Brainstorming: Weaknesses of the AI

[Kael]

Really? I don't see how it ruins the immersion at all - it's perfectly realistic even in fantasy settings. You see examples of it everywhere. In the Warcraft 3 campaign, humans and orcs, which are bitter enemies of each other, unite against the Burning Legion. In LoTR, elves and dwarves dissolve their distrust for each other to stand against Sauron.

I don't see the distinction betwen "not playing to win" and "just being opponents". It's all a matter of challenge, which besides the immersion is what I play game for. And as I've just pointed out above, it's perfectly immersive.

Perhaps you're misunderstanding what I mean when I say "gang up on you". I don't mean that once you reach some critical threshold they'll all start attacking you. Now that's unrealistic and just annoying. Instead they'll be far more hesitant to trade techs, start forming closer relations with others instead of you, and in general start distrusting you. Naturally, you can just do what the US does today - no matter how much the world hates the US, the US holds them by the power of trade and economy. So a nation that focuses on trade and diplomacy can avoid other nations from forming an alliance against it.

I personally want AIs to be as human-like as possible at the highest difficulties, within the bounds of the lore of course. It's much more satisfying defeating a human-like opponent than having contrived handicaps to compensate for inferior AI.

The adjustments for this already exist, they are iWorseRankDifferenceAttitudeChange and iBetterRankDifferenceAttitudeChange. The are set per leader and can be adjusted with the editor. In general good leaders dont adjust their attitude based on rank, netureal leader sdo slightly (dislaking people more poerful than them) and evil leaders do greatly (and are therefor more likely to jump on high ranked players). The adjustments are very slight but you may be able to tune to your prefered level of resistence by increasing the values.

 

[Maian]

How about making it switchabe with the default beeing off. Kind of an option like Aggressive AI?

That would work for me

 

[Sureshot]

I think for the "all gang up on dominator" strategy to work requires more action than simple dislike. Even when many people dislike you, none of them will declare war when they know individually they are incapable of challenging the dominator (which makes it fairly easy to broker peace with many of them).

The thing that needs to be done for such a strategy to work is a shift in the thinking of the non-dominators as a whole. Like Maians examples, the Orcs and Humans Unite against the legion, and the Humans, Elves and Dwarves Unite against Sauron, all despite ancient grudges or hatreds, because of the greater threat.
There needs to be Uniting, maybe force/allow a permanent alliance of all non-dominators?

Like, upon one Civ reaching critical dominion (maybe 50% higher score/whatnot than next highest rival, making them the Dominator), the next highest rival (or the greatest enemy of the dominator) has the option to do a Rally against the dominator (maybe requiring using a great leader, or building a wonder, whatnot).
This could create a temporary alliance lasting X turns or until threat is reduced to not being the top (whichever is longest). Maybe have joining the alliance variable (popup window for human players), dependent on whether that civs relationship with the Dominator and the Rallier.

 

[Xyshi]

Some obvious problems with the AI include the fact that they dont understand the abilities of spellcasters. This will take a lot of work and frankly i dont envy you.

Also, the AI doesnt ever seem to try and build better units. I almost always see clone armies of assassins and mercenaries running around (even if they have resources to build better units). Also I'm pretty sure the AI doesnt really understand that it gets a cool archmage if it levels its mage....

 

[Bad Player]

In my game as the Ljosafar I was getting my ass kicked badly (barbs at start then Vampire nation attacked me). However then I built a couple of Elven assassins - they wiped the floor with Moroi blood! The AI problem is that they just kept attacking me with Moroi hordes but I kept beating them with a handful of assassins (and the world wonder Elf guy)! My assassins got very powerful of course and I burnt nearly all their cities to the ground (just left them their capital in exchange for all their techs). The AI also made hunters but that didn't help them much. Therefore the problem seems to be the AI is not responsive to the type of units I'm using. They should realise they need to build units that can combat my assassins.

 

[Chandrasekhar]

Yes, I believe that the AIs' biggest problem (in this more than in vanilla Civ IV) is that they don't play to win. It's moreso a problem here because they don't have any Space Race to fall back to. I'd say that, since conquest is really the only way for a person to get a victory, the AIs should be much, much more aggressive and have some sort of script to follow to gang up on people.

They shouldn't only gang up on strong people, as I know that many players have a bad habbit of just tech'ing up in the early and mid game and only building a military to go conquesting. The AI should definitely war in the mid game more often.

 

[Chalid]

Concerning the playing to win...

I have some (i think) good idears what has be done to archive this, but in FfH we want the AIs not fixed as enemies, but as other reaces living on the same words, so to speak some as frieds some as enemies. But if we get one time to the Point where we can declare FfH as nearly finished i will implement those ideas in combination with a Game_Setting to switch it on/off. I hope you won't complain then if you loose every game and have to switch back to Settler

 

[eerr]

i usually play so that the game reccomends techs to me, but many times it reccomends techs i have no need for (fishing in a civ with no costal or freshwater cities, the tech deception which doesn't actually grant anything yet...)

with so many choices, the ai needs to be able to make descisions-like choosing farming and mining early on if they have hills and corn in their radius.

 

[Chalid]

They check for some (not all) of those things as well, the AI_value of fishing for example depends on the number of costal cities plus some other aspects.
the AI also considers techs farther down the tree as next goal (whereas the proposal for the player only chooses between the available techs).

And the question is of course when does it propose this tech. Right at the beginning, or did you already have most of the tier 2 techs?

It could be needed to adjust some of the tech Values through (right now all tier 1 techs have the same price, all tier 2 techs as well and so on. So it might be needed to do some rebalancing there (for the players good as well as for the AIs))

 

[Chandrasekhar]

Concerning the playing to win...

I have some (i think) good idears what has be done to archive this, but in FfH we want the AIs not fixed as enemies, but as other reaces living on the same words, so to speak some as frieds some as enemies. But if we get one time to the Point where we can declare FfH as nearly finished i will implement those ideas in combination with a Game_Setting to switch it on/off. I hope you won't complain then if you loose every game and have to switch back to Settler

Am I to take that as a challenge?

You know, vanilla Civ IV is supposed to be the same way, which is the whole reason for Diplomatic victory. Meh, maybe I'll just have to get into some multiplayer games and see how that runs...

 

[Bad Player]

I was playing a game as Bannor and one of the good civs (who was human) built 2 or 3 earth mana nodes (more than one of the same type is useless currently). I saw 2 of their earth mana nodes but they had 3 earth mana (I hadn't seen all of their territory).

I would think it would be relatively easy to make ai only build their preference mana for the 1st one and then work out (or even random) which second mana type they want. When multiple mana of the same type becomes useful it might be good but the ai would need to calculate whether they can still use multiple mana of the same type.

 

[Chalid]

The code is already in but commented out at the moment. it'll be activated in 0.15 probably.

 

[snarko]

I was watching the AI and Basium started building Basium right away. Built a worker when the city reached size 2 but then it was back to building Basium. By the time he was built they had lost everything the worker built but one piece of road to barbarians. As a result of spending so much time building him they were behind the other civs in every other way.

Also the AI doesn't know how to defend agains barbarians (or even that it needs to do so). They will send out units to explore and keep other units home to defend the cities, but they have no units to defend the border or improvements... resulting in many lost improvements and workers sitting idle in cities waiting for the nearby barbarian to be killed.

When it comes to warring I see two major flaws in what the AI does. It leaves too many defenders in their cities, allowing the enemy to come to them, and it doesn't bombard city defense enough. This leaves them with too few units attacking well defended cities having high defensive bonus' from culture... no wonder the AI have trouble taking cities middle/late game. Early they are much better at it, before culture defense and number of defenders are too high for them to deal with. I *think* FfH is better than vanilla here, problably thanks to heroes, but there's still a long way to go before the AI is as good as a human is with this.

 

[Sarisin]

Also the AI doesn't know how to defend agains barbarians (or even that it needs to do so). They will send out units to explore and keep other units home to defend the cities, but they have no units to defend the border or improvements... resulting in many lost improvements and workers sitting idle in cities waiting for the nearby barbarian to be killed.

Excellent point, snarko!

I think this affects you in your game, especially if you are playing raging barbs. What happens is the the AI civs are gradually wiped out until there are just a few left. If you have the misfortune to be with AI civs with the BAR trait, then you are in deep kimchi. This is because all of the barbs on the map come after you. Stacks of double digit barbs are very common.

Strengthening the AI of civs against the barbs will help spread out the barbs more evenly (unless the AI cheats by going after the human player with more barbs ) and make it easier to survive and let your civ grow.

Now, I am pinned down in one city fighting like mad just to defend my improvements - forget about settling another city. Also, the barbs have conveniently set up cities along my borders.

 

[Chalid]

Good points but really difficult. The AI does at the moment not defend any Improvements, abd that usually with good reason. If the Ai will be allowed to defend the Borders and such it might spred out its units to far. That is good agains Barbariand that come in small number, but if it has spread out its units to defend its improvement when the player comes with a big stack they are easy to pick up and kill.

So deciding when to allow the Ai to defend Boarders and when not is not an easy task.

 

[snarko]

I let the AI play 1000 turns while I slept and here's some observations for when I came back (large map, 17 ai civs, pangea, raging barbarians).

Barbarians won't pass through the territory of someone with barb trait to get to someone they can attack. This late in the game there are few plots which are still neutral and a whole pack of barbarians stuck there because the land is surrounded by an AI with barb trait and ice.

Inflation hurt the AI late game. They are paying more for inflation than everything else combined. I guess it would be the same for a human, but a human would try to do something about it (ie win) instead of sit there waiting for it to increase even more, slowly being forced to go closer to 100% tax...
Also all the AIs are paying more for their military than their cities, but they don't use the military for anything.

The AI have run out of things to build, including units (they don't want more). With the exception of one civ (there are 9 left alive) they are building wealth in most of their cities. Some of these civs should easily be able to conquer one of their neighbours, but they don't. Not even if they don't like them. It's not because of defensive pacts either, there's only one such pact in the game. They're all pretty much doing nothing...


The kuriotates doesn't know that it should grow it's cities large. They have 40+ happiness but their cities are only size 9, 9 and 23. 23 because it's near a fish resource and three ancient forests on grassland. Using serfdom doesn't help them with this either, but they would not get the hammers back from switching to something else, so the main problem is not building more farms.

 

[Kael]

I let the AI play 1000 turns while I slept and here's some observations for when I came back (large map, 17 ai civs, pangea, raging barbarians).

Barbarians won't pass through the territory of someone with barb trait to get to someone they can attack. This late in the game there are few plots which are still neutral and a whole pack of barbarians stuck there because the land is surrounded by an AI with barb trait and ice.

This is changed in 0.14.

Inflation hurt the AI late game. They are paying more for inflation than everything else combined. I guess it would be the same for a human, but a human would try to do something about it (ie win) instead of sit there waiting for it to increase even more, slowly being forced to go closer to 100% tax...
Also all the AIs are paying more for their military than their cities, but they don't use the military for anything.

The AI have run out of things to build, including units (they don't want more). With the exception of one civ (there are 9 left alive) they are building wealth in most of their cities. Some of these civs should easily be able to conquer one of their neighbours, but they don't. Not even if they don't like them. It's not because of defensive pacts either, there's only one such pact in the game. They're all pretty much doing nothing...

I suspect this is vanilla behavior. But regardless I think we need to define our own process to help some wars happen. It will probably occur in "Fire" as we start to deal more closely with the endgame.

The kuriotates doesn't know that it should grow it's cities large. They have 40+ happiness but their cities are only size 9, 9 and 23. 23 because it's near a fish resource and three ancient forests on grassland. Using serfdom doesn't help them with this either, but they would not get the hammers back from switching to something else, so the main problem is not building more farms.

Thats a great point, I'll boost Cardiths chances of building farms in 0.14.

 

[Sureshot]

I've been playing a 3 vs. 3 team game of neutral/goods in automatic constant war with evils, and I've been noticing that the AI's on my team don't know how to deal with the raging barbs or the enemy teams (the enemies were doing pretty well since one had the barb trait and it was on deity which i think handicapps human players teams).

One of my allies (Amurites) was nearly wiped out and only had 1 city left, and they were only saved because I managed to get some werewolves to them and take over two cities and gave them them. With these werewolves she soon scoured the countryside and managed to stave off the barbs (less fog of war with so many units patrolling) and expand quickly.

My other ally (Bannor) did alright against the raging barbs due to sea "walls" that made them less accessible to most barbarians (lizardmen got through but thats still manageable), and by chance they built a city or two on hills (cities built on hills are a great early advantage for archers).
But, once they expanded they started making weak defense cities and stretched themselves too thin (most cities had 1 or 2 units) and got bowled over by the other team.
I had to go in and build cities then hand them over to them so they had defensible cities, plus I gave them werewolves.

I think their strategy is good for non raging barbarians, but given how much more difficult raging barbs is in FfH compared to Vanilla civ, I think they need to understand its "red alert" when raging barbs is on (and also when permanent war is on, as there are some powerful offensive civ units) and that their building, scouting, and improving need changes that reflect this.

During these "red alert" scenarios (or even temporary wars) I figure some guidelines are necessary:
-only build in highly defensible areas (on hills and tiles with high concentrations of impassible tiles as adjacent tiles, with greater weight on directly above or beside as opposed to diagonal)
-send out settlers in moderately calm spells
-send out stronger entourages with settlers
-recon units remaining near borders (not going out far, which usually just gets them killed; instead adding a buffer around borders in which barbs can't appear)
-possible entourages with recon units and ending turns in defensible tiles (like forests or hill forests, and tiles with lots of impassible tiles nearby)
Most importantly, extreme production of defensive units. In a previous game the Lanun were capable of making archers but kept outputting tons of adepts, and the hippus (also capable of making archers) were making warriors, while both were losing cities every turn (they kept tryin to make new ones often too).

I'm not sure if the AI for all that can be changed, but currently AI civs seem incapable of managing defense with raging barbarians or constant war (both sides in my current game wipe out each others cities regularly - but its constant war and raging barbs, but I've managed to keep all mine with no reloading... lost all my heroes though lol).

 

[Bad Player]

Good points - building on defensive tiles and having stronger entourages seems fairly easy to code surely?

 

[Chalid]

Nope not as easy as it sounds. The problem is it would need differents AI and switch them dependend of the game setting. As you now know that Ai can't handle those settings you'll have to avoid them either or live with the not so good AI.

The main focus on teh AI will be laid at the "normal" settings for the time beeing. If that Ai works in a acceptable way one can think about specially adapted AIs.