So.... A.I. why hasn't been any info, focus or DD about it?

[Krafty]

And yes, infact, I think anyone with an interest in the subject, would like to know the details of the AI "thought process"

Going back to say, Civilization Call to Power, the AI was basically given a giant "script" to unfold (Like, the Iriqouis was scripted to perfer building its unique warrior over archers) and had a very basic set of algorithms that worked on 'same turn' basis (that is that the ai only thought about the situation as it stood when the player hit "End Turn").

Is that still basically how the AI will function in HOI4?

As a "real time game" HOI is, if im not mistaken, really just "turns" or "ticks" that you dont have to hit end turn for. At speed 1, the ticks are normal, at each higher speed, a multiplier is attached. Time isnt moving faster persay, just where if you were getting 1 iron a day (making this up) on speed 1, and on speed 5 you went through 5 days a "tick" youd basically, in a progamming sense just be getting "5 iron a day", there is still only one tick in the same time frame, but its multiplied, giving you the sense that time is passing faster.

What im wondering, and im sure I already know the answer, is HOI going to be 64 bit only, and will the AI "think" ticks or turns, into the future. Also, if the AI is going to be using scripts to control its most likely behavior, can we adjust those scripts as modders, and will those scripts be documented and commented (//). That would be just as good as telling us now how the AI behaves.

All that said. Thank you very much for creating the AI, and sharing this information with us. Were literally foaming at the mouth to get our hands on Hoi4 and start fighting world war two

PS:

Interestingly, Battlegoat's games released by Paradox, those Supreme Commander games, have some interesting AI. Those games are, to put it mildly, an acquired taste, and the AI is braindead. However, this modder named Ruge came along, and noticed that the AI was able to be edited, and that the developers and purposefully gimped the AI (whhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhy). So in a simple 500kb mod, he was able to make a completely placid and flacid AI, into something that actively tried to annihilate you. It was actually really good. The problem there was that the AI couldnt handle the economic system, political system, or build province (hex) improvements. So ultimately you could beat it. But, if you just say, took Russia, and declared war immediately on NATO, the AI really did a pretty good job of acting just like a human being that was overly aggressive. It didnt use units any better, or act any more strategically, but it was unrelenting on all fronts and exploited every weakness.

Ultimately it would fail, as it didnt build, research, or properly manage its economy, but aggression level was great. There were never units just sitting in some far off place protecting something youd never attack ALA HOI's France. The AI was very good at deciding, we can win this, and throwing everything at you, or deciding, we cant win, and fighting a very good active defense.

 

[Krafty]

I think really it boils down to "Can the AI think about what the situation MIGHT be in the future and make a decision now, even if its wrong, or is the AI stuck thinking about the situation as the game ticks in real time".

The most recent Galactic Civilizations has an AI that does that, but, as said, its still bad and not even remotely as good as a player. The only way its a challenge, is to give it arbitrary (cheating) bonuses.

However the AI will throw some curveballs at you in GalCiv, which is the best I can say for any AI ive ever seen outside of Gary Grigsby's War in the East. The German AI was fundamentally flawed, but the Soviet AI was masterful, without any bonuses.

That was way easy for them to do however, as essentially, a two player game, is chess...with more rules. They were able to take a cheap outsourced AI routine, and plug it into their game. Which still took 5 years to be competent.

 

[Krafty]

And for what its worth I think Empire Deluxe's AI is better than C-Evos

But C-Evo is still being worked on. So time will tell.

But seriously, try this, Go to google and type in "best AI in video games".

Yeah its a pretty paltry list. To the point that id like think us savvy gamers, realize, that AI, is a misnomer, and does not exist. What you have, is pretty much a scripted machina. As was mentioned, a 'set piece' to keep the player amused and immersed. Its simply there to provide something to keep the player clicking. Not something that is actively trying to win. Its just there to give you something to interact with, and given some percentage chance for things to happen, to give the appearance of unpredictability (in some games) or of "purpose".

Theres less than 10 games I can think of, where the AI is actually trying to beat you. Empire, C-evo, TOWA, Chess, etc...

The most obvious thing that kills AI for me, is the inability to make "pacts" that make sense. Even in Eu4.

The reason Europe today, isnt one country, but has been fought over for centuries with no clear winner, is because of alliances. One person gets too strong, so a coalition is formed against them keeping the balance of power and eventually knocking the leader down a peg. Eu4 tries to do that, but its not dynamic, its simply a function of how "bad" you are. Stay under the "bad limit" and no one cares that youre gobbling up the world.

Sorry for ranting on for 4 posts....but AI in video games is something im deeply interested in, and ive played literally thousands of games over the years, alot of them strategy, and ive never seen a good AI.

So im a tad bit worried about this HOI, that seems to hang so much of its impressiveness, on said AI.

 

[Secret Master]

The reason Europe today, isnt one country, but has been fought over for centuries with no clear winner, is because of alliances. One person gets too strong, so a coalition is formed against them keeping the balance of power and eventually knocking the leader down a peg. Eu4 tries to do that, but its not dynamic, its simply a function of how "bad" you are. Stay under the "bad limit" and no one cares that youre gobbling up the world.

Do you remember the earlier days of EU4? I do.

I remember when the AI would form vast coalitions to stop anyone from getting too big. I remember when even vaguely resembling Louis XIV or Napoleon would result in a series of coalition wars.

Do you remember why those hard-core coalitions were ended? It wasn't because the AI was stupid. It was because players whined that it was unrealistic (it wasn't, and even a cursory list of coalitions from the 15th Century on demonstrates the point) and that it wasn't fun. So, the AI coalition behavior was toned down a bunch.

It wasn't perfect, but the end result was that if the AI was eager to form coalitions against expanding powers, players cried about it. Make an AI that even tries to run a foreign policy on par with Talleyrand or Bismark, and players will complain more than they do when the AI is being stupid.

 

[Krafty]

And this is the last post I swear.

One of my concerns, seeing the press videos, is that the AI is perfectly willing to make bad attacks. NorthernLion for instance, put some Mountain and Cav into an army and sent it into Romania where it stalled out crossing the Danube.

This kind of sums of my concern. Now, if youre the Soviets, with hundreds and hundreds of divisions, who cares if youre throwing Mountain and Cav at an uncrossable river when theres a hill province to attack instead. Youre going to win in that situation eventually, no matter what. Some inconsistency and errors improves immersion.

If youre Romania, with the same AI thought process...welp...youre going to be pulling your hair out and micro managing just like you did in Hoi3, or 2, or any other game, and the entire system is superfluous.

Is the AI going to be capable of "knowing" things like, were going to win no matter what, dont bother with efficiency, were going to lose no matter what, dont bother with efficiency, were slightly losing, increase efficiency, etc.

I just want to avoid that hair pulling moment when my SNLF forces land on a heavily defended port, when I told them to invade the Phillipines, instead of a nice undefended or lightly defended port down on Luzon. Then im reloading a save game, turning off the automation, and microing it, in which case why even bother creating the thing in the first place. And if the US AI is using the same "logic" to reconquer the Phillipines, im going to have a silly easy time defeating them. Which sucks.

Tons of changes have me jumping for joy (real equipment figures!!! were a wargame finally!! Move over SSI, Avalon Hill, and Gary Grigsby, Hearts of Iron is in town!!) but this whole AI thing has me a bit concerned, since it was such a pointless part of Hoi3.

 

[Krafty]

Do you remember the earlier days of EU4? I do.

I remember when the AI would form vast coalitions to stop anyone from getting too big. I remember when even vaguely resembling Louis XIV or Napoleon would result in a series of coalition wars.

Do you remember why those hard-core coalitions were ended? It wasn't because the AI was stupid. It was because players whined that it was unrealistic (it wasn't, and even a cursory list of coalitions from the 15th Century on demonstrates the point) and that it wasn't fun. So, the AI coalition behavior was toned down a bunch.

It wasn't perfect, but the end result was that if the AI was eager to form coalitions against expanding powers, players cried about it. Make an AI that even tries to run a foreign policy on par with Talleyrand or Bismark, and players will complain more than they do when the AI is being stupid.

I got EU4 after the first expansion. So I dont recall that.

Why on earth isnt there an option to enable, or disable that. Id love to play a game where im actually challenge over taking territory.

Crying should have led to an option. Im one of those people who is extremely annoyed with how easy it is to blob out as OPMs.

One would think on "Normal" it would function as it currently does, in "hard" they would form coalitions like crazy. Isnt it possible to have to best of both worlds as options?

I think no matter what you do, someone will cry. Its too hard for some, too easy for others. That seems to be the realm of difficulty settings. And difficulty settings that actually raise the AIs abilities. Not something that just throws at -15% at you.

Like, Arcade (super dumbed down) Easy (gimped ai) Normal (ai working at full steam, but with an arm tied behind its back, IE, its not trying to beat you, just survive) Hard (ai is trying to win) and Impossible (ai is trying to win, AND cheats with bonuses).

Crazy right? *shrug* seems more worth the time and effort to develop, than an all encompassing ai that is just the right balance between good and aggressive, and terrible, to let the player win. (since youll never find a balance that pleases everyone)

Its kind of annoying that you can go play Black Ice on Grofraz and still WC with Albania. The AI is terrible. Even with bonuses out the ying yang, you can game it into the ground because at the end of the day, its not trying to beat you, its just trying to look active enough that youre engrossed in the game.

Just once, id like to see AI germany with hordes of Tac bombers, destroy my infrastructure and roflstomp me into the ground. Just once. Like the first ever Soviet game I played in Multiplayer HOI2. Heck...id just like to see the AI make a sensible logistical bombing. Preferably on a province theyre not about to move into... If you can get the AI to randomly (1 in every 100 games) go with an entire Paratrooper army, and take every VP in a few days...that would be just grand.

Easy should be easy. Hard should be hard.

 

[SteelVolt]

NOTE: this is more of a general discussion and not necessarily indicative of HoI4 AI behaviour (just to make sure no one reads to much into this)

AI is bad. AI has always been bad. AI will always be bad.

Empires Deluxe, after hundreds of people have gone at the open sourced title (AI programming experts) over the past 15 years...the AI is still woefully inadequate.

HOI4s AI will be bad. Its just a fact of life. The AI that controls your units, and the AI that controls the other nations units, will be terrible. (terrible on a scale of a rock, to a human, all AI will fall closer to the rock, than the human)

I can see where you are coming from with this, but am not willing to agree with such an absolute (and somewhat generalizing) statement. Good vs bad are relative terms, particularly when it comes to game AI. Academic AI is interested in actual intelligence while the primary goal of game AI is entertainment. Chess AI, for example: "Current chess engines are able to defeat even the strongest human players under normal conditions." (Wikipedia). While in a sense "intelligent", most players would not be having a lot of fun if they were guaranteed defeat 100% of the games they play. So "good" AI, but "bad" game AI.

At any rate, I would argue that things are good or bad depending on what standard you hold them to. Compared to humans AI still has a long way to go, of course. And comparing between games I feel, as stated, never be fully fair unless the games are extremely similar.

As was mentioned, a 'set piece' to keep the player amused and immersed. Its simply there to provide something to keep the player clicking. Not something that is actively trying to win. Its just there to give you something to interact with, and given some percentage chance for things to happen, to give the appearance of unpredictability (in some games) or of "purpose".

Some games may do non-competitive AI, others do not. How would you know the difference between an AI that actively backs off to allow the player to win and an AI that is just bad at playing the game? As for propability driven decisions; well, a random selection does, by definition (I would argue), lead to unpredictability (albeit not an intentional decision by the AI to mislead their opponent). But (hypothetically): would you be able to tell the difference between a system that "intelligently" selects good options and one that randomly happens to make the exact same decision?

A philosophical question, i suppose: how would you tell the difference (just as with the concept of free will) between an actually intelligent system and one that just appears to behave intelligently? Is there a relevant difference?

All that said. Thank you very much for creating the AI, and sharing this information with us. Were literally foaming at the mouth to get our hands on Hoi4 and start fighting world war two

Noted. And thank you for your input

That was way easy for them to do however, as essentially, a two player game, is chess...with more rules. They were able to take a cheap outsourced AI routine, and plug it into their game. Which still took 5 years to be competent.

If "easy" takes 5 years, I would say those are pretty high standards. Regardless, the percieved quality of AI is also dependent on the game complexity. A typical chess (or general board game) AI has a fairly simple algorithm: build a tree of possible moves, traverse it based on scoring of each possibility and make moves that leads to generally better scores down the tree (very simplified). If "more rules" mean more possible options each turn, the AI will seem to perform less well, because it will not be able to build a large enough tree (as is the case with the game go).
Sadly, memory and processor budgets are often what hold you back.

EDIT: As for your more specific question, I am afraid we are not yet ready to give them proper answers as things are still moving around quite a bit.

 

[General Sandman]

The Battle-Plan-System requires a well performing AI even more than in previous versions. I have to admit that I am pretty much pessimistic about this matter. Letting the AI manage secondary parts of theatres often lead to an unnessecairy loss of formations, which in turn caused much worse conditions to suceed at all.
Here is supposed to be a high frustration-potencial.

 

[Chromius]

AI is always the key thing for me regardless of the genre, but after reading a bit about it, I conclude the current computer instruction sets and the limits of ones and zeroes, that it will take more than the current technology we have to really create a better AI.

It would simply take to many lines of code and too much time for any one person to make a decent AI for a game in the limited time span of a single game development cycle.

 

[fabius]

For me the AI to be 'good' is not straight difficulty level.

It's about doing things that pose credible challenges to me as a player.

e.g.So I eventually defeat an allied invasion in the soft underbelly of Europe; but while the AI was doing it, it provided a decent resource management challenge will the East Front is going on.


The credible part is linking to UI and information the player gets. So in some games EU IV does a good job of this, so we can see why countries dislike or like us in the tool-tips.

I wonder- if the AI uses or will use some pre-mad operational plans, with x% of them using different plans in diferrent situations.


If so, in a game like HoI, there could be some benefit from intel information pop-ups or whatever, giving a reasoning for X AI plan against us. Could be our General offering his opinions. eg. Allies hold their reserves back so this amphib landing is a distraction or probe; Soviets are desperate to relive Stalingrad so have left the north sector weak .... "Attack now".

 

[Hans_Schnitzel]

pre-mad operational plans

If we make our soldiers mad before the offensive they'll fight more effectively, but more recklessly.

 

[tazaaron]

NOTE: this is more of a general discussion and not necessarily indicative of HoI4 AI behaviour (just to make sure no one reads to much into this)



I can see where you are coming from with this, but am not willing to agree with such an absolute (and somewhat generalizing) statement. Good vs bad are relative terms, particularly when it comes to game AI. Academic AI is interested in actual intelligence while the primary goal of game AI is entertainment. Chess AI, for example: "Current chess engines are able to defeat even the strongest human players under normal conditions." (Wikipedia). While in a sense "intelligent", most players would not be having a lot of fun if they were guaranteed defeat 100% of the games they play. So "good" AI, but "bad" game AI.

At any rate, I would argue that things are good or bad depending on what standard you hold them to. Compared to humans AI still has a long way to go, of course. And comparing between games I feel, as stated, never be fully fair unless the games are extremely similar.



Some games may do non-competitive AI, others do not. How would you know the difference between an AI that actively backs off to allow the player to win and an AI that is just bad at playing the game? As for propability driven decisions; well, a random selection does, by definition (I would argue), lead to unpredictability (albeit not an intentional decision by the AI to mislead their opponent). But (hypothetically): would you be able to tell the difference between a system that "intelligently" selects good options and one that randomly happens to make the exact same decision?

A philosophical question, i suppose: how would you tell the difference (just as with the concept of free will) between an actually intelligent system and one that just appears to behave intelligently? Is there a relevant difference?



Noted. And thank you for your input



If "easy" takes 5 years, I would say those are pretty high standards. Regardless, the percieved quality of AI is also dependent on the game complexity. A typical chess (or general board game) AI has a fairly simple algorithm: build a tree of possible moves, traverse it based on scoring of each possibility and make moves that leads to generally better scores down the tree (very simplified). If "more rules" mean more possible options each turn, the AI will seem to perform less well, because it will not be able to build a large enough tree (as is the case with the game go).
Sadly, memory and processor budgets are often what hold you back.

EDIT: As for your more specific question, I am afraid we are not yet ready to give them proper answers as things are still moving around quite a bit.

Nice answers. You must be the smartest one in the office and the rest of them play jokes and make fun of you. Lol

 

[fabius]

If we make our soldiers mad before the offensive they'll fight more effectively, but more recklessly.

Yeah sure, it's called a harangue . And cheaper than Dutch courage for all.

 

[aruon]

Yeah sure, it's called a harangue . And cheaper than Dutch courage for all.

i just use beer. works everytime.

 

[Axe99]

Cheers for all the info @SteelVolt . I saw your comment about memory and processor budgets, and it reminded me of something a whiles back - there was a game I played once where you could choose to select 'advanced AI' or similar, where it would use more system resources (but be slower). Would it be possible in HoI4 (or other PDS titles) to have some kind of 'tick box for more resources for AI, but this will make the game run slower' option? So the people who tick the box don't get a new set of AI algorithms or anything like that, they just get the current system used more deeply? I know there's a risk of segregating the playbase taking this option, and deffo not 'pushing' for it or anything like that, just throwing the idea around, as for me and at least a few others the quality of the AI is probably the single most important feature in the game.

Either way, am sure you'll do a great job of it, looking forward to getting beaten up by your work .

 

[SteelVolt]

Would it be possible in HoI4 (or other PDS titles) to have some kind of 'tick box for more resources for AI, but this will make the game run slower' option?

An interesting idea, but not something I see myself getting any time to look in to any time soon.

Either way, am sure you'll do a great job of it, looking forward to getting beaten up by your work .

Thanks for the confidence. Most of all I am happy to have Wiz to lean on when the going gets tough