• Our friends from AlphaCentauri2.info are in need of technical assistance. If you have experience with the LAMP stack and some hours to spare, please help them out and post here.

Status of Civ 6 AI

Joined
Oct 6, 2010
Messages
301
Has the AI improved significantly since launch? Are there still major flaws in the tactical and strategic AI?
 
At the tactical level, based on the first 90 turns in my first post-patch Emperor game, it's still a long way behind Civ V and includes some issues that should be easy to fix. AI missile units appear to be literally unable to fire when garrisoned unless the city has walls - it's encouraging that now the AI is garrisoning most cities, and usually with ranged units, but they keep walking outside to fire (not the best plan vs. a horsemen army).

Target selection remains terrible - on multiple occasions I had units attacking a city and the associated archer unit would turn around and fire at a barbarian scout or galley - and the AI is still mostly incapable of preferentially selecting wounded units to attack, or using units of a type that counters the attacker.

EDIT: It's also as bad as ever in its efforts to attack. Before declaring its not-very-surprising surprise wars it marches towards the capital ignoring any closer cities, and still sends very few units and no backup. Peter initiated the war in which I wiped him out by turn 90 by sending two Warriors at my capital and one damaged Warrior at my second city.
 
Last edited:
Thanks for the update. Still sounds completely busted to me. I don't get why a large developer can't get a decent AI built. I'll continue to wait until it's fixed.
 
Honestly, I've not experienced this even before the patch. Okay, garrisoned units rarely made a sorte, but that was it. The cities always garrison. The city almost always takes a shot at my ranged attackers. And now even garrisoned melee will attack.

Its also been my experience that the ai loves (loved) conquering cities. Never had trouble doing it. But people say in their games the ai can't do it. Well, I've spent hundreds of hours liberating.

I don't use an ai mod.
 
I have many circumstances where the city will not take a shot at me, even when it could. Maybe it thought it couldn't do enough damage or something, but there is a bug somewhere that prevents ranged attacks after certain events. For example, when the AI asks for peace, and then you say no, then after it seems like no ranged attacks happen. I'm sure it's more specific than that, but you get the idea.
 
Have naval AI become any better? I remember that before the patch AI didn't use ships properly- i.e. no stacks of ships (like in civ 5), prepared attacks. it looked like ships were just randomly passing through the oceans, occasionally shooting targets. Civ 5 was much better in this component.
 
Military is still horrible, I ran over two complete Civs in the BC era with 5 archers only loosing a warrior to a chariot in a bad dice roll. Ended up at 20+ cities by 1 AD, on Deity... even civ 5 AI would put up more of a fight.

More importantly AI city management is still a mess, farms everywhere, and always less districts than the cap allows... Holy Sites bloody everywhere, just one Campus built in over 15 cities I took. I understand that combat is hard to code, but econ management? That's just not acceptable at all. Civ 5 AI is dumb as a brick, but it can at least manage its economy.
 
For me, the frustration stems from the excuse that the AI has a hard time with 1 upt. Well, we've been 1 upt for almost 7 years now....

Tactical AI is definitely harder to program for and more processor intensive.

But practically, they've had one person on the AI for basically each of the civ games - possibly less money on AI then on Sean Bean. I don't think it's every been seen as a 'customer' priority outside of priorities like these. I've mentioned before - based on the achievements, the significant majority of players play at Prince or lower. Animated leaders and the like clearly sell the game better than the AI.

However, fingers crossed this might be the game that changes that: in that a lot of criticisms of the AI are now bubbling up into steam reviews and the like, where they really didn't as much before (see Civ 5 reviews).
 
Military is still horrible, I ran over two complete Civs in the BC era with 5 archers only loosing a warrior to a chariot in a bad dice roll. Ended up at 20+ cities by 1 AD, on Deity... even civ 5 AI would put up more of a fight.

More importantly AI city management is still a mess, farms everywhere, and always less districts than the cap allows... Holy Sites bloody everywhere, just one Campus built in over 15 cities I took. I understand that combat is hard to code, but econ management? That's just not acceptable at all. Civ 5 AI is dumb as a brick, but it can at least manage its economy.

Yes, they really should have a bigger variety of 'openers' for the AI. Ideally there would be some logic (i.e. a religious civ might go for an holy site first, otherwise only if they had a spot generated more than +2 in adjacency bonuses). But honestly, I think even if they had a variety of openers (i.e. these civs are military oriented and going for encampments as a district priority, these are scientific oriented and going for campuses as a priority, etc.), it would help.
 
Still waiting for a CQUI update so haven't tried the new stuff yet.

I don't ever expect the AI, any AI, to be good at tactics.

Though the broader game, also known as strategy and also of which the game is part of a genre (TBS), needs to be a thing the AI can do well. As in, undisturbed the AI shouldn't fall over anyways, which it does. On Prince difficulty I see the AI getting rekt by Barbarians and even worse sometimes an allied city state beats them down and razes a city before I can even get to it. That is a fundamental failure if the AI can't even fight stupid NPCs....

Strategy is something like reaching a certain tech so you will be able to field stronger counter units and upgrade the ones you have to defend against an attack. Or deciding, "I think i can take this enemy, I'll declare war", or "Maybe not, this would lose everything" Tactics is like putting your melee units in front of archers and while that sounds simple in human terms, I guess it can be difficult to set hard and fast rules for that while you certainly can for strategy and build orders.

Basically, every vs AI strategy game I've played, the AI usually manages to match my numbers or outnumber me. Because that's easy to do. Now they might suffer a very bad KDR but that's fine; that's why the AI gets bonuses and stuff. And to me that's more acceptable than AIs who has already lost before the battle starts.
 
Yes, there are definitely things that could be done to improve the AI's unit handling to some degree, but I don't think comprehensive tactical AI approaches have really evolved much beyond "look at all possible options of you and your opponent and then choose the best one", which would be computer-nuking for something like Civ. Though one has to assume that there is some progress on this being made - it's essentially what an autonomous vehicle has to do in many respects, so maybe that will trickle down eventually.

Strategy should be much more feasible with an 'algorithmic' approach, and just straight up decision trees, and the AI should be able to operate with those much better - assuming the right ones are found.
 
So it seems the AI can't manage cities, can't defend cities, can't use air units or naval units, etc. What will get 2Ks attention and get this fixed?
 
Same here. Tested the new patch and already waiting for the next one. Looks like releasing DLCs is much more important than coding a decent AI...
 
For me, the frustration stems from the excuse that the AI has a hard time with 1 upt. Well, we've been 1 upt for almost 7 years now....

That's not a real excuse since I can't recall Civ V's AI ever being this bad. It could target appropriately based on unit type and would preferentially focus on wounded units (at least unless you ran them back), it could fire when garrisoned, and it could execute an attack in force well. The latter may partly just be a function of the fact that the Civ V AI built many more units than in Civ VI, but that itself is an issue. But the Civ V AI also seemed better at choosing target cities (it wouldn't always beeline the capital and in fact would usually just attack the closest city) and identifying when it had sufficient force to attack. It wouldn't take account of the enemy units in the vicinity but would at least attack with an army which could credibly capture an undefended city. Of course it was much harder to take undefended cities in Civ V, so the AI had to send more units.

Civ V also never tried to attack ships with embarked units.

Honestly, I've not experienced this even before the patch. Okay, garrisoned units rarely made a sorte, but that was it. The cities always garrison. The city almost always takes a shot at my ranged attackers. And now even garrisoned melee will attack.

Its also been my experience that the ai loves (loved) conquering cities. Never had trouble doing it. But people say in their games the ai can't do it. Well, I've spent hundreds of hours liberating.

I don't use an ai mod.

They're much better at garrisoning than I recall, and often even with ranged units, but I attacked three cities and all exhibited the same behavioiur. The only time a unit fired from within a city was when the city had walls, which suggests to me there's a coding problem with garrisons in unwalled cities. And even then the unit would sortie out to fire as often as it would stay behind the walls and fire. Of course the AI is better at building walls than it was pre-Spring Patch, but when early war comes before the AI has Masonry that doesn't much matter.

Based on defeat notifications the AI has been fairly good at taking AI cities since the Spring Patch (and in my current game Kongo at least has Brussels and a former Indian city), but given that the AI can't defend I'm not sure how much that really says about its offensive capabilities beyond the fact that it's more aggressive than it was. vs. a human player that level of aggression, with too few units to back it up, is extremely ineffective.

Military is still horrible, I ran over two complete Civs in the BC era with 5 archers only loosing a warrior to a chariot in a bad dice roll. Ended up at 20+ cities by 1 AD, on Deity... even civ 5 AI would put up more of a fight.

More importantly AI city management is still a mess, farms everywhere, and always less districts than the cap allows... Holy Sites bloody everywhere, just one Campus built in over 15 cities I took. I understand that combat is hard to code, but econ management? That's just not acceptable at all. Civ 5 AI is dumb as a brick, but it can at least manage its economy.

I did actually lose units in my attacks, admittedly - most to barbarians. It's a sad comment on Civ VI that the barbarians are more of a military challenge than the AI civs.

It doesn't seem that anything's been done about spamming useless districts - the AI definitely places them better, but getting better adjacency bonuses doesn't solve the issue that it builds the wrong districts.

Still waiting for a CQUI update so haven't tried the new stuff yet.

I don't ever expect the AI, any AI, to be good at tactics.

Though the broader game, also known as strategy and also of which the game is part of a genre (TBS), needs to be a thing the AI can do well. As in, undisturbed the AI shouldn't fall over anyways, which it does. On Prince difficulty I see the AI getting rekt by Barbarians and even worse sometimes an allied city state beats them down and razes a city before I can even get to it. That is a fundamental failure if the AI can't even fight stupid NPCs....

Strategy is something like reaching a certain tech so you will be able to field stronger counter units and upgrade the ones you have to defend against an attack. Or deciding, "I think i can take this enemy, I'll declare war", or "Maybe not, this would lose everything" Tactics is like putting your melee units in front of archers and while that sounds simple in human terms, I guess it can be difficult to set hard and fast rules for that while you certainly can for strategy and build orders.

Basically, every vs AI strategy game I've played, the AI usually manages to match my numbers or outnumber me. Because that's easy to do. Now they might suffer a very bad KDR but that's fine; that's why the AI gets bonuses and stuff. And to me that's more acceptable than AIs who has already lost before the battle starts.

Prince seems too low a difficulty to get much idea of the AI's competitiveness. Bear in mind that Civ games have been getting progressively easier over at least the last couple of iterations - Civ V was about 1-2 levels easier than an equivalent difficulty in Civ IV, and Civ VI is at least a difficulty level below Civ V. So you're playing Warlord in Civ V terms and perhaps Settler in Civ IV terms.

Your criticisms still seem largely valid at higher levels, and on Emperor I came close to victory on my first playthrough using a random walk approach through the tech tree and mechanics with which I was unfamiliar, but if the AI isn't capable of keeping pace with you at any game stage you should increase the difficulty.
 
Last edited by a moderator:
Prince seems too low a difficulty to get much idea of the AI's competitiveness. Bear in mind that Civ games have been getting progressively easier over at least the last couple of iterations - Civ V was about 1-2 levels easier than an equivalent difficulty in Civ IV, and Civ VI is at least a difficulty level below Civ V. So you're playing Warlord in Civ V terms and perhaps Settler in Civ IV terms.

Your criticisms still seem largely valid at higher levels, and on Emperor I came close to victory on my first playthrough using a random walk approach through the tech tree and mechanics with which I was unfamiliar, but if the AI isn't capable of keeping pace with you at any game stage you should increase the difficulty.

Sure, but I think that's the problem in itself. Remember, I'm talking about no interaction in my part so my concern wasn't it keeping pace. This is the equivalent of needing bonuses simply to breathe. And given other people's anecdotes, well, higher difficulties show a similar problem.

And yea I should move up difficulties, though we will see when I want to actually put in some real effort. (Probably when CQUI gets updated)
 
Last edited:
I don't know, I've seen a major improvement since before the patch. Playing as Nubia on Immortal, Gilgy actually managed to steamroll my captured Kumasi right after he plowed through two other city states, and although I managed to turn the tide and liberate Zanzibar, by Jerusalem my classical army was getting ripped by gilgy's brand new knights and crossbowmen. I was overzealous based on previous dealings with the AI, but there definitely seems to be improvement. And in my other game I've noticed the AI using air units a little better.

It's still not where I'd like it, but it's definitely better. As mentioned, the AI picking the best target to fire upon is still an issue.

I don't really have much of a problem not exploiting the AI, though. Steamrolling cities gets boring.

On Prince difficulty I see the AI getting rekt by Barbarians and even worse sometimes an allied city state beats them down and razes a city before I can even get to it. That is a fundamental failure if the AI can't even fight stupid NPCs....
If a civ mobilizes an army against a target and ends up leaving their flank open, they deserve what's coming to them. Not like that never happened in reality.

And my guess would be that the AI would handle barbarians and city states much better on higher difficulties. Playing on emperor and above, the NPC civs definitely seem to push around the city states much more than on lower difficulties.
 
If a civ mobilizes an army against a target and ends up leaving their flank open, they deserve what's coming to them. Not like that never happened in reality.

That is what I would call stupid.

Just because they get bonuses to mask said stupidity, doesn't change that.

EDIT: Started an emperor game. This is a good example of something that would be a bad idea, independent of bonuses. Can you spot two things wrong with the AI already? And that DoF was just declared, so I doubt it was thinking that far ahead. (the DoFs are good though)

TxCdJ9u.jpg
 
Last edited:
Back
Top Bottom