Better AI in BNW?

[ptoss1]

This makes no sense, since the AI already goes for other victory conditions besides World Conquest.

What he probably meant is that the AI shouldn't be so aggressive just because they settle cities near you and gain a "we want your land now" and "nearby border" negative modifier. This leads to war in most cases and forces you to build a defensive army, damaging trading and alliance relations.

 

[Putmalk]

What he probably meant is that the AI shouldn't be so aggressive just because they settle cities near you and gain a "we want your land now" and "nearby border" negative modifier. This leads to war in most cases or just damages trading and ally relations.

The AI as it stands right now (in gods and kings) does not directly want war because of the covets lands modifier (which, fyi, is very bad wording - covets lands = "land dispute" and land dispute is computed based on player proximity and plot buying aggression).

However, the covets lands modifier does contribute to a negative opinion of you, and opinion does affect the AI's desire to go to war here.

 

[ptoss1]

The AI as it stands right now (in gods and kings) does not directly want war because of the covets lands modifier (which, fyi, is very bad wording - covets lands = "land dispute" and land dispute is computed based on player proximity and plot buying aggression).

However, the covets lands modifier does contribute to a negative opinion of you, and opinion does affect the AI's desire to go to war here.

Yes, but the point is that because the AI was the one that settled near the player, and not the other way around, technically the AI shouldn't be all upset since it was the one being the aggressor. It's like the AI wants to ruin the relationship when it was all dandelions and flowers before.

 

[Putmalk]

Yes, but the point is that because the AI was the one that settled near the player, and not the other way around, technically the AI shouldn't be all upset since it was the one being the aggressor. It's like the AI wants to ruin the relationship when it was all dandelions and flowers before.

The AI doesn't work like that.

The AI calculates expansion aggressiveness (for land dispute levels) by checking your closest city's distance in relation to their capital, and your capital. So if you settle a city closer to their capital than yours, it's considered medium aggressiveness (already). And if that city is further from your capital than it is from the distance between both cities, the it's really aggressive. But that's the key: it's your city, not theirs. So they can't settle near you and then blame you for being aggressive. You settled near them.

 

[ptoss1]

The AI doesn't work like that.

The AI calculates expansion aggressiveness (for land dispute levels) by checking your closest city's distance in relation to their capital, and your capital. So if you settle a city closer to their capital than yours, it's considered medium aggressiveness (already). And if that city is further from your capital than it is from the distance between both cities, the it's really aggressive. But that's the key: it's your city, not theirs. So they can't settle near you and then blame you for being aggressive. You settled near them.

I wondered how that worked.
Is fog of war taken into account though?
I've seen Alexander "covet my lands" even though he probably didn't know where my capital or any of my cities were. Like on the first turn when we've met.

 

[Putmalk]

I wondered how that worked.
Is fog of war taken into account though?
I've seen Alexander "covet my lands" even though he probably didn't know where my capital or any of my cities were. Like on the first turn when we've met.

No, for the purposes of proximity evaluation the AI does not care if they can see the city or not. You can say this is a bug or intended. Not knowing their proximity from you gimps their diplomacy in many areas.

 

[George Orwell]

Although naval warfare became more relevant in G&K, one thing that bugs me is the AI's complete blindness to naval ranged units. So, I take a coastal city and then decimate their land forces with battleships as they try to take it back. One amusement recently was that I liberated Cape Town so many times my influence was over 1000 and Japan's very pointy sticks became a small twig.

Another thing that bugs me is being called a warmonger by various Civs when all I'm doing is stubbornly defending myself.

 

[Trias]

[...] I personally think that Firaxis should drop 1UPT for the next Civ game as this is an area where the AI struggles - instead maybe they should bring in an army system that allows limited stacking?

I don't think I can go back to stacks tbh. At the least have limited stacks or 2.

@ptoss1
[...]
One thing I do like about Civ 5's 1UPT is there is less tedium moving units compared to Civ 4 so I would like it if they consolidated movement into army groups for the next one as I think the AI would handle it better.

Limited stacking will only cause further problem for the AI, since:
1) It introduces a new optimalization problem which the AI is going to be bad at: stack composition.
2) It is still bad at unit placement (with mistakes having become more harmful given the increased value of the stack.

To help the AI you can do either of two things:
a) Return to unlimited stacking (in which case the AIs lack of talent at stack composition is mitigated by having more units).
b) Allow for stacking, but in such a way that having 1 unit on a tile is always favorable over having multiple units on the same tile. (This allows the AI to resolve unit log jams more easily, but gives it an easy optimal stack (a single unit) to strive after.)

 

[ptoss1]

Limited stacking will only cause further problem for the AI, since:
1) It introduces a new optimalization problem which the AI is going to be bad at: stack composition.
2) It is still bad at unit placement (with mistakes having become more harmful given the increased value of the stack.

To help the AI you can do either of two things:
a) Return to unlimited stacking (in which case the AIs lack of talent at stack composition is mitigated by having more units).
b) Allow for stacking, but in such a way that having 1 unit on a tile is always favorable over having multiple units on the same tile. (This allows the AI to resolve unit log jams more easily, but gives it an easy optimal stack (a single unit) to strive after.)

So sacrifice complexity for AI?
Why don't we go back to chess and checkers...

 

[Winston]

@Trias

Other strategy games use army stacks just fine and I'm sure Firaxis could programme the AI to successfully do the same. The 1UPT system demands alot from the AI and the system was lifted from other games like Panzer General which have much less choke points.

@ptoss1

The current system isn't complex IMO - creating choke points and kill zones with lots of ranged units isn't really that sophisticated whereas an army system done right has plenty of scope for nuance and strategy.

 

[apocalypse105]

You guys seem to have forgotten how Vanilla AI told you they weren't happy when you got close to victory, which was a major immersion breaker. They fixed that. I only get backstabbed when I'm a warmonger, and that was because I kept annexing allied city states, threw 2 nuclear missiles at Babylon, and 3 atomic bombs at Memphis. The nearby city state which I had allied declared war on me immediately which made lots of sense because I just created a nuclear wasteland not 3 tiles away from their territory. Seems pretty legit to me.

I''ve got backstabbed because I declared war on babylon who conquered a city state who i was ally with And I am a warmonger for liberating it? No putting a penalty for declaring war doesn't work

This makes no sense, since the AI already goes for other victory conditions besides World Conquest.

They only go for science not the culture victory or diplomatic

 

[apocalypse105]

The AI as it stands right now (in gods and kings) does not directly want war because of the covets lands modifier (which, fyi, is very bad wording - covets lands = "land dispute" and land dispute is computed based on player proximity and plot buying aggression).

However, the covets lands modifier does contribute to a negative opinion of you, and opinion does affect the AI's desire to go to war here.

Thats just awefully not right if they have a higher military then you and you don't have a lot of positif modifiers which you will have because you're neighbours usally don't aks for a DOF early on. so you only get we are trading ... They just DOW and there is nothing you can do to prevent because

there aren't enough positif modifiers maybe the positif modifier for trading with trade routes will help we will see

 

[Trias]

So sacrifice complexity for AI?
Why don't we go back to chess and checkers...

Neither of those options is any less complex than what we have now in civ5. (If anything option b) is more complex than what we have now.)

In more general considerations. When designing a computer game meant to be played against a computer, then you should not make it any more complex than you can write an AI to handle.

@Trias

Other strategy games use army stacks just fine and I'm sure Firaxis could programme the AI to successfully do the same. The 1UPT system demands alot from the AI and the system was lifted from other games like Panzer General which have much less choke points.

What games with limited stack have an AI that handles it fine? Precious civ titles (Call to Power) have shown just how problematic this can be for the AI.

Limited stacking means that unit/stack positioning is still important, and because the limited stacks cannot end their turn on the same tile, the AI is faced with the same log jam issues when it has lots of stacks in an area.

The only thing that limited stacking does is reduce the number of effective 1upt units on the map. (The same effect can be reached by simply reducing the number of units on the map). The add disadvantage for the AI is that its stacks are inherently less powerful than a human's because they are not good at finding an ideal unit composition for the stack.

 

[Winston]

@ Trias

I was thinking of the newer Total War games as well as older games like GalCiv2, both of which use army stacking mechanics fine IMO - Europa Universalis also appears to use army stacking mechanics though I have no interest in those games so I don't know the specifics. I actually think Firaxis could put together something new and interesting and more tailored to the kind of maps in Civ that finds a happy medium between Stack o Doom and Carpet o Doom.

Edit: I also think Firaxis could pick up ideas from FFGs Civ Board Game too

 

[Greizer85]

The Total War games have atrocious AI, particularly the newer ones. Well - I hear Shogun II is a bit better in this regard but I haven't played it. To speak nothing of the actual battle AI, usually the AI has either too much or too little ranged units in its stacks, enabling you to decimate them with relative ease. And its weakness to mounted archers is legendary; in Medieval I I needed two stacks of 9 unique Horse Archers to sweep the entire map clean.

I'm in favor of unlimited stacking but with negative effects (collateral damage springs to mind), so that 1upt will still be the best in most situations (as per Trias' suggestion). It would help the player too, in clearing the tremendous log-jam issues that Civ V currently has. Someone mentioned Civ V being less tedious than Civ IV in this regard; frankly I want whatever you're smoking. Have you ever tried moving 20 units across the freaking ocean, dodging small islands and enemy ships and nudging them along one by one? Or taking a city that's in the mountains and has 100+ defense and a ranged unit? I'll take my single stack every time, tyvm.

 

[BluegrassGeek]

To be honest, I'd love to see the barbarian units and AI reworked. I see melee barbarian ships just circling continents endlessly until they're shot, and I don't think I've ever seen a barbarian attack a city.

They should beeline to attack cities, and they should get that promotion that lets them steal half the damage they do in gold.

I've had them attack cities, but their priority is units. Even a single worker is higher priority for them. If there's nothing else around, though, they'll sack your city.

 

[Winston]

@Greizer85

I'm not talking about the battle AI for Total War, just the army selections on the campaign maps - with Shogun 2 I think they do a great job here - Civ doesn't need to worry about the battle AI anyway.
What I meant about less tedium was simply that because you build less units you make less moves overall - I think we can all agree the log jam issue is a pain in the ass. Although I think Civ 4 did introduce commands that let you move stacks if I remember correctly. I definitely think there's room for a happy medium between Civs 4 & 5.

 

[Ryoga]

Are you sure these would not simply create a new set of exploitable AI moves?


I am weary of any instruction to the AI using the words "never" or "always" because they scream "exploit me". Nonetheless, I agree that the AI could benefit from better threat mapping.


Of the rules you name, this is the only that is (relatively) simple in terms of AI coding. Simply increase target priority for wounded targets. (This could however lead to the AI being easy to bait.)


Given that one of the design requirements of the AI is that it needs to work with any set of xml rules, this rule may not be implementable.
Also note that even now the AI often finds itself in the position that it cannot take a city because it has run out of melee units. Building fewer melee units is bound to exacerbate that problem.)


#Parsing error unknown variable "front".
The notion of "front" is a rather high level abstraction on a map that is a proiri isotropic. Implementing these rule may not just be "hard" but impossible within the available computing resources.



Possibly, given that the team has no dedicated AI programmers.

But hiring some computer science grad students as summer internes could potentially move the AI along quite a bit.

These are the fixes that I have in mind and that should be relatively easy to do:


1) Learn to use fighters more effectively. Count the number of intercepting units and do airsweeps accordingly before sending bombers. I wouldn't mind if the AI "cheated" here (at least on higher difficulties) by knowing exactly how many intercepting units are active on a certain area even if normally it would be impossible to know with precision.

2) Set fighters in intercepting mode when under attack by bombers. Send in and build aintiaircraft units accordingly

3) Prioritize focused fire when using bombers, and don't let them sit in cities without doing a thing, unless using them would cause the unit to die (again cheating is all right here to know whether it will die or not in advance). Use them on unprotected targets if no better option is available.

4) Prioritize ranged units over melee with a ratio of grossly 2-1

5) Either make submarines truly invisible for the player (you can guess their position as soon as they fire) or make them automatically visible for the AI as soon as you fire for the duration of the next turn.

6) Rewrite the encircling tactic routine for naval units and rather prioritize focused ranged attacks from afar.

7) Make naval units work in groups as soon as a DoW is in place. at least 3-5 ranged with a melee or two for support.

8) have naval units automatically retreat inside the closest available port city when HP are yellow or red. If city cannot be reached in 2 turns, choose nearest friendly border instead.

9) Have naval units retreat inside port cities when isolated or unable to face the enemy (same rules as above). (right now they simply stay near the coast and do nothing waiting for the slaughter). When no port city is available, send them to a suicide attack against a near threat.

10) Learn to use nukes against something other than cities. Prioritize high concentrations of aircrafts/missiles and high concentrations of enemy units when near or even inside own borders (but avoid hitting your city directly)

11) Create a routine to calculate the rate of defensive bonuses of a city in early eras. Hills, forests, mountains, rivers should all have some "points" when directly around a city. Water tiles should be included if no naval unit is part of the conquering force.
If the defensive values are too high, the AI should refrain from attempting an assault until it gets better units or an adequate numbers of them. Prioritize less defended cities when available.

12) When moving calculate the tiles that will be potentially under threat the next turn. Always prioritize the tiles that aren't threatened first and the least threatened after, even if that means not using all the moves available.

13) If a ranged unit can attack a target, let it attack that target rather than simply not attacking at all and waste a turn. No buts unless a retreat action is in order.

 

[Putmalk]

Thats just awefully not right if they have a higher military then you and you don't have a lot of positif modifiers which you will have because you're neighbours usally don't aks for a DOF early on. so you only get we are trading ... They just DOW and there is nothing you can do to prevent because

there aren't enough positif modifiers maybe the positif modifier for trading with trade routes will help we will see

Oh yeah? Where's your proof? I'm prepared to pull source code examples (and logs which show the AI pursuing every victory condition). Are you?

 

[sickVisionz]

and AI is meant to be a lot more loyal or it's easier to make allies.

That's good.

Also doesn't handle naval invasions well yet.

They definitely need to work on that. At times it feels like the AI simply isn't even aware of what a Navy is. Almost like they have AI for land units but when water is introduced they freeze up and are completely baffled.