instead of AI what about mimic Intelligence?

[unmerged(83175)]

Well, if the human-human "competitive" multiplayer games database is available, then at least AI could probably deduce both what is the likely continuation of current player unit build strategy, and what would be a better unit composition to beat it.

It is much more smaller thing, and AI would only need tables of unit of all sizes, and a score of how good did the game turned out for a particular player.

That is assuming AI knows how to handle basic unit types. Obviously a smaller taks, but much more doable, and would improve HOI AI difficulty by a huge lot.

Yes, an AI that knows which unit to use for X terrain type would make the AI already much better than HoI3.
Than make certain units unmovable (so the AI can protect it's borders much better) as in HoI2 and voila the AI would be already much better than in HoI3.

 

[Imaginary Star]

Personally, I am more partial to the idea of an AI that is leaning heavily onto certain types of moves. That is, each nation's AI has a set of predetermined doctrines that it will try to use, imitating the doctrines that were used in real life. Even if these doctrines are sub-optimal, just like in real life. That would make each nation a unique character.
Or at least, give a pregame option to turn this on.

 

[winterwatchers]

I would be pleasantly surprised if the AI can use deception, such as feints, baiting, and traps. Right now the AI is predictable only because it follows a linear thinking path which is pretty much line them up and shoot. Then again, if Paradox really made a sneaky AI, alot of normal players would fail miserably and complain about the AI being too hard. In the real world, you have generals with a wide range of competence and abilities, and throw in things like luck and probabilities and trying to emulate that in a game that runs on 3gbs of ram seems a little far fetched.

 

[Beagá]

I think the most important part is AI that reacts to build orders. The obvious example is UK building less carriers and battleships if Germany is REALLY spamming subs, but there are other examples.

 

[Dark Jakkaru]

Actually, I always did want the hard and hardest difficulty be more about the AI playing to win rather than be constrained historically to repeat some mistakes. What I mean is, the AI should do its best to win with the most benefit playing the game aggressively regardless of the player is around or not. This doesn't necessarily have to entail a giant zerg rush by a cheating AI as is normally done. Rather, the AI should make best use of every means whether it's military, diplomacy, trade, or working together with partners and friends to achieve the victory conditions set by the game. This way, if any of the 3 major AI factions are busy building up working relationships with other nations they could potentially change the game's outlook each and everytime without necessarily having Ethiopia conquer Italy.

 

[Imaginary Star]

Actually, I always did want the hard and hardest difficulty be more about the AI playing to win rather than be constrained historically to repeat some mistakes. What I mean is, the AI should do its best to win with the most benefit playing the game aggressively regardless of the player is around or not. This doesn't necessarily have to entail a giant zerg rush by a cheating AI as is normally done. Rather, the AI should make best use of every means whether it's military, diplomacy, trade, or working together with partners and friends to achieve the victory conditions set by the game. This way, if any of the 3 major AI factions are busy building up working relationships with other nations they could potentially change the game's outlook each and everytime without necessarily having Ethiopia conquer Italy.

That's the kind of AI that is most difficult to make. In fact, I would like to posit, that no game of considerable complexity has ever managed to develop one that would fit this description.
In other words, yes principle is admirable, however reality tells us that this is an unrealistic expectation. Having an AI that follows certain pre-rendered patterns is more feasible, and more of a semblance of real history. History teaches quite plainly that nations do not act according to pure logic, try as they might, they are also driven by customs, emotions, vices, virtues, etc. Most of those are hard to code (habit/custom is one exception, for example), but by imitating real events, we can at least imitate some of these.
I think "To End All Wars" did it rather well - each nation has several historical "plans", and every game, AI chooses one of them and attempts to execute it throughout the game. Simple, elegant, and doable, leaves room for alternative history of the war.

 

[Dunbal]

how about have every player send his play
to paradox server

This poses a few problems, not the least of which being that there are always a not insignificant number of individuals in a given population who are more than happy to exploit such a bad design by intentionally feeding Paradox "bad" moves either out of malice or simply curiosity. As Terry Pratchett once pointed out about human nature, to paraphrase:- even if you took a big red button and placed a sign on it saying "END OF THE WORLD SWITCH DO NOT PRESS" and placed it in a hidden cave somewhere, the paint would not have time to dry before someone pressed it just to see what happens...

Your method could produce something worse than even bad AI could ever come up with. Rule 1 of programming: NEVER trust the user!

 

[Rastrigin]

Hoi3 is a very difficult game to project an AI for. The amount of variables is huge and there is no penality for overextension like in EU. Also, even very little mistake can lead to crushing defeats against human players, as MP games showed so many times.

Some things are already quite decent, like the air war, the sub warfare and the management of land fronts against a player with and equal or inferior force that distributes his forces evenly. If the front moves slowly

In my opinion some the biggest flaws of the hoi3 AI are:

_Inability of having a focused build up strategy (Hoi3 really discourage dispersive builds)
_Too many garrisons even where not needed, so the forces are too spread out and many divisions are very vulnerable. The AI should be able to distinguish cases in which a port garrison is not needed, like when there is naval superiority, or when an inland reaction force is better than divisions in the ports. If the choice is to garrison the coast, than it is needed more than some divisions in the ports, but also garrisons in the adiacent provinces and something inland.
_More in general, (related to the previous point) the ability to prioritize some theatre and leave others undefended (too dispersive)
_Also related with the previous, the ability of chosing which theatre to prioritize (Germany commits way to much forces in Norway everytime)
_The ability forming a sparehead for offensive warfare and countering the enemy sparhead in defensive warfare. As said, the AI alwais work with an evenly spread force, which is very ineffective.
_The ability of keeping the forces focused on a target. The AI management of units is a mess. always strategic reployments everywhere (which, btw, creates a lot of supply problems), troops that take a province and than retreat instead of pushing to next one. I've often seen Germany stopping it's adavance on at the gates of Moscow for no reason, with the Soviet defence already shattered.
_Awful management of HQs, alwais out of range.
_Inability of breaking well built defensive lines, even with a crushing superiority in numbers/quality, the AI can't make sacrificing attacks, for example to cross a river. It just stands there.
_while newly built units have a quite OKish composition (with exeptions), the AI can't rework existing units compositions.
_I don't even want to talk about the naval warfare. I know it is difficult, but at least a proper fleet conposition and bigger fleets would be nice. And the AI probably needs to be a little more conservative.

 

[Chromos]

Isn't "mimicking" not basic ai coding?

You think about what should do if this or that happens, and then act like this or that..
Taking more peoples experience will help in getting smarter, if the basic coded ai is smart enough to use it. ->If you can incorporate it into the ai easily enough.
And easily enough also means the amount of cpu time needed. As too much cpu time needed, is very likely a KO-criteria fro a game like HoI.


Besides of that, I think that even the HoI3 ai would be able to be much better with just a "few tweaks" to the exe and Lua.
I can only repeat that most never saw how good the warfare ai of HoI3 can be, because it needs some more units to shine. Some modder like Dr.Johnson proved that years ago.

As it is said that HoI4 is done completely anew, it should be able to do a much better ai for HoI4 as we had with HoI3.
HoI3 was also better then HoI2(not AoD/DH, but vanilla HoI2!) in that regard.

 

[Dakk]

just type "selflearningai" in the console to enable it.

Actually they nerfed the AI to cater for casual players. But you can just add "unnerf_fullAI" in defines.lua AI-section, and the game is almost unbeatable.

 

[Imaginary Star]

Actually they nerfed the AI to cater for casual players. But you can just add "unnerf_fullAI" in defines.lua AI-section, and the game is almost unbeatable.

Shhh! Casuals will hear you.

 

[Pugmak]

I would like to see, I think, the intel efforts, battlefield radio security and intercept research and battle plans to be part of the "enemy" AI (single player non player AI).

IF the AI gets ahead on the espionage efforts and the radio security/intercept research, then there's a chance the AI would gain a copy of the player's battleplans. The AI could then adjust it's plans to fit.

For this to work, there'd have to be some significant penalties to the player for either deviating much from his plans or not using plans at all. Penalties in org, morale, supply flow, etc.

 

[Rastrigin]

Actually they nerfed the AI to cater for casual players. But you can just add "unnerf_fullAI" in defines.lua AI-section, and the game is almost unbeatable.

Are you talking about EU4 or HOI3?

 

[Secret Master]

IF the AI gets ahead on the espionage efforts and the radio security/intercept research, then there's a chance the AI would gain a copy of the player's battleplans. The AI could then adjust it's plans to fit.

They have already said it is possible to steal battleplans AND produce fake plans for others to steal.

How this works with AI is unknown.

 

[Dark Jakkaru]

That's the kind of AI that is most difficult to make. In fact, I would like to posit, that no game of considerable complexity has ever managed to develop one that would fit this description.
In other words, yes principle is admirable, however reality tells us that this is an unrealistic expectation. Having an AI that follows certain pre-rendered patterns is more feasible, and more of a semblance of real history. History teaches quite plainly that nations do not act according to pure logic, try as they might, they are also driven by customs, emotions, vices, virtues, etc. Most of those are hard to code (habit/custom is one exception, for example), but by imitating real events, we can at least imitate some of these.

Crusader Kings did it well and somewhat in a manner as I would have liked. Granted, this type of AI behavior is actually difficult to beat rather than to emulate if they behaved the way they did in HOI 4! So, the Ethiopia/Italy that the dev team mentioned is an example where the AI will play to its strengths and really toss up the game around even it seems bugged for it to do that. The only problem with that is most folks would feel like it is out of place even though the AI has no predisposition on what is out of place or not, it just does routines and can do them well enough. So it's really not an AI issue since you could make the AI into a difficult battle-chess like opponent that will beat the player most games and not really make it that fun. The other flip-side of the issue is if the AI does pre-rendered plans too much, then it seems the AI is too formulaic in its approach and loses its repeatability.

Of course, what I really wanted was that the difficulty modes entailed making the AI stronger and more of a challenge for repeatability more than anything else AND less about a game of the AI doing one or the other making it easy to trick the AI into loses time and time again. I had in a way not wanted it just to be locked into just making a historical scenario more difficult but rather the challenge be broader and more strategic with less predictability against the AI.

 

[Imaginary Star]

Crusader Kings did it well and somewhat in a manner as I would have liked. Granted, this type of AI behavior is actually difficult to beat rather than to emulate if they behaved the way they did in HOI 4! So, the Ethiopia/Italy that the dev team mentioned is an example where the AI will play to its strengths and really toss up the game around even it seems bugged for it to do that. The only problem with that is most folks would feel like it is out of place even though the AI has no predisposition on what is out of place or not, it just does routines and can do them well enough. So it's really not an AI issue since you could make the AI into a difficult battle-chess like opponent that will beat the player most games and not really make it that fun. The other flip-side of the issue is if the AI does pre-rendered plans too much, then it seems the AI is too formulaic in its approach and loses its repeatability.

Of course, what I really wanted was that the difficulty modes entailed making the AI stronger and more of a challenge for repeatability more than anything else AND less about a game of the AI doing one or the other making it easy to trick the AI into loses time and time again. I had in a way not wanted it just to be locked into just making a historical scenario more difficult but rather the challenge be broader and more strategic with less predictability against the AI.

What about the "To End All Wars" solution to the issue. Granting AI several predetermined (actual historical, even) paths to pursue solves the problem of repetition for the most part.

I also want to point out that while CK2 did a splendid job of emulating the kingly behavior and webs of succession, the job of handling the war in HOI is so very complicated, at the very least in a very different way than CK2. After all, war on a grand scale is the reason HOI series exist.
Some lessons from CK2 can certainly aid making a better AI in HOI, but the extent of said lessons is limited, I believe.

 

[ringhloth]

What about the "To End All Wars" solution to the issue. Granting AI several predetermined (actual historical, even) paths to pursue solves the problem of repetition for the most part.

Because it didn't work? In 1.01, victory was assured by the Germans, even if they didn't advance past Belgium, by the end of 1914 because the French were incredibly stupid about attacking. It didn't really change how the AI performed. It certainly didn't make the AI better. It'd be near impossible to replicate with HoI the way it was done in TEAW.