AI Improvement Project

[Chaingun]

Anyone interested in an AI improvement project since the EU2 source code has been released? The goal would be to make an AI that plays more competently according to the game's rules, rather than a historical AI. This in order to make it offer more of a challenge, making it more comparable to a multiplayer opponent (to the extent that's possible - it won't ever be as good as a [competent] human).

 

[MichaelM]

Hmm... I think that a lot of people would like to see such a project, but very few could help. I personally have taken a class in machine learning, but I don't remember enough of it to try to do it in a real game.

Still, I'll stick around just in case I can help.

 

[unmerged(1047)]

I would think there's a lot that could be done to improve AI's without necessarily editing the engine. More specific AI's for minors and events that shift AIs based on time period, size of territory, etc. could easily be done within previous limits.

For example, there's no reason France or England or Spain should necessarily have the same AI throughout. You might even be able to "steer" colonization or wars in more historical directions by AI tweeks and switching events. Also handling ahistoric powers - Aragon or Genoa, for example, surviving and becoming a serious power, should want to colonize.

I would, however, like to see a general reduction in the costs of colonization and missionaries - these were often carried out outside of the government budget. I'd think that missionaries should generally be 20-30% cheaper than they are, and colonists 50-60% cheaper.

Also, one that I've always thought should be looke at is the "wrong-religion" tax penalty for provinces, which is currently constant. I think, in light of historical examples like the OE, that it should be somewhat fluid, based on either religious tolerance or narrowminded/innovative slider (or both). The mechanics could be similar to how the overseas penalty is toed to the land/naval slider.

 

[MichaelM]

I would think there's a lot that could be done to improve AI's without necessarily editing the engine. More specific AI's for minors and events that shift AIs based on time period, size of territory, etc. could easily be done within previous limits.

For example, there's no reason France or England or Spain should necessarily have the same AI throughout. You might even be able to "steer" colonization or wars in more historical directions by AI tweeks and switching events. Also handling ahistoric powers - Aragon or Genoa, for example, surviving and becoming a serious power, should want to colonize.

Like what several mods (AGCEEP, EP, Daywalker's TGAI, and possibly others) have done already? Or something more?

I would, however, like to see a general reduction in the costs of colonization and missionaries - these were often carried out outside of the government budget. I'd think that missionaries should generally be 20-30% cheaper than they are, and colonists 50-60% cheaper.

Possibly so, but I think this project is just for AI improvement. Perhaps someone should start a general improvement project.

Also, one that I've always thought should be looke at is the "wrong-religion" tax penalty for provinces, which is currently constant. I think, in light of historical examples like the OE, that it should be somewhat fluid, based on either religious tolerance or narrowminded/innovative slider (or both). The mechanics could be similar to how the overseas penalty is toed to the land/naval slider.

I like this idea a lot. I would also like to have the narrowminded/innovative slider affect your overall amount of tolerance. But again, I don't know that these changes are relevant to an AI project.

 

[unmerged(40707)]

Possibly so, but I think this project is just for AI improvement. Perhaps someone should start a general improvement project.

or resurrect old ones (but it is against the rules).

It is a fact many things could be done and not only for AI.

 

[unmerged(1047)]

Perhaps I got a bit carried away tossing out ideas. However, I think the point was made, in that there are areas other than AI that could be addressed. And (this one is a bit more grey) that there are things that could be done to improve the AIs without necessarily tinkering with the engine, while some of the other points are things that cannot be done any other way.

If anyone wants to throw out an idea for a more general project, feel free to use my ideas. I'm not at a place in my life, nor computer-skilled enough, where I could take an active role in such a project beyond tossing out concepts like these.

I'd start a thread, but it seems to me that around here people expect those who start such a discussion to take a leading role in the resulting project.

 

[Chaingun]

I can work on AI in a more general project too. Putting this in a more general project probably has the advantage of more easily getting approved by Paradox, as I don't think they can sell an "AI improvement expansion".

AI is just something I'd be very interested to work with, because I've thought of how stupid EU2 AI is for years, and what could happen if it was smarter. I used to play a lot of EU2 MP and the AI is making some really basic mistakes that ruin it. Example: Piling up 100k soldiers in a province with 10k supply limit, spending its money in odd ways, etc.

And yeah, there is a mod (Daywalker's) which basically maximizes the capability of the AI from the modding side already. However, there are a lot of things he couldn't change because they are in the source code.

 

[unmerged(40707)]

I can work on AI in a more general project too. Putting this in a more general project probably has the advantage of more easily getting approved by Paradox, as I don't think they can sell an "AI improvement expansion".

AI is just something I'd be very interested to work with, because I've thought of how stupid EU2 AI is for years, and what could happen if it was smarter. I used to play a lot of EU2 MP and the AI is making some really basic mistakes that ruin it.

So true.

And yeah, there is a mod (Daywalker's) which basically maximizes the capability of the AI from the modding side already. However, there are a lot of things he couldn't change because they are in the source code.

And this mod is not the only one.

 

[Hive]

Isn't AI some of the hardest stuff to work with, though?

But I recall the AI in games like Victoria and HoI being a great deal more clever than in EU2, maybe examining those could be a good basis?

 

[unmerged(40707)]

AI is arguably the hardest thing to work with.

This is maybe for this that "cross-game" source code will be very interesting.

 

[Chaingun]

Isn't AI some of the hardest stuff to work with, though?

That's exactly why I'd like to work with it. I've basically done everything else (though admittedly I never finished anything). I believe developing oneself should be a key objective in whatever personal project one persues (since it's such a lot of invested effort). I wouldn't exactly do this to make money; in fact if I tried to make money I wouldn't be involved in game development at all.

 

[Hive]

Well I would definately love to see an improved AI, if someone is able to pull it off. I sincerely doubt I myself can.

 

[Chaingun]

I believe best chance for this is if I join someone else's team and work on AI there. Anyone interested? I could probably help with a few other things too if required.

Well I would definately love to see an improved AI, if someone is able to pull it off. I sincerely doubt I myself can.

I do think it's very possible to improve the AI by a huge amount by the fixing its worst shortcomings.

For example the AI tends to lose 90% of its soldiers in a war against me due to attrition. Consider that again; 90% is totally HUGE. It means only 10% of AI troops actually inflict combat losses on me. The corresponding attrition/total loss ratio for a human player averages at 30-50% (some of it is unavoidable). If the AI can be improved in this respect, Very Hard will become Very Utterly Insanely Impossible difficulty (for minors), because the AIs damage output on me in battles may rise by a factor 5.

The problem is because the AI is absolutely terrible at taking province supply limits into aspect when sieging (especially with multiple allies). Of course, nothing is as simple as it seems, since the AI is easier to attack if it puts 6k to siege a minimal fort. However, putting 30k in a 5 supply province just makes the human wait until AI army goes down to attrition. AI should station troops in neighbouring friendly provinces instead to be ready to rush to the defence when human attacks.

I also think the economic AI can be improved a bit, very easily, by simply assigning better build priorities for troops/colonies/manufactories etc. AI needs for instance consider how costly a war is, and act thereafter when negotiating peace/deciding whether to continue. (Some humans don't do this rationally admittedly, which leads to some hilarious disasters. :rofl: )

Another thing I would love to work on is the diplomatic model. I'd completely remove influence of -200/200 relation scale on AI decisions (except for things that have direct influence on gameplay such as good relations for diplo-annexation etc.), and replace with true geopolitical diplomacy. France too big? Attack them because they're big, not because they have -200 with you.

 

[cool-toxic]

Well I think it would make sence if you joined this team:

http://forum.paradoxplaza.com/forum/showthread.php?t=352140

I think it's almost all EU2 players team.

 

[unmerged(40707)]

I believe best chance for this is if I join someone else's team and work on AI there. Anyone interested? I could probably help with a few other things too if required.

As cool-toxic said, see here and especially post #5 and PM me if you want. You'll have to sign up here.

I do think it's very possible to improve the AI by a huge amount by the fixing its worst shortcomings.

For example the AI tends to lose 90% of its soldiers in a war against me due to attrition. Consider that again; 90% is totally HUGE. It means only 10% of AI troops actually inflict combat losses on me. The corresponding attrition/total loss ratio for a human player averages at 40-50% (some of it is unavoidable). If the AI can be improved in this respect, Very Hard will become Very Utterly Insanely Impossible difficulty.

The problem is because the AI is absolutely terrible at taking province supply limits into aspect when sieging (especially with multiple allies). Of course, nothing is as simple as it seems, since the AI is easier to attack if it puts 6k to siege a minimal fort. However, putting 30k in a 5 supply province just makes the human wait until AI army goes down to attrition. AI should station troops in neighbouring friendly provinces instead to be ready to rush to the defence when human attacks.

I also think the economic AI can be improved a bit, very easily, by simply assigning better build priorities for troops/colonies/manufactories etc. AI needs for instance consider how costly a war is, and act thereafter when negotiating peace/deciding whether to continue. (Some humans don't do this rationally admittedly, which leads to some hilarious disasters. :rofl: )

Yes, you're right. Not sure if this will be easy but we can have a try.

 

[MichaelM]

Well I think it would make sence if you joined this team:

http://forum.paradoxplaza.com/forum/showthread.php?t=352140

I think it's almost all EU2 players team.

As cool-toxic said, see here and especially post #5 and PM me if you want.

Well, that project is more for AGCEEP players than for vanilla players (there still are some, right?). A lot of the changes proposed in that thread are adding abilities for modders to use, rather than enhancing core game play. Is an all-in-one project really good for EU2?

 

[Chaingun]

Well, that project is more for AGCEEP players than for vanilla players (there still are some, right?). A lot of the changes proposed in that thread are adding abilities for modders to use, rather than enhancing core game play. Is an all-in-one project really good for EU2?

That is a good question. My idea of "AI improvement" might not correspond to 100% historically accurate, as I just want it to play as efficiently as possible without cheating. Though I suppose one could develop a configurable AI to have an on/off setting to respect historical boundaries.

On the other hand I don't really think the playerbase is big enough for too many ambitious projects.

 

[unmerged(40707)]

Well, that project is more for AGCEEP players than for vanilla players (there still are some, right?). A lot of the changes proposed in that thread are adding abilities for modders to use, rather than enhancing core game play. Is an all-in-one project really good for EU2?

This is why I expressed two parallel goals. EU2 engine deserves enhancement that will be used by AGCEEP but not only. And these enhancements are not just about script engine and moddability.
What I wrote first in post #25 of the other thread and other suggestions are obviously for modders but some aspects are already about the game as a whole.

IMHO, it is only a matter of schedule. We can start with "easy" enhancements, see how situation evolves, and plan more deeper ones for a future version.

On the other hand I don't really think the playerbase is big enough for too many ambitious projects.

I don't know for the global playerbase but I think projects will probably focus on the same evolutions, at least at start.
OTOH, who would, as a consumer, want to choose between a bunch of expansions that differ only on different parts from the original game(s) in the end, not to mention required resources to check them all for Paradox?

Better have as few as possible projects especially if they go in the same direction or complementary directions. This is not about killing competition but rather promoting cooperation and consensus if possible.
About existing mods in general and AGCEEP in particular, main question will be adopting or not expansions or being compatible with them. It is a danger but also an opportunity especially if one of the expansions is considered to be the long awaited and logical continuation of EU2 1.09.

Main goal must remain bringing the most enjoyable experience to players in keeping what makes the uniqueness of the game.

 

[Toio]

I believe best chance for this is if I join someone else's team and work on AI there. Anyone interested? I could probably help with a few other things too if required.



I do think it's very possible to improve the AI by a huge amount by the fixing its worst shortcomings.

For example the AI tends to lose 90% of its soldiers in a war against me due to attrition. Consider that again; 90% is totally HUGE. It means only 10% of AI troops actually inflict combat losses on me. The corresponding attrition/total loss ratio for a human player averages at 30-50% (some of it is unavoidable). If the AI can be improved in this respect, Very Hard will become Very Utterly Insanely Impossible difficulty (for minors), because the AIs damage output on me in battles may rise by a factor 5.

The problem is because the AI is absolutely terrible at taking province supply limits into aspect when sieging (especially with multiple allies). Of course, nothing is as simple as it seems, since the AI is easier to attack if it puts 6k to siege a minimal fort. However, putting 30k in a 5 supply province just makes the human wait until AI army goes down to attrition. AI should station troops in neighbouring friendly provinces instead to be ready to rush to the defence when human attacks.

I also think the economic AI can be improved a bit, very easily, by simply assigning better build priorities for troops/colonies/manufactories etc. AI needs for instance consider how costly a war is, and act thereafter when negotiating peace/deciding whether to continue. (Some humans don't do this rationally admittedly, which leads to some hilarious disasters. :rofl: )

Another thing I would love to work on is the diplomatic model. I'd completely remove influence of -200/200 relation scale on AI decisions (except for things that have direct influence on gameplay such as good relations for diplo-annexation etc.), and replace with true geopolitical diplomacy. France too big? Attack them because they're big, not because they have -200 with you.

the issue on why the AI is so terrible might be due to the fact that we do not know what all numbers are supposed to do for each section , yes we have a good idea on base and front , war, trade and a few others ,but the combat and garrison sections are a hit and miss affair,
maybe with access to the source code we might really see what they are suppose to do and MAYBE with this knowledge the AI does not really need that much attention to it

 

[Chaingun]

Don't think you'll find an AI which works perfectly underneath, even with optimal modifiers.