Features of AI on hard difficulty
- Thread starter jarpiekar
- Start date
-
We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
That sounds nice, if it just didn't revolve around targeting the player. I would love to have the AI more likely to attack countries when they're weak and be more open about giving Condotierri, but I don't see how it's fair that a higher difficulty results in the AI targeting the player to the point where they refuse to ally on difficulty basis alone.
That, in my opinion, is very bad design for increased difficulty. Could you at least add these options in defines.lua so some of these AI "features" (less likely to ally player) can be turned off in a non-ironman game?
The AI player doesn't currently do any planning, i.e. considering (ahead into the future) the effects of a sequence of actions it might take. A number of independent subsystems perform a series of (what you could call) emotionally driven actions, coordinated (for the most part) only implicitly through observed changes in the gamestate. Sometimes it manages to form a coherent strategy nonetheless.
Overall, there is limited value of looking more than 1 step into the future in this game any way due to the vast number of possible outcomes. If you can quantify how much you will immediately be better off by doing something, that's enough to make decisions that are reasonable in most cases. However, it does lead to things like FRA easily conquering Iberia because CAS/ARA/POR don't realize the future threat of destroying each other in intra-peninsular wars.
I don't think we'll ever attempt to incorporate a general planning system in EU4 due to performance reasons (disregarding the small detail of its implementation).
Algorithms don't design themselves you know (actually, they can, but...)Currently up thinking about the army AI again at this hour because it's my big headache.
So...Why you guys merge special bonus and AI difficulty and separate them? Like I want to try special positive modifier on player but play on hard or medium.
- 3
.png){kind=avatar}
I'm very curious on wether you guys are planning on improving the AI logic for choosing idea groups?
Because i think a large amount of the AI's downfalls come from them choosing all the wrong idea groups at the wrong points in time, i mean, many tiny/small nations choose groups like Administrative, Innovative, Diplomatic and Influence as their first picks, which usually doesn't benefit them much due to their size.
Because i think a large amount of the AI's downfalls come from them choosing all the wrong idea groups at the wrong points in time, i mean, many tiny/small nations choose groups like Administrative, Innovative, Diplomatic and Influence as their first picks, which usually doesn't benefit them much due to their size.
Yes, but to be done for a different purpose. Wanting or signing a NAP means that there is another threat incoming that you may have to deal with and don't want problems at the moment. The AI could be programmed to want to sign NAPs if they are made aware that they may be the next target if the other nation they signed the NAP with falls. It can be interrelated with the region system.
Example: if X is a threat in Lower Bavaria, then Y in Upper Bavaria may want to sign a NAP with Z in Venetia. Z in Venetia will accept because they know that X will be close to them if Y is weakened too much. If Z knows Y is stronger than them then Z will seek an alliance with X.
Maybe you could increase difficulty with slowing the game down, while giving more calculation opportunities to the AI nations?
Increase the number of intelligent stacks?
Maybe doing the diplo thingy which happens after reloades every 5 years?
Increase the number of intelligent stacks?
Maybe doing the diplo thingy which happens after reloades every 5 years?
.png){kind=avatar}
How about a check box we can enable for more AI calculations for those of us now using a toaster PC? I think Civ games did this.
Condottieri on Hard isn't aimed specifically against human, it also makes chances you can hire them greater as AI is more willing to offer them in general. Greater overall Condottieri use makes the game more dynamic and thus harder.
The game can be different in other ways on hard; AI does for instance cheat on revolt risk on Hard which leads to Timurids often holding together unlike on Normal. (This specific bonus we might find reason to nerf.)
I would eventually like to move most human specific checks to Very Hard. My vision for difficulty levels is:
Very Easy: AI extremely nice to human. We have a category of players for whom the game on Easy is too hard..
Easy: Same as now, perhaps disabling difficult features (and for Very Easy too).
Normal: Same as now.
Hard: Some light bonuses to AI in areas where it has problems (same as now, possibly reduced), advanced AI features, but "fair".
Very Hard: Very heavy bonuses to AI (similar to the bonuses of old) and checks to make it cooperate against human to a limited degree. For the masochists.
This is just want I want to do, and it's only a matter of time to be available.
Thanks for answer. Could you add two different settings in option. One for AI bonuses and second for AI features tuned on? So I could play on full AI capable algorithms but without bonuses they get on hard ?
- 3
Agreed. Putting these much needed AI features under Hard difficulty, under which they get unfair other bonuses or additions (ex. targeting the player, which I hate as a poor mechanic unless it's a competitive/ironman session, which I never play, or getting 50% manpower recovery speed boost, which is hilariously unfair), is pretty ridiculous.
- 1
@Chaingun
Any chance since we are between releases that one of the device diary topics could cover a dummies version of AI programming for EU4 ? I think Wiz mentioned that as a possible topic. Something that could give the average player some insight on the process and a better idea on why certain ideas work while others are too problematic.
Any chance since we are between releases that one of the device diary topics could cover a dummies version of AI programming for EU4 ? I think Wiz mentioned that as a possible topic. Something that could give the average player some insight on the process and a better idea on why certain ideas work while others are too problematic.
- 1
I'd also suggest adding a "World Conquest Mode" so that whenever people complain that some game mechanic makes WC too hard/easy/boring/stressful/unfun/unrealistic/[insert other complaint], the Devs can just strip that feature out of "WC Mode" and move on to other things.
- 2
I'm not sure this would be effective without crippling game speed. Adding a few seconds per turn for 20 odd ai entities is a different world than trying to add cycles to 300+ ais running in real time...
His earlier post about how the AI is a collection of simple instincts with no capacity for foresight or planning more or less covers it. In order for the AI to "reason" about anything you first have to turn the high-level conceptual problem ("what idea group should I pick at this moment?") into a low-level math problem, because math is all a computer understands.
Imagine every piece of data that factors into picking idea groups. Now turn all of that data into numbers. Now see if you can figure out how to add those numbers up in a way that is consistently better than just hard-coding a list for each tag to use.
Sounds like the AI plays like actual rulers of the time period, then. With a general strategy of what they want but, in general, living mostly in the moment.
Condottieri on Hard isn't aimed specifically against human, it also makes chances you can hire them greater as AI is more willing to offer them in general. Greater overall Condottieri use makes the game more dynamic and thus harder.
The game can be different in other ways on hard; AI does for instance cheat on revolt risk on Hard which leads to Timurids often holding together unlike on Normal. (This specific bonus we might find reason to nerf.)
I would eventually like to move most human specific checks to Very Hard. My vision for difficulty levels is:
Very Easy: AI extremely nice to human. We have a category of players for whom the game on Easy is too hard..
Easy: Same as now, perhaps disabling difficult features (and for Very Easy too).
Normal: Same as now.
Hard: Some light bonuses to AI in areas where it has problems (same as now, possibly reduced), advanced AI features, but "fair".
Very Hard: Very heavy bonuses to AI (similar to the bonuses of old) and checks to make it cooperate against human to a limited degree. For the masochists.
This is just want I want to do, and it's only a matter of time to be available.
Regarding Very Hard, have you considered player maluses together with moderately hard AI bonuses? Very heavy AI bonuses can mess up the AI vs. AI balance. Player maluses also scale better and will always be on target. EU3 used this and it worked very well.
- 1