Hearts of Iron 4 announced


Yeah. I do commend the team for wanting to make the AI smart enough instead of relying on scripted behavior or cheats, but sometimes you just need to take the brute force approach. Maybe it’s not elegant and it’ll offend your programmers’ sensibilities, but sometimes you just need to hack something in while you search for the Perfect Solution to the AI.

I appreciate the CPU concerns and I appreciate that it’s a difficult problem they’re still looking at, but just throw in a dirty hack of a fix until then. It’s not like this sort of thing can have a minor impact on the game or is a weird edge case.


Somewhat serious question. What if they were to offload the AI from the player to WOPR? For a nominal monthly fee players can dial into WOPR and play against AI that does not have limitations imposed on it because of WOPRs nearly unlimited CPU power and memory. WOPR will offer indistinguishable from human AI so players can have challenging games that don’t require the usual hostels of playing against other players. With a WOPR subscription you can save your games and come back to them anytime you like and at any hour. WOPR is tireless and available 24/7/365 excepting downtime for maintenance.


Is WOPR a real thing or are you asking why they don’t use the fictional computer from the movie Wargames?

I think cloud-based AI for a real-time game would be a significant technical challenge.


The expert AI mod author shares my opinion about the downsides of spurning scripts: http://steamcommunity.com/workshop/filedetails/discussion/741805475/360672137533958058/

[quote]This mod implements an AI in the fashion that I always felt would work best in the HOI series. It is a mix of predetermined guidelines for the AI with enough freedom for the AI to adapt to most situations. This is in contrast to the vanilla AI that essentially has a ‘one size fits all’ dynamic approach to the game. The issue with that approach is that the AI doesn’t and likely never will understand the subtleties of proper strategy that depends heavily on each specific country, it’s situation and the current meta.

This mod replaces several vanilla AI functions partially or completely with event script based systems that run on the background for all AI nations which can create highly specialized and efficient strategies for each country with the least amount of cheating possible (parts where it cheats are listed above). Other aspects of the AI are handled with heavy use of AI strategies to control its behaviour or other tricks.

Generally speaking when playing this mod, you can expect the AI to be twice as effective with its entire economy. A good management of laws, research, advisors, construction and occupation laws means the AI has as much as double the military factories AND the resources to support it at certain points in the game compared to vanilla (not to mention the massively increased lend-lease between AIs that helps them with deficits). It will put this into good use by building effective divisions and aircraft. Countries like SOV and GER have armor production many times higher than in the vanilla and the AI will produce up-to-date and upgraded aircraft in massive quantities.[/quote]


If it’s as advertised, that sounds great. I’ll have to try it myself for a few games to see how close reality is to intent. I vastly prefer dynamic AI instead of scripted, but I don’t think being a purist about it is helpful and that’s what the HOI4 team seems to be.


The lack of specific scripts to solve specific AI problem areas is indeed baffling. There’s no way a single dynamic algorithm is ever going to play a game of this complexity competently on today’s top hardware (and even getting an algorithm that could play on a supercomputer is a really difficult task).

I understand the lack of script is due to difficulty of scaling it as the game develops (the game is due for a lot of changes over it’s lifetime, and maintaining those scripts so they keep being relevant would be a significant workload) but they are trying to solve the problem from purely an engineering standpoint, instead a game design one, and that has a lot of chances of not working out.

And besides, even a dynamic generic, one size fits all algorithm is going to need adapting as the game changes (and even if it doesn’t given the current issues).

Does anybody know what paradigm are they using to building the AI?

I think nobody is talking about a purely scripted AI (that would be madness) but about specific scripted solutions to common tasks that make sense. You are going to have scripting at some point of the AI layering, no matter what, even in a purely dynamic AI (unless they have HAL 9000 as their AI engine), but right now it’s probably just at the very lowest level and they probably need to raise it a couple of notches on some instances.


So the next DLC will release in a couple of weeks. I figure I’ll try out HoI4 again in a month or so, once the expert AI author has had time to make use of the new features.

I guess on some level it makes sense for the Paradox AI dev to focus on fixing stuff that’s not moddable, rather than finding the right scriptable settings for competitive AI. If they have talented people willing to do the work for free, why not make use of it? Though I wonder if it wouldn’t make sense to basically import the AI fixes from a mod like this into the base game and pay the author a decent lump sum. Most of the people who have tried the mod swear by it.


This is my angry face on.

To generalize , in order to please the very loud anti-ai cheating crowd, they have made an AI that just can’t play the game at all. In other words, this game is meaningless in SP if you like a challenge…unless you like minors…

Always struggled to understand the anti-cheat crowd, I guess they just don’t like to loose.


I think it comes from a fundamental misunderstanding on what the AI role is in a game’s design. Which is not to simulate a human opponent but to put forward an entertaining opposing force in games that rely on more or less symmetric sides (nobody criticizes AI in shmups).

I see were they come from, though. If the bonuses and cheating creates suspension of disbelief or invalidates otherwise valid tactics mechanic-wise (like an ever seeing AI not allowing for surprise attacks) then the AI is failing to be entertaining. But that has little to do with it cheating or not. The perception of it being related to cheating is because game AI cheats too frequently that most times an AI fails the suspension of disbelief/smoke and mirrors test is related to a cheat. What they don’t realize is that those times the AI really works it’s probably also cheating.

Nobody likes playing against a supercomputer-driven state of the art chess program (unless you are a grandmaster) and that’s not cheating either.


The devil’s in the details. “Cheating AI” can mean lots of things. If they get some extra resources and stuff to make up for them not being as smart/efficient as a player? I’m fully on board.

The problems for me arise when the AI players are playing a fundamentally different game than everyone else. When GalCiv3 came out, I was really annoyed by the full map vision the AI had from the start. I’d be getting carpet bombed by starbase constructors and colony ships right at the word go. Meanwhile, I’m having to build scout ships to find the resources/planets, then build the constructors, then colonize.

Or if the AI can ignore strategic resource requirements and just miracle up armies/fleets/units as needed, that eliminates a lot of the strategy of a game. You can’t starve them out, cut them off from resupply, etc. The game is worse due to the AI cheating.

HOI4 isn’t opposed to a cheating AI per se. They added those sliders you can use to give varying levels of bonuses to the different countries. What they’re adverse to is allowing the AI to break fundamental game rules or hard-coding behaviors. I am in agreement with that philosophy… up to a point. I think they’re just being overly dogmatic about it because they’re looking for the perfect solution. They want the AI smart enough so that it understands not to do this things without being explicitly told not to do it, because that will make for a much better game and the AI can respond to similar situations in other regions (as opposed to a hardcoded rule regarding the English Channel situation).

The problem I have is that such a goal might not be attainable. Or if it is attainable, it could likely be a long way off. I think they need to be a little more flexible in the meantime because it would improve the game as it currently stands. It’s a shipped product that people are playing, they can’t act like this is a project that is still behind closed doors that they are perfecting.

Still, kudos to them for providing the modding tools and a flexible framework for customers to change the behavior to their liking via mods. A lot of my disdain for Civ6 could be alleviated if Firaxis had done the same.


I once played with two sets of chess pieces against a guy with 2300+ Elo rating, when I was like 16, it was a tense grueling match for both, neither cared one bit that we broke the rules…as it was still pretty even match …



I am cautiously optimistic that AI, after taking a step backward in 1.3.3 will be quite a bit better in this patch. It sounds like the AI guy (SteelVolt) spent a fair amount of time looking on what the AI actually produced and made changes to how the AI uses templates. He also made it possible for modders to differentiate how Garrison troops and armor units are used.

Production license are a nice addition and should make minor more fun as well providing a source of income (aka civilian factories) for countries like Germany and some other majors.

I’m not a big fan of minors, but I think I’ll give Yugoslavia a shot after the expansion is out.


$18 for the bundle is kind of a no-brainer for me.


The good news is that Paradox is coming around. For instance, they started with trying to teach the AI about attrition in places like Africa after a few patches they simply did what game designer have done for decades and made areas like the Quattro depression impassable. Likewise in the patch after trying to get the AI make generalized improvement to templates they got smarter and made some pre made templates and now they AI is going to have explicit orders to treat garrison units, Marines, and Armor as different divisions. This will also be good for modders.

Still reading the HOI forums is a good reminder why I’m glad I’m not in the game business. People are more upset about the AI in this game, than most of us are for Civ VI. Now it is true that AI certainly fails at many things in HOI IV, but it is also succeeds more often than in fails. Unlike Civ VI which literally can’t take over a city.


The only thing that really bugs me about the AI in HOI4 is its repeated effort to launch D-Day in 1941 that the Germans easily throw back into the sea. Sometimes they get lucky and establish a small beachhead in Italy and then you get these big announcements like “Rome Falls to the Allies” like the end of the Reich is just around the corner, but then 50 Axis divisions roll over them.


Can you win a civ 6 game on a huge map on immortal?

For clarification, I can win but it’s hard, and you can easily loose…

In hoi 4 on hardest punishing myself and boosting everyone else, I win without breaking a sweat with Germany …pointless to play Soviet, UK is a tad fun, but I can stop them in France…I guess Poland usually is too hard , but I’ve held out quite a long time…


I know they looked into this but, it doesn’t seem liked they’ve been able to fix it.

Of course, in many ways it is really really hard problem to fix because there is no easy answer. I have had the allied invade with 20 plus divisions and by the time I’d notice Bremen and the Keil canal or Hannover were captured and I had to respond with training divisions. They invasion ended up damaging lots of factories and generally screwed up stuff but was it worth losing 20+ divisions probably not. I’ve also seen the all to common attempts to invade a fairly well garrison city added a reinforcement division or two and watched the troops die on the land. It is hard to fault the AI in the first case because if the AI isn’t somewhat aggressive about invading the Axis simply won’t do much in the way of garrisoning the France, and the lowlands or Italy, which makes it easier to conquer Russia.

Over the decade’s I’ve played many game of 3rd Reich and World at Flames with experienced gamers, and more often than not the first allied invasion is repeal and often so is the second.


I’ve never tried and since I literally fall asleep on standard size map at either Immortal or Emperor. I’m sure I’d die of boredom before I finish it… (I don’t like playing at immortal because you are almost forced to use the military which I find really boring )

I agree with the current patch the AI isn’t good and Germany is too easy. On the other hand with the previous patch I had challenging games, as US, India, and Japan (with max boost to the other guys). I couldn’t hold out as Poland either.

I mostly play Germany (it is my test case when they first release a patch and so I’ve got better at playing it than other countries)


Ironically a failing naval invasion is one situation where suddenly abandoning the front and shipping all the troops to a low-priority theater would actually be a good move!

Yeah I don’t fault the AI for failed naval invasions. It’s a tough problem. When I do an invasion I try to find regions I can cut off and defend a narrow front against the inevitable counter-attack. But there are so many factors for the AI to consider in this case, and besides, what German player doesn’t want to throw the Allies back into the sea?

On the other hand, blithely pathing land division transports through or near regions completely dominated by hostile naval forces seems like something that should be easier to solve, even if a rough solution results in some missed overseas reinforcement opportunities.


It just needs a few lines of code:

<GERMANY at war with USSR>
<if YES: launch DDay>
<if NO: put the kettle on>