AI 'handicap' on border forts goes too far.

[Chaingun]

If you're OK with modding (no ironman), change defines.lua:

FORT_MAINTENANCE_CHEAT = 1, -- Set to 0 to disable AI fort maintenance cheating. Warning: Will make AI suck.

 

[Chaingun]

If you are the AI guy now, could you update the FAQ thread on AI cheats when you have the time?

https://forum.paradoxplaza.com/forum/index.php?threads/ai-cheats-facts-and-misunderstandings.713930/

Yes, I'm pretty sure that's fairly out of date.

I'd even find it more comfortable if the AI got an insta-garrison on war declaration than for them to get zero-maintenance forts everywhere.

If the problem they are trying to solve is the AI getting forts ninja'd at the start of wars, that fixes that WITHOUT giving them free army tradition AND extra money.

It's both the tactical problem, and the fact AI easily bogs down its budget in forts. The player can majorly outclass the AI by changing strategy in response to the static choice made by the AI and adjusted for the circumstances. Even after fixing AI loan repayment last patch AI was still easily getting caught in fort expense spirals. Although Wiz suggested a full cheat, the "border only" cheat as seen here was created to let small AI countries have comparatively larger benefits than large AI countries (with a greater proportion of internal provinces). Sea counted as a border since the human player might rather easily perform a naval invasion. Ottomans do have a lot coast, unfortunately, leading to the result seen by Arumba.

I've removed it for wastelands in 1.16 as that obviously wasn't intended.

 

[Chaingun]

Wouldn't the general consistency at which the AI gets bogged down due to fort expenses be a sign that the fort expenses are too high, though? Players are deleting them in favor of spending that money on more useful things, and the AI is literally dying due to it, so that's generally indicative of such a problem, I'd imagine.

Yes, Johan/Wiz could certainly change the design. IDD that fort maintenance feels rather risky to me and it's difficult to not overdo it in MP. On the other hand, the difficulty can be interpreted as depth in gameplay. Nonetheless, if you're going to have any luck with design changes, Johan is the person you need to convince.

I'm fairly sure it would be impossible to stage a naval invasion of a coastal fort before a monthly tick could recover some garrison.

Sure, it would still be severely understrength though with associated implications. The current solution may be too generous, but it was quick without tacking on additional exceptions. It's especially nice in AI code for the AI to be able to ignore maintenance of a specific fort altogether as it makes the problem vastly more simple.

Giving the AI freebies on an over-priced stalling mechanic seems like the last place you'd want to give the AI freebies to make the game more entertaining. If you want the game to be more challenging/engaging, extra siege times is probably not the best path.

Again, this is more of a game design question. I basically share your sentiment, but then again I've always been the conqueror type playing the game (EU2+) at almost perpetual war. Late game fort balance may be an entirely different beast, referring to the multiple threads on the excitement of sieging a level 8 fort.

Wouldn't it make more sense to set a maximum amount of money the AI is allowed to spend on fort maitenence (10-20% of total expenses) and have them auto-disband forts if they go over that limit? Or just give them an automatic -20% reduced fort maitenence?

I could write a long explanation but you will have to trust me when I say simple solutions have been tried. (The current AI is also indeed simple but a fair bit more complex than what you describe here.) For a start, consider that country income is not a fixed number but often varies from month to month.

 

[Chaingun]

Yes, it is obvious that I as an AI programmer have no interest in improving the AI whatsoever.

Now, if there were 10 more AI programmers on EU4 we could indeed speed things up by a factor 2-3 or so.

 

[Chaingun]

It's great to see so many design suggestions in response to the posts I made. However, my capability to see these things through from idea stage to implementation is realistically rather limited, both because of time constraints and my mandate. I've updated my forum signature to remind everyone of this.

I haven't really seen many of the devs replying on the EU4 forums recently (barring, obviously, Chaingun), so I do have to wonder how often they read nowadays.

Pretty sure Johan and most QAs read at least once a day, but because discussions are a potential time sink they are sometimes better avoided (I am writing this at 3.45 AM).

It's a problem with AI in general, and how it fits into the game development cycle.

Work on the AI calculations on any given feature can't really start until the feature is complete, and by then the company developing the game wants it to be released (it's ready and functional, companies are about business and making money, let's release).

This happens to all and every game developer out there. The time to improve AI would be once the game is complete and not being developed anymore - but by that time the companies behind the game cut funding and start focusing other projects.

Even ongoing games like EUIV, which are lucky enough to have a company behind like PDX who really cares about the product, fall into this trap.

You're certainly right. I should add that as a dev it's a challenge to keep up with the new features being added and understanding them in time. Even features that I implemented myself, because it's unclear how much they impact the 'efficient' way to play the game, until after release when I can read players' AARs and discussions on strategies.

 

[Chaingun]

I worked quite extensively with making the AI follow missions a bit better and similar things for the 1.15 patch (there were Christmas holidays in-between which helped). I wouldn't blame it all on forts.

To be clear, my focus has now shifted to de-blobbing mechanisms, but there are yet no holidays in sight.

 

[Chaingun]

Side thought, Chaingun you mentioned you're trying to make the AI play the game well as it currently is, could we perhaps consider teaching the AI to please god stop hiring mercenary cavalry and cannons? I think even the newest player in EU4 quickly realizes how atrocious the cost-benefit is on them, yet I constantly see the AI raising mercs in dangerous locations, I go to kill them and its a solo merc cannon. I mean.. why? Just say no! You can only hire merc infantry. You hire them and you use your manpower on your other units.

Sure, low hanging fruit like this is always welcome. Can't really recall why it hasn't been addressed (maybe the solution is more complicated than it would appear) but it really shouldn't be at first glance. My guess is that AI is somehow needing to maintain unit balance because it doesn't understand it can compensate via regular regiments, or it's simply to be compatible with mods (but one can leave a define for that).

 

[Chaingun]

Don't worry, this kind of thing passes through QA first. I am aware that how human plays isn't always necessarily the best for AI.

No, I am not at all an expert at EU4 which I've played little in comparison to EU2/3. Hence I'm not in charge of design.

 

[Chaingun]

Yeah, I wasn't saying it was realistic. I'm fully aware that there are dreams that cannot be but damnit I can still hope eventually

Actually had to move some of the Condottieri AI to Hard after QA got destroyed by it in their MP campaign. So Normal will definitely not have the hardest AI the way it looks now. (We're also looking at adding Very Hard as mentioned earlier which will be a pure modifier boosted AI in most cases, but unsure of whether it will get into 1.16 or the followup patch.)

Don't get your hopes up though regarding the AI's competence, the fact QA lost is probably because they're unfamiliar with counter-strategies.

 

[Chaingun]

Nice work Arumba. There's currently a modder on our beta team looking at improving the buildings script, so I will forward it to him.

 

[Chaingun]

If it's a performance thief we can always cache it (pre-calculate and store the result when some input changes). Caching is more complicated and risks introducing bugs so it's usually not done until there is a confirmed problem.