I don't have EU3, but I do have EU2, and one of the things I was hoping for most with EU3 is a decent AI (the other being a better economic/tech system, but it sounds like the old fudges of 'ahead-of-time penalty' and 'inflation' are still in place).
A lot of people seem to be saying "a perfect AI is impossible, so we should be content with a feeble one". This is a false dichotomy. The purpose of AIs in this kind of game is not to make Peter Ebbesen struggle to survive while playing as France, it's to give the bulk of players an entertaining challenge. For that, you just need an AI that avoids the worst mistakes, ie it plays as well as someone who's just learnt the basics of the game.
I don't know how it is in EU3, but in EU2, here's a list of things that even average players ought to know about, but the AI doesn't:
- supply limits
- transporting armies, specifically that you might need to split them up to fit in the boats
- terrain, in particular the basics of which situations give a big advantage to the defender, and which terrain is good for cavalry
- as long as you're not getting battered, economy is more important than military in the long term
- when expanding, focus on rich, right-religion, right-culture provinces
- only start wars if you have a very good chance of winning, and where you have a specific goal, such as capturing cores or 'good' provinces
- don't be dogmatic about peace offers, especially if the enemy makes a more generous offer than you would have suggested (I'm talking about when the AI refuses an offer to give up 1 province, then offers the same province plus 4 more the next day - that's just totally retarded)
- rebels should be crushed ASAP, especially when at peace as your armies have little else to do
- In peacetime, build troops to deal with rebels/deter aggressive neighbours if you don't have enough, and disband some if you have too many
- the 'empty slot' strategy, and an attitude to trade that's similar to war (if you can't win, don't be aggressive, but if you can, press the advantage).
- if you're big enough that you'd refuse vassalisation, you should also refuse diploannexation. Obviously a player would never accept annexation, but for the AI, I'd be happy if it was merely as hard as diplovassalisation, so that 'force-vassalise, diplo-annex' stopped being the expansionist's mantra. In fact, it would be great if the AI remembered how it had been vassalised, and the answer 'by force' made diplo-annexation even more unlikely, to the point of being nearly impossible.
- minting is not just for making armies, it should also be used to make manus (which should be prioritised more by the AI) and colonies. There could be a simple 'inflation factor' to stop this getting out of hand - the more inflation, the less inclined the AI is to mint
- don't colonise a given province once it has reached city status; instead, start a new colony in another province, or if there isn't anywhere worthwhile, stop colonising (the expense of unprofitable colonisation can really hurt the AI)
All of these are really major problems that cripple the AI's ability to compete, and put together they make the game much too easy. But all of them should have been possible to fix, without being any more sophisticated than things the AI could do already. For example, the AI should be able to use its 'war' algorithms when facing rebellions, only it should be more aggressive than in an actual war. The diploannexation, retarded peace offers and excessive warmongering problems could be mostly fixed by simply adjusting a couple of parameters in an existing formula.
I'm not saying that the AI should know about gamey exploits like synchronised looting, that it should show historical prescience when making event choices, that it should have a grand strategic vision and plan centuries ahead, or that it should get all the details spot-on. Good players will always have the upper hand here, and so they should. But the AI needs to play as if it has some basic 'common sense' about the game at a broad level.
The great thing about good AIs is that if they're too good, it's very easy to 'damage' them a bit so they play worse. That way the player can choose between 'full AI', 'moderate AI' and 'crippled AI', and so get a more genuine difficulty control.
A lot of people seem to be saying "a perfect AI is impossible, so we should be content with a feeble one". This is a false dichotomy. The purpose of AIs in this kind of game is not to make Peter Ebbesen struggle to survive while playing as France, it's to give the bulk of players an entertaining challenge. For that, you just need an AI that avoids the worst mistakes, ie it plays as well as someone who's just learnt the basics of the game.
I don't know how it is in EU3, but in EU2, here's a list of things that even average players ought to know about, but the AI doesn't:
- supply limits
- transporting armies, specifically that you might need to split them up to fit in the boats
- terrain, in particular the basics of which situations give a big advantage to the defender, and which terrain is good for cavalry
- as long as you're not getting battered, economy is more important than military in the long term
- when expanding, focus on rich, right-religion, right-culture provinces
- only start wars if you have a very good chance of winning, and where you have a specific goal, such as capturing cores or 'good' provinces
- don't be dogmatic about peace offers, especially if the enemy makes a more generous offer than you would have suggested (I'm talking about when the AI refuses an offer to give up 1 province, then offers the same province plus 4 more the next day - that's just totally retarded)
- rebels should be crushed ASAP, especially when at peace as your armies have little else to do
- In peacetime, build troops to deal with rebels/deter aggressive neighbours if you don't have enough, and disband some if you have too many
- the 'empty slot' strategy, and an attitude to trade that's similar to war (if you can't win, don't be aggressive, but if you can, press the advantage).
- if you're big enough that you'd refuse vassalisation, you should also refuse diploannexation. Obviously a player would never accept annexation, but for the AI, I'd be happy if it was merely as hard as diplovassalisation, so that 'force-vassalise, diplo-annex' stopped being the expansionist's mantra. In fact, it would be great if the AI remembered how it had been vassalised, and the answer 'by force' made diplo-annexation even more unlikely, to the point of being nearly impossible.
- minting is not just for making armies, it should also be used to make manus (which should be prioritised more by the AI) and colonies. There could be a simple 'inflation factor' to stop this getting out of hand - the more inflation, the less inclined the AI is to mint
- don't colonise a given province once it has reached city status; instead, start a new colony in another province, or if there isn't anywhere worthwhile, stop colonising (the expense of unprofitable colonisation can really hurt the AI)
All of these are really major problems that cripple the AI's ability to compete, and put together they make the game much too easy. But all of them should have been possible to fix, without being any more sophisticated than things the AI could do already. For example, the AI should be able to use its 'war' algorithms when facing rebellions, only it should be more aggressive than in an actual war. The diploannexation, retarded peace offers and excessive warmongering problems could be mostly fixed by simply adjusting a couple of parameters in an existing formula.
I'm not saying that the AI should know about gamey exploits like synchronised looting, that it should show historical prescience when making event choices, that it should have a grand strategic vision and plan centuries ahead, or that it should get all the details spot-on. Good players will always have the upper hand here, and so they should. But the AI needs to play as if it has some basic 'common sense' about the game at a broad level.
The great thing about good AIs is that if they're too good, it's very easy to 'damage' them a bit so they play worse. That way the player can choose between 'full AI', 'moderate AI' and 'crippled AI', and so get a more genuine difficulty control.