Hope for Combat AI Improvement?

[Eternalsteelfan]

The biggest impasse for me between completely giving up CIV IV for V has been the AI's extremely poor use of combat units. The AI just doesn't seem to be able to wrap it's processors around 1UPT as it currently stands (and has stood). Has there been any indication or reason to hope that the AI will get some work in the new expansion?

 

[Veneke]

We can (hopefully) expect a better AI but in the form of diplomacy not warfare. It's possible that the enhanced diplomacy will mean that you will encounter more wars with and against coalitions of AIs, especially in the late-game. They've said nothing about the tactical AI and personally I think they can't do anything to significantly improve it; the AI just can't deal with 1UPT given the constraints placed on it.

 

[otaman1]

I do hope AI combat improves. I always hate it when AI's start sieging a city with like 20 units but they still cannot siege that city. Also, when an AI declares war on me, it seems like they never wanna give up even though I pretty much destroy all their units

 

[Eternalsteelfan]

I wouldn't say "impossible", there's plenty of room to improve the AI's decision-making and logic; they very poorly manage ranged units and how they set up combat lines and armies. I agree that siege warfare is especially poorly handled, I think it's inherently problematic with how defensible cities are. Taking cities can be costly even for a well-planned player assault, of course the AI will stumble.

 

[Aussie_Lurker]

Actually, I've seen some examples of very intelligent combat from the AI in my current game. Korea, Russia & I were only able to conquer England as an alliance, & even then we lost a lot of units in the process. Korea, in particular, showed excellent use of its siege units, followed up by a combo of melee & mounted units to finish Elizabeth off. It was beautiful to watch.

 

[Eternalsteelfan]

Actually, I've seen some examples of very intelligent combat from the AI in my current game. Korea, Russia & I were only able to conquer England as an alliance, & even then we lost a lot of units in the process. Korea, in particular, showed excellent use of its siege units, followed up by a combo of melee & mounted units to finish Elizabeth off. It was beautiful to watch.

I think Gods and Kings improved it some, I've seen fortresses from Great Generals and a lot more civs wiping out other civs. Unfortunately, that's the AI vs AI component; in 150 hours recorded by Steam I've not once seen the AI pose anything remotely close to a threat to a player. A few mixed units in the tiles around a city is basically an unassailable position for the AI to threaten.

You can bump up the difficulty but that only inflates the AI civs, the root cause of terrible combat AI is still there.

The hope would be the patch coming with Brave New World helps, as thee patches with Warlords and Beyond the Sword greatly improved the AI in CIV IV and G&K's may have for V.

 

[Veneke]

Actually, I've seen some examples of very intelligent combat from the AI in my current game. Korea, Russia & I were only able to conquer England as an alliance, & even then we lost a lot of units in the process. Korea, in particular, showed excellent use of its siege units, followed up by a combo of melee & mounted units to finish Elizabeth off. It was beautiful to watch.

Oh aye, the AI does well at times. The problem is that it's incapable of replicated those examples of good gameplay consistently during war, especially if the terrain is broken or constrained by water. Even there, however, I've seen some reasonably decent attempts at amphibious invasions, and decent navies to boot, but it seems unable to replicate that consistently and it breaks down much more often if faced with anything unusual, like choke-points. Don't suppose you've a couple of screenshots of taking on Lizzie? I've noticed that England AI tends to do pretty well on the defensive if they have rough terrain on their side and a number of Longbows. Assaulting range 3 units under normal conditions is always a messy affair, doing it across hills/forests/mountains is rather more difficult.

 

[otaman1]

In one of my game, I was playing the Danish. All of a sudden Egypt declared war on me for no apparent reason. After setting up defense lines, I haven't seen a single Egyptian unit yet. After 20 turns, I negotiated peace and Ramesses agreed and he was willing to give up Heliopolis. No one lost a unit in the battle.

 

[Zaimejs]

Otaman, I'd say that's just great playing on your part!

Maybe it is the difficulty of taking cities that is the BIG issue with the AI. I mean, I have to sit there an pound, pound pound a city with siege... holding the line with supporting melee... it is very complex and I have to carefully plan the attack or I risk losing all of my troops.

But there is a math to it. To take a decently defended city, you probably need about an 8-1 unit advantage (if the units are of the same era). If I'm ahead by an era, I can do this with half. If I'm behind, I don't even want to try...

But there is a formula. 2-3 siege... 2-3 melee... 2-3 mounted or bowmen.

If you don't have enough... even by 1, then you risk losing all your units and their city heals to full health, and it was a complete waste of time.

 

[Seek]

Otaman, I'd say that's just great playing on your part!

Maybe it is the difficulty of taking cities that is the BIG issue with the AI. I mean, I have to sit there an pound, pound pound a city with siege... holding the line with supporting melee... it is very complex and I have to carefully plan the attack or I risk losing all of my troops.

But there is a math to it. To take a decently defended city, you probably need about an 8-1 unit advantage (if the units are of the same era). If I'm ahead by an era, I can do this with half. If I'm behind, I don't even want to try...

But there is a formula. 2-3 siege... 2-3 melee... 2-3 mounted or bowmen.

If you don't have enough... even by 1, then you risk losing all your units and their city heals to full health, and it was a complete waste of time.

I think the current formula is 4-5 bows, 1 mounted (and maybe 1-2 siege if you have the extra hammers).

 

[Eternalsteelfan]

I agree, city-taking can be very costly and difficult even as a player. Rather than go into a separate debate about balance, however, there is still an AI issue: assuming the defense of a city is powerful and easily done with a few units, the AI fails to do this itself.

 

[Hulking Lummox]

What I don't understand is why Leader personalities and flavors have no effect (or seemingly no effect) on AI combat tactics. A civ like the Mongols should do maximum damage from out of reach, a civ like England should blockade all of your ports and claim the seas, a civ like Rome or Greece should approach in a calculated and dominant fashion, a civ like China should play you like a game of Go with give-and-take but a strategy of surrounding you. My favorite would be if Napoleon tirelessly jockeyed for positioning with you before attacking (but then he could get stubborn and charge too).

Alas no major tactical differences between AI's, just Unique Units. Random personalities could be all the more interesting.

Also there should be flavors on a game-by-game basis for whether that civ is going to play defensive, passive or aggressive with it's approaches. If the AI detects range units it ought to set up and hold position out of range if defensive - or charge if aggressive. Passive would be more of a unit by unit assessment of what seems good. Simple.

 

[nokmirt]

In my current game. I a playing as Germany. England just declared war on me. It looks like they are on a separate continent or large island. I have it on continents plus. Anyway I am interested in seeing if England performs a real amphib landing against me. There are only about 4 water tiles between us.

 

[Aussie_Lurker]

In one of my game, I was playing the Danish. All of a sudden Egypt declared war on me for no apparent reason. After setting up defense lines, I haven't seen a single Egyptian unit yet. After 20 turns, I negotiated peace and Ramesses agreed and he was willing to give up Heliopolis. No one lost a unit in the battle.

Actually, its possible that Egypt might have attacked a city-state you pledged to protect. That happened to me in my current game.

 

[otaman1]

Actually, its possible that Egypt might have attacked a city-state you pledged to protect. That happened to me in my current game.

Nope. Egypt was on the right side of the map. My CS allies were no where near Egypt.

 

[Veneke]

What I don't understand is why Leader personalities and flavors have no effect (or seemingly no effect) on AI combat tactics. ...
Alas no major tactical differences between AI's, just Unique Units. Random personalities could be all the more interesting.

...

I think they tried to do that, but I agree that it's never really worked or if it has then it's entirely too subtle. The different leaders, however, do have different flavour preferences for unit mixes. That's not quite what you're getting at, but it's close. You can see them here: http://forums.civfanatics.com/showthread.php?t=409062 (they're in orange).

 

[Hulking Lummox]

I think they tried to do that, but I agree that it's never really worked or if it has then it's entirely too subtle. The different leaders, however, do have different flavour preferences for unit mixes. That's not quite what you're getting at, but it's close. You can see them here: http://forums.civfanatics.com/showthread.php?t=409062 (they're in orange).

Thanks for sharing that! Very good information. But you're exactly right, they didn't do enough of it because the AI varies strategy-wise but behaves pretty uniformly with it's tactical style - barring army composition. There always seems to be the same goal in the "mind" of the AI's which is just "I want this city, here is a formulaic attack." It only gets interesting when a second front opens up with another civ and you have to defend against two formulaic attacks on separate cities.

Like the OP indicated, it would be really nice to get some improvements in this expansion but I would personally pay $20-$30 just for an AI overhaul DLC. It would be so great if the chart that Veneke linked to could be expanded for the game to include additional attributes that affect things such as: the likelihood the AI will aim a surprise attack against a very out of the way city in your empire, or that they will spread out and focus on occupying your territory and fighting a war of attrition and starving you rather than capturing your cities (going for an favorable peace treaty), or that the AI will attempt an ingenious and unconventional attack such as an army spearheaded by excessive numbers of tanks. Their flavors could even change by era. Anyways, it seems like there is more than enough room for improvement to validate a pricey DLC (and enough interest!)

 

[Veneke]

If an AI overhaul DLC was released I think that they'd receive a ton of flak for it, and rightfully so. If Civ was primarily a multiplayer thing then a DLC to get singleplayer to work with decent AI would be one thing; to turn the AI component of a predominantly singleplayer strategy game into paid DLC though is altogether too much. If they did that, and it worked, they'd probably just ignore the AI totally for Civ 6 and get us to pay double for a fully functioning game. Of course, it's reflective of the sad state of the gaming industry that things are leaning that way. That said nobody could possibly object to an improvement in the combat AI, but as a patch or inclusion in an expansion pack.

Different manners of attack would certainly be nice, and probably more doable than a flat-out improvement to the combat AI. It was speculated, in another thread, that the World Congress and enhanced diplomacy (and later on with distinct ideological blocs) that the AI might get better at war simply by having more chances to cooperate together against the player. The biggest problem the AI faces right now is that even if it's a runaway civ it's typically unable to bring enough units to bear at any one point to beat a human player. If a bunch of AIs though all bring their stuff against a couple of points there are better odds that the human player can't defend against all of these attacks. So while the AIs still might not be able to take out a human player, they could probably whittle away his support base which is looking far more important than it did previously. This is all very speculative though at this point.

 

[alpha2117]

I have used City States to kill off other civs and they have always seemed to use their units okay. I think the larger a civ is the worse the AI uses it's units. In general if it's a small empire with limited borders and focus it isn't too bad - it's when it gets larger that it simply seems to make tactical errors.

 

[Trias]

The problem with improving the AI is that it is a LOT of work.

The improvements to the civ4 AI in the warlords and BTS patches were for a large part due to the effort of the community. The contributors to the Better AI project literally sunk in thousands of man hours. If they would have needed to be paid, that would have cost a fortune.

Out best hope for an improvement of the combat AI is a community effort. Note that this is something everybody can help with, by identifying situations in which the AI systematically makes the wrong choice. (Although it helps to keep in mind that the AI makes his decisions almost purely on a unit by unit basis.)