Lets list AI stupid behaviors

Ninakoru

A deity on Emperor
Joined
Nov 14, 2002
Messages
658
Location
Madrid, Spain, Europe
Hi, I'm seriously considering making a dll mod that makes the AI more coherent and competitive, first I need to know what are the common silly moves and errors the AI make near every game, so with your aid lets make a list with all the AI bad decisions.

Here is an update on my progress, also I categorized all the issues.

Production AI:
---------------
- AI sometimes mass carriers, when they may use one or two, I often go on a naval assault to find lots of useless carriers near under their borders doing nothing.

- Some AI focuses on training too many units and then don't use them, going broke and killing their game.

- Some balance on unit types trained by AI: AI sometimes do many of one type of unit (Artillery), too little of another type (Melee).

- Bad use of Archeologists, the get too little works, they get some unused ones doing nothing.

- Possible overuse of prophets to convert cities instead of making holy site.

- Possible overuse of religion units on modern era and afterwards, when is better to save faith for GP purchase.

- Is posible AI doesn't even know what is an AA unit and use them as standard melee.

- Later in the game, AI doesn't use the best units like the XCOM squad.

- Bad use of great people: Using musician concert mid-game, bulbing GSs early.

- On higher diffs AI still spawn too much units hurting them more than helping due to the science/culture penalties.

- AI seems to never want to upgrade some of their recon ships. Like still having triremes in the atomic era.

- Sometimes AI with high GPT is unable to spend the gold for no reason, reaching 6k, 8k gold with no use even if the AI is on a critical moment.

Combat AI
----------

- When AI attack a coastal city, or any battle happens near water tiles, they have no worries to throw lots of units to the water, only to lose them by ranged units.

- Air units tend to rebase a lot instead on attacking.

- Air units very rarely use the sweep and intercept options.

- When you capture a city or you have a city in range damaged, they tend to bombard the city even if there's no chance to later take the city.

- Ranged units moving OR firing when they can do both. FIXED: Now AI can move to target AND shoot as part of a multiple attack. I still have to potimize code and check out the previous analysis of the combat outcome properly.

- Ranged don't attack if isn't a city take or a focus-kill, they should always fire if the can get a shot to any target, as they don't receive damage doing so.

- Possible problem with units with no defense from terrain promotion going prioritizing moving into a hill to get the defense bonus.

- AI sometimes left the Generals unprotected on the field, they sould have an scort near always.

- Sometimes AI abuse of sitting and healing when they could instead easily wipe the barbarian ranged attacker / clear a camp, and instead dies.

- Embarking Great people without naval escorts or protection.

- AI sends GP/missioners trough territory of opponent/CS who is at war.

- Fail on priorities when AI choose to focus on targets. Always go for the weakest one, while could try to shut down a nastier one.

- AI should be able to detect when they can stomp roll over some threat (IE it have a much much bigger army strength), and instead of focusing, try to max out damage done to the enemy lines.

- AI doesn't have "memory" of past actions on the field, thus not being able to see planes out of line of sight, submarines, and isn't able to detect where is the hottest war zone to react accordignly.

- AI still lose too much settlers due to barbs.

Diplomacy AI
-------------

- AI often ask for resources at 0 value for them as part of a trade.

- AI acepting a city deal only to raze it afterwards.

- There no diplo effect on war/peace bribe.

- Asking the AI to not do something should have more pressure in them with a diplo effects associated, as it is now, is just a waste if it's against their willingness.

Global AI
----------

- Few AI Grand Strategies, one per victory type, no combination like culture wide, culture tall, cuture, war.

- Seems AI always get the same promotions based on unit type: Melee go Drill + Medic, Calvary go Shock, Ranged ships go Bombardment, Carriers go Armor plating, always. - FIXED: Now AI will choose promotions more randomly, still prioritize on higher tier promotions.

- Naturals wonders may have not enough priority for the culture governor of the city.

- Not clearing the marsh on calendar luxury tiles.

- Building roads between 2 or 3 pop cities.

- AI Strongy desire to go culture the entire game, with kills diversity.

- AI will choose order as Ideology most of the time.

- End game AI Tech path seems to be streamline.
 
Building too many artillery units, without enough units to put in front of them to shield them. My infantry kills them without my artillery needing to shoot at anything other than the enemy city.
 
their range units cannot move a space AND attack, which is crazy easy to beat in battle. they have a really hard time understanding that just because there are hills, it doesn't mean going into the water and around them in war time is a better plan. lots of UU's have issues like China's version of a Catapult that refuses to fire more than once.

heavy focus on getting once of each unit type it seems, and leaves them about 1-2 units short in terms of melee. I just won a game where the AI had me BEAT. to the point where I almost just quit before they put me away, yet I was able to with 2 units and NO WALLS, keep away a ridiculously large army and rush thanks to Germany having only 3 ground units and not 4. I feel like with melee units being so key, and the AI being so bad at understanding how to protect them until ready to take the city, that they need MORE melee units, less range.
 
They often move their units instead of firing.

The AI often leaves units near the back of their empire, even when their city is under siege. I often see the computers just have units standing back not engaging me, even when that unit could save or prolong the life of one of their cities. The AI masses up decently well for a first strike, but fails at figuring out what to do with their troops after that, and can't respond well to counter-aggression.

City states constantly make their ranged units leave the city. A city state should ALWAYS have at least one ranged unit in their city when they are at war, no matter what the situation on the ground is. Since city states should play pretty defensively anyway, this seems like a relatively easy thing to code which would make the game a lot better.

AIs don't always adjust what they're building when they're under siege. They do rush buy walls at times I think, but they're often building settlers, wonders, and buildings on cities that are under attack and about to be taken.
 
Denouncing or DoWing someone who is two times as powerful or more.
 
Denouncing or DoWing someone who is two times as powerful or more.

I'm not sure that one is valid. The AI usually only does this when they have an ally who declares war with them, and some civs are just really bold and willing to take risks. Alexander may be an idiot sometimes, but I'm not sure that this is the type of idiotic behavior I want to take out of the game.

Also, denouncing a powerful civilization is not a bad move in the slightest. I do it all the time if I think it will earn me points among other civs.

That said, changing the combat strength calculations to account a little more for ranged units and upgraded units would be a nice change, since right now the computer isn't very good at calculating true army strength.
 
This is good and may help hugely in our dream of a BetterAI mod. Nikanoru, can I task you to be the "official" organizer of this list? I could add you to the BetterAI team list if you want...

What we would need is for you to keep an organized, prioritized list from what other forumites and yourself are adding... we are aiming at going from the most specific (and high priority) to the most general in terms of improving the AI. We should probably follow the AI levels as coded, that is, attack the TacticalAi first and then go higher. The list should then be ordered in at least 3 levels, starting with Tactical bluffs (for example, units to swim!), then Strategic blunders, and then higher level blunders (victory pursuit, etc).

Can you do that?
 
Archeologist spam. Seriously, I've invaded AI capitals where they have like 10 Archaeologists lying around not doing anything. At least get the AI to use them instead of just being a waste of production.
 
This is good and may help hugely in our dream of a BetterAI mod. Nikanoru, can I task you to be the "official" organizer of this list? I could add you to the BetterAI team list if you want...

What we would need is for you to keep an organized, prioritized list from what other forumites and yourself are adding... we are aiming at going from the most specific (and high priority) to the most general in terms of improving the AI. We should probably follow the AI levels as coded, that is, attack the TacticalAi first and then go higher. The list should then be ordered in at least 3 levels, starting with Tactical bluffs (for example, units to swim!), then Strategic blunders, and then higher level blunders (victory pursuit, etc).

Can you do that?

Sure, that would be great for sure, I was thinking on doing the mod all on my own but any type of support would be greatly appreciated. I'm taking my own steps ATM:

- Compiled the BNW core dll, modified some debug logs and added to a mod project, built solution and looked into those logs with my custom code in an automated game: WORKED. So there's no problem to prepare a mod using the core dll from my side.

- I'm an experienced programmer but I'm more into Java and C# / VB .NET languages, however I know a little of C++ and shouldn't be an issue if I don't mess too much with the current dll architecture.

- For now I'm trying to identify where are the problems and how to solve them without entering into coding details, as example I have this kind of notes:

"Too many carrier ships on AI naval fleet -> AI checks all their carrier cargo capacity vs number of fighters type planes. If it's lower, makes carriers. This goes worse as they don't use the Flight deck promotion. CvMilitaryAI.IsTestStrategy_NeedAirCarriers. A possible solution is to also check number of combat ships vs number of carriers the AI has, and don't go carriers if the proportion is less than 80% vs 20%."

Adding this to the list:

- Seems AI always get the same promotions based on unit type: Melee go Drill + Medic, Calvary go Shock, Ranged ships go Bombardment, Carriers go Armor plating, always.

- Bad use of Archeologists, the get too little works, they get some unused ones doing nothing: My guess is that seems they stop doing archeologists at some point, but as they only go to neutral land or their land archeologic ruins, so if they lost all the coded valid targets they get their archeologists sitting on a city doing nothing.

- Few AI Grand Strategies, one per victory type: They should have more variety, as example: Culture going turtle-tall, Culture going expansive peacefully and abusing wide for religion and land grab for later ruins, Culture going warmonger and stealing cities for the extra culture and works...
 
Unability to use all those aircraft missions. On the other hand, I have played this game for 1000+ hours in singleplayer, multiplayer, scenarios and immortal/deity and I still have no clue what's going on with all those interceptions and so on, aircraft for me is BUILT BOMBERS -> BOMB CITIES and I never had problems with aircraft so I have no motivation to use more advanced tactics :D


Oh, and one important AI problem - tendency towards "dancing" in front of cities. You know, changing formations and moving and changing formations and moving for a couple of turns, especially before the declaration of war :p
I have seen this problem also in case of AI vs AI war - Chinese troops regrouping for 10 turns while being few tiles from easy to capture Mongolian capital (although when they finally attacked they took it in 2 turns :D )
 
How much for 1 coal? That'll be 1 gpt, 20 gold, 5 aluminium, 5 iron, 5 horses please.
How about 1 gpt? Okay, sure.
 
Don't think this was mentioned, but if it was, oops!
-Bombing a city when it is at zero HP (i.e. Berlin is at 0 HP and Shaka has 5 bombers left to use, instead of bombing units or other cities he continues to bomb the 0 HP city and then take it after all the times he can attack it). Goes with all cannon and bomber units.

-Not attacking twice (i.e. Chu-Ko-Nu (I think that's the right way to spell it) only attacks once even though it has the ability to attack twice
 
-Not attacking twice (i.e. Chu-Ko-Nu (I think that's the right way to spell it) only attacks once even though it has the ability to attack twice

This is just another version of the 'can't move AND attack with ranged units' problem, ranged units only perform one action per turn as far as I can tell.
 
Also on air power, the AI makes poor targeting decisions - it's better since BNW, but it will still send bombers to attack cities when it has no units in position to capture them, often losing planes repeatedly to AA fire and defences. It routinely does the same with nukes, hitting targets a long way from the front line.

It rarely uses fighters on intercept (or, as you note, air sweep), and uses them against ground/sea units too frequently.

It over-prioritises building mobile SAMs, and uses them as front-line combat units. The AI should be programmed to assign high priority to anti-air only if a nearby non-friendly city has aircraft based there, and to use them mainly defensively.

It appears not to register AA attacks as a threat, so will use planes against units even when AA weapons are nearby.

AIs will use all their ranged units to shoot cities, even if the city is already at 0 health and there are enemy units nearby.

It is still far too common for an AI to mass ranged to attack cities and have no melee in the vicinity - sometimes it will even place ranged units all around the city, blocking access for melee units behind.

In the early game it uses too few siege weapons - rarely more than two in an attack, and one alone is very vulnerable to city garrisons.

The AI still consistently uses melee units to launch embarked attacks even when there are available tiles to disembark onto where they would be better-protected and more effective.

The AI will use embarked attackers one at a time, usually without a naval vessel protecting them, and so consistently loses them to defensive fire.

Not sure if this is specifically programmable as it affects only one unit, but the Hun AI will use battering rams as defenders and to protect settlers/workers, as well as moving them preferentially through rough terrain when attacking a city (which is good generally with other units for the defence bonus, but fatal for rams).

Bombing a city when it is at zero HP (i.e. Berlin is at 0 HP and Shaka has 5 bombers left to use, instead of bombing units or other cities he continues to bomb the 0 HP city and then take it after all the times he can attack it). Goes with all cannon and bomber units.

I've noted the general version of this phenomenon above - it will do it with any ranged unit, from archers onwards.

- Few AI Grand Strategies, one per victory type, no combination like culture wide, culture tall, cuture, war

The really noticeable lack here is the absence of any clear AI domination strategy - play Civ IV and you'll see an aggressive AI take on and start to gobble up neighbours based on their relative strength, taking out the weakest, then the next weakest, then... I've seen similar behaviour maybe twice in Civ V over almost 1,000 hours, and in one of those the strategy just stopped after the first two conquered civs despite Russia being on the aggressor's (Shaka's) landmass and still alive. The other was way back in vanilla, suggesting the behaviour may have been largely patched out. The AI plays domination (when it does go for it) as complete elimination - it doesn't seem able to recognise that it only needs certain key cities (i.e. capitals) - in that game with the Zulu, while Shaka went to war with Catherine eventually (at my urging), he never made any move for Moscow even after the Russians had ceased to be any kind of threat.

At the same time, the AI is almost equally bad at any kind of targeted war beyond "I want the city that's closest to mine"; it lacks any ability to evaluate the value of specific target cities it may be able to take, even if they aren't the closest.

Archeologist spam. Seriously, I've invaded AI capitals where they have like 10 Archaeologists lying around not doing anything. At least get the AI to use them instead of just being a waste of production.

This was absolutely everywhere when BNW was released, but in my games it's almost nonexistent since the patch.

It is however still very common with missionaries and prophets; now the AI is more reluctant to use these against civs that have their own religion, but this doesn't appear to have modified its missionary production behaviour, so I'm seeing a lot of missionaries just wandering around my and other civs' territories and often dying of attrition.

On the subject of civilians and wasted resources though ... barbarian worker/settler/missionary capture. The AI is better at protecting settlers and workers than once it was, but will very rarely protect other civilians (archaeologists, missionaries, or Great People of any kind), and if a settler, worker or other civilian is captured the AI 'forgets' it. That's production you want back - players can be careful but will still lose these civilians occasionally, but the player won't just say "oh well" and build another one - they'll actively target the camp and bring the hostages out. The AI will not do the same, which both wastes its resources and gives free civilians to the player who liberates them.

aircraft for me is BUILT BOMBERS -> BOMB CITIES and I never had problems with aircraft so I have no motivation to use more advanced tactics

Yes, but you have the advantage that the AI isn't terribly good at anti-air, while a player can be. The AI is successful at AA more by luck than judgment: it just happens to favour AA guns and SAMs at a particular point in the game. It doesn't actually actively use these as AA weapons. And while it now builds more or less enough fighters, as above it's much more inclined to use them as attackers than interceptors.

Oh, and one important AI problem - tendency towards "dancing" in front of cities. You know, changing formations and moving and changing formations and moving for a couple of turns, especially before the declaration of war
I have seen this problem also in case of AI vs AI war - Chinese troops regrouping for 10 turns while being few tiles from easy to capture Mongolian capital (although when they finally attacked they took it in 2 turns )

The AI plays by the same rules in combat whether it's fighting an AI or a human - I had one game where I was spying on a capital (I forget which) that Darius was attacking and he did a lot of this, and also a lot of bazooka-spam when the city was at 0 health anyway. And my observations about AA guns are based on watching AI vs. AI since I don't build them, only fighters.
 
[there was an error posting your message. your message cannot 938984762331133 characters long]
 
- mass missionary spam to convert your cities, even when their religion gains nothing by converting you (e.g. beliefs are pagodas and cathedrAps and would always be better off buying a building)

- using prophets to convert your cities, when they would benefit more from a holy site with piety beliefs (see above)

-post-industrial missionary spam, when the faith is better spent on great persons.

-overall, too much focus on piety and religion to the detriment of other things.
 
One of the most annoying AI things for me is when they go to war, but don't even try to attack the opposition.

It's the same when they persuade you to attack X enemy together, but then never make any proper attack towards them.
 
Top Bottom