part of a whole. A brighter AI would try to keep only 2 or 3 friends max (ideally people who are friendly with each other) and be indifferent to or actively antagonize anyone beyond that.

This would more or less mirror how real players play. Being friendly with everyone, or angry at everyone, is pretty much always a losing strategy.

 

[Kibikus]

This would more or less mirror how real players play. Being friendly with everyone, or angry at everyone, is pretty much always a losing strategy.

For culture and religious victories, being friends/allies with everyone is usually very helpful.

 

[darko82]

@darko82

Imagine thinking stylised graphics are down to "the kids" (which seems to be the default stand-in for "new audience" when people mention it) instead of a growing realisation amongst industry professionals (and marketing, and so on) that the endpoint of hyperrealisation in graphics is a) unsustainable and b) not actually desireable wrt. psychology and the uncanny valley effect. Nevermind scaling that to a 4x or other TBS game where you have to fit so much on the screen at once.

I wonder if you think the same about Endless Legend? Heavily stylised, definitely complex, definitely the same kind of market.

We seem to be in an odd age. We live in a world where we demand more complicated mechanics, more polished mechanics, higher visual fidelity, but when the one thing that isn't any of those (the AI) doesn't perform up to scratch (because it's rarely one of the things asked for) we get these weird sweeping generalisations about who games are made for, in some amazingly negative context.

Endless Legend has a fantasy setting. A different vibe. Colorful and stylized graphics is more acceptable to me for obvious reasons. Civ games that are related to history, war, politics, science, development, real world issues etc. are harder to get immersed with that kind of vibe. Stylised graphics are not down to the kids, but I think that stylised graphics is not the right path for Civ games. The new audience is more likely to grab it. That's why it is stylized in case of Civ games.

However, I really do not like this ugly Civ V graphics. It looks very ugly today. So I prefer Civ VI graphics because it is a lot more pleasant to look at. It is TOO colorful, but I was able to reshade it to my own taste.

Civ I has the best graphics in the series, to my mind. 3D is not the best for many games, and it costs so much.

 

[acluewithout]

@SupremacyKing2 The behavior you describe in the post is a result of how Operations and partially Tactics work in Civ6. It has nothing to do with "role playing" (there is no such thing in Civ6) nor "not to crush human player" (I wish it was that simple).

Every city attack the AI performs is a separate operation that recruits its own units. The target for the attack is chosen based on city valuation and distance. Issues:
1. No bigger picture when running a war - this one is true. Each operation is independent and runs as long AI calculates it can win. Once it determines it cannot win, it either retreats or looks for another target.
2. The vanilla game allows for operations as far as 30 tiles away from you. That explains why AI makes stupid attacks deep into your territory. It didn't matter in vanilla because you could keep that city buy with R&F it changes drastically as such a city flips over anyway.
3. City valuation is quite complex, but in general Original Capitals are valued 200-300 pts. more to incetivize Dom victory. It so happens that each Minor is also an Original Capital and that is why AI values them so much (at least early and mid game, later this value is less important). There is no easy fix for that as there is no parameter to distinguish Majors from Minors when running an operation.
4. The actual behavior on the map is controlled by Behavior Trees. In one of latest patches Firaxis added a shortcut to faster take cities called "CanAlphaCity". I suspect that there is a bug in the implementation here that AI only executes it correctly when a melee unit is adjacent to the city. It is how the attack was performed before that change - melee units are surrounding a city, AI takes its health to 0 and then captures the city.

I’m constantly amazed this site has so many clever people. Thanks for this. Super interesting.

It's a tough gig for the AI developer to guess at what the best approach should be when the game is still in development. The dev team will tell you how they think it should work, but that's not always going to be the best advice. Plus, as the rules change, some of the AI programming should change, too, which isn't always going to happen (see conversation about re targeting cities and new impact of Loyalty).

In the end, we need:
(a) a final set of rules (which is a big reason I'm personally hoping there is no 3rd expansion)
(b) time in the field to identify what the AI does well and where it's approach is completely at odds with what better players will do
(c) the tools for motivated modders to make adjustments based on (b).

In the interim, what might help most at this time is better balance between options, particularly with respect to production costs. Wonders being more valuable, on average, relative to their production costs (including the risk of losing 50% of those costs), would really help the AI, for example.

No! Third Expansion please! There’s just ... so many more things to do!

But yeah. Hard to really nail the AI when the rules keep changing. If we get another expansion (#onemoreexpansion ... that’s not going to become a thing is it? Sigh...) then I hope after that we’ll get a few more just cosmetic dlc, so FXS can finally just focus on balance and the AI.

 

[darko82]

But yeah. Hard to really nail the AI when the rules keep changing.

This argument is false. The AI has never been at any satisfactory level. The changing rules have never made the AI worse, broke it or whatever. It's been always bad. When the game was released, the AI should have managed the systems it was designed for. Regardless of the new systems. That being said. No excuses. I would understand this argument if the AI became worse after releasing the new expansion. It's not the case... on the contrary, many people say it's better now (it was improved) after the expansion was released ... LMFAO. And the best part, it will never be fixed by Firaxis, no matter how many expansions get released. Sad but true.

 

[Leathaface]

But Civ VI still has quite a lot of areas where it's just broken. I just had a game where I played an island map, took over one of Australia's cities on the coast though I was pretty roughed up at the end of it. And, as it turned out, Australia just kept churning out galleys in the rest of his cities (this was early on in the game). He had a much bigger navy than I did and what did he do with it? Absolutely nothing. He kept building them and then they just sat there, not moving, not doing anything except allowing my pitiful archers sink them on by one. They could've easily taken their city back, no problem whatsoever. No strategy needed, just smashing into the city.
As a sidenote, he also kept building archers in the abovementioned city before I conquered it, which is good... but, he never let it stay in the city, he always moved the archer out so I could easily kill it. Again, what the hell? It had targets to shoot from within the city.

I don't think that's asking for "great AI", it's just asking for an AI that works. But it's broken, in many ways. And it is in my experience not uncommon at all to see this kind of strange behavoir.

I agree with your post, why does the AI move Archer units out of it's cities? That is just simply terrible AI, the Archer should stay in the city centre at all times and shoot.

I believe we are bashing the AI a little bit too hard here. I think it works fine for the religious victory as well as the culture victory. domination and science game is weak , i agree but it is good at building stuff ( exception being the ridicolous wonders in cities that dont benefit at all ) . Also they ally themselves against warmongers although since they are weak at army control ( domination game ) this aspect doesnt benefit them. overall the bad domination game/army control hurts a lot since there is no penalty for going wide and a lot of incentives the AI falls behind the longer the game continues. Hopefully if they or the modders can get that part just a little bit better the game would become a lot more challenging.

In close to 1000 hours of playing Civ VI (much of it on higher difficulty) I've never seen the AI win a Culture victory, not even once. So the AI should get better at winning at it.

The only way a Civ with high Culture/Tourism is a challenge is if you are going for a Culture victory and another AI is strong in Science, the AI with lots of Culture could keep you from winning long enough that the Science focused Civ gets a Science victory.

 

[Gorbles]

Endless Legend has a fantasy setting. A different vibe. Colorful and stylized graphics is more acceptable to me for obvious reasons. Civ games that are related to history, war, politics, science, development, real world issues etc. are harder to get immersed with that kind of vibe. Stylised graphics are not down to the kids, but I think that stylised graphics is not the right path for Civ games. The new audience is more likely to grab it. That's why it is stylized in case of Civ games.

However, I really do not like this ugly Civ V graphics. It looks very ugly today. So I prefer Civ VI graphics because it is a lot more pleasant to look at. It is TOO colorful, but I was able to reshade it to my own taste.

Civ I has the best graphics in the series, to my mind. 3D is not the best for many games, and it costs so much.

Fair enough! I still don't get what you mean by "new audience", but I think I've gone a bit too off-track, hah

 

[darko82]

Fair enough! I still don't get what you mean by "new audience", but I think I've gone a bit too off-track, hah

New audience = new young generations that like colors, animations, cartoons. Yeah. And maybe not the best AI...

 

[Kruos]

Like everyone I hope also that the AI will be greatly improved with GS.

Since release the AI has made big progress on both fronts, tactical and strategical, but there are still big issues pending.

Personaly I think its tactical level is still mediocre but now 'playable', compared to vanilla release the AI is now able to capture city and can be a threat now at deity level. There are still big room for improvments of course, like a fix to this stupid trend of puting its archers on the front line in some situation (seen on the Ottomans stream of last week) or the aircraft management (carriers not carrying aircraft) for example.

On the strategic side I would say that the AI needs a bigger effort now. It needs to be better 'focused' on the victory objective pursued. It's not normal that on deity level, with all its bonuses and with a powerhouse civ like the corean for example, the AI is not able to win space before ~ T300 (on standard continental game).

 

[Victoria]

My 18 year old said 2 night ago that phones have shortened his attention span and pointed out the phone games like Clash of clans are designed for fast win gratification. It may be that you only win 50% of your games but you play so many that it feels like a lot of wins... he was rather insightful. He used to like civ but now says it takes too long.
I guess with single player, they are longer games so you need more wins?

 

[Gorbles]

20 years ago, folks were saying TV shortened peoples' attention spans and made them less communicative and more prone to instant gratification.

60 years ago, folks were saying the same about radio.

And apparently a generation or two before that, books were claimed to have a similar effect.

I'm an adult who played the first Civilisation, and use a smartphone every day (I work with them, it's somewhat unavoidable even if I didn't like them). Video games have always had a focus on the graphical, and while preference is absolutely subjective, it's always been a marketing (and thus development) focus to make an appealing art style that fits the product and sells enough copies. It's why DOOM looks like it does (and why the remake found so much success). It's why Dishonoured looks like it does.

"AI" is never asked for outside of dedicated fan forums where players on average vastly exceed the average consumer's investment in the franchise. It's never about attention span, or "the kids", or whatever. It's literally never been a priority in any game where AI is a thing. RTS games are rife with AI issues. Most of them feature scaling economic benefits to higher AI difficulties. The same goes for TBS games, but it's easier to exploit because without the mimicry of real-time movement it's easier to observe the patterns the AI exhibits, and thus take advantage (all units can only move X in a specific turn, which means it limits their resulting end positions. RTS games don't suffer from this specific limitation, and therefore pathfinding becomes more of the visible bottleneck).

By all means, ask for better AI. There have been some great posts in this thread from people who know what some of the limitations are in the modding tools, and understand well how the existing AI can process the game. I've found them very informative. But don't get stuck on pop psychology, because 20 years ago (I'm taking an average here, don't shoot me, haha) the generation before you were saying the same things . . . but about you!

 

[Victoria]

But don't get stuck on pop psychology

And don’t assume it’s just pop, there are studies that show attention spans have dropped just like there are studies that show violent games do not make violent children. My 18 year old also now thanks us for banning phones in their rooms and limiting TV to 1 hour a week.... while I was of that generation when people ate dinners in front of the TV we eat at the table and we all now appreciate the value this has bought the family. However he is 18 he has started using his phone in his room. We may know what is right but what’s the fun in following it? Fact is, before he used a phone he liked civ, now it takes too long.

 

[acluewithout]

And don’t assume it’s just pop, there are studies that show attention spans have dropped just like there are studies that show violent games do not make violent children. My 18 year old also now thanks us for banning phones in their rooms and limiting TV to 1 hour a week.... while I was of that generation when people ate dinners in front of the TV we eat at the table and we all now appreciate the value this has bought the family. However he is 18 he has started using his phone in his room. We may know what is right but what’s the fun in following it? Fact is, before he used a phone he liked civ, now it takes too long.

Cal Newport has some interesting stuff on this if you can tolerate his writing style. His book Deep Work has influenced me a lot. He’s why my phone doesn’t have internet and is set to gray scale.

There’s a lot of silly stuff written about attention. Really, the core insight is that attention is to some extent skill, or really a set of related skills. Practice controlling attention and it improves. Incentivise paying attention and you’ll practice. So, go on a holiday without internet, and you’ll find you’re motivated to read, and will then get better at reading.

Phones. iPads. Tv. Whatever. They incentivise not paying attention, but instead moving from thing to thing. So, you end up less good at controlling your attention.