Is it really that hard to program AI

[Sverd]

That is marginally more intelligent than a braindead skunk? The AI just pulled out 3 divisions in a defensive line for the second time in a row while sending volunteer corps to aid Japan's wars, resulting in my 5 divisions of panzers getting encircled and killed. Again. Because the AI doesn't seem to be intelligent enough to realize "Hmmm, if I pull out every division guarding this advancement line, then the other side can literally just walk in and encircle the massive tank force that's been single handedly winning this war due to my complete inability to fight".

Absolutely infuriating. If you're going to make the AI this ♥♥♥♥♥♥♥ braindead, at least make the enemy AI braindead enough to not realize that they can encircle and massacre my volunteer forces when my allies decide they don't like having my volunteers in their country.

[flyboymb]

Postat inițial de foolmanchoo:

We eagerly await your mod that easily fixes the AI.

So if you buy a brand new car and it has a major defect that requires a recall you're going to go ahead and not only engineer a fix but also give it out to everybody for free?

See how dumb that sounds in the auto industry? Just why do you think it suddenly sounds smart in the video game industry?

[Sandvich]

Postat inițial de flyboymb:

Postat inițial de foolmanchoo:

We eagerly await your mod that easily fixes the AI.

So if you buy a brand new car and it has a major defect that requires a recall you're going to go ahead and not only engineer a fix but also give it out to everybody for free?

See how dumb that sounds in the auto industry? Just why do you think it suddenly sounds smart in the video game industry?

OP said that making AI is super duper easy.

He could probably do it in less than 10 minutes, right?

[The Doctor]

Postat inițial de Bored Peon:

Is unit attached to a army? If yes, then find new unit.
Is unit unit assigned to a front? If yes, find new unit.
Is unit unnassigned? If yes assign to expedition

Simple if then statement.

When I send expeditions and volunteers I have to manually detached them form existing armies, the AI should not be skipping that requirement.

I'd appreciate it if you fixed your post so that the quote is not falsely attributed to me. [Edit to add: boy, quoting your post in fuill really screws up the quotation function. ;)]

Second, yes, to us, and that includes me, who know next to nothing about writing AI algorithms to manage every single playable faction in the game world in real time and that it doesn't slow the program down too much, it appears to be a simple thing to do, doesn't it? It's very easy to type that out and think that you've solved the problem.

Finally, and not to you but to the discussion in general, I believe that the AI could be improved by giving it better division templates to work with, something that modders have already done. I see no good reason why Paradox can't look at these mods and code the features that most benefit AI play into the game themselves. However, with respect to other improvements that modders have made, I understand they don't want to railroad the AI too much as they want the game to be a sandbox.

[The Doctor]

Postat inițial de GoldenTalon:

Postat inițial de flyboymb:

I think you guys are blowing this up to make your argument valid. There were computer chess games in the 90s that could be programmed not to make a stupid move that would put it at a disadvantage.

How difficult would it truly be to add in a line of code saying that if the AI's units are on the border with a faction they are at war with, they are not utilized as volunteers. If all units are on the border, no units will be volunteered.

He's asking for the 3 laws of robotics, not the programming of the grand totality of human ethics.

And before anybody else says 'why don't you do it yourself', Paradox hasn't waved a check in front of me.

The devs decided to add in all these complex mechanics. That was there decision. Now it's too complicted for an AI to handle. That's their fault. Bottom line is AI is braindead.

I'd argue that HOI 1 was too complicated for the AI to handle as the AI was easy to beat there too. It wasn't any better in HOI 2 either but it definitely got a whole lot worse when they released HOI 3 because they increased the number of provinces the AI had to consider by a very large amount. The AI in HOI 3 TFH is still as bad as HOI 4's. Yes, perhaps there are some things the HOI 3 TFH's AI could do better but I've played every iteration of this franchise and not one of them had a decent AI.

Would any of us have bought HOI 2, 3 or 4 if it had no new, complex mechanics? It hasn't stopped us in the past and it won't in the future because quite a few folks play the game with other humans while others use the sandbox to role play and have adventures. The BEST thing TFH brought to the game was allowing the AI to manage our units for us. IMO, the battleplanner is the one feature that redeemed the franchise for me. If I control my units myself, as I did in HOI 1 and 2, I can beat the AI on the highest difficulty with ease.

There is no brain to be dead either. The AI is just smoke and mirrors, the illusion of an AI opponent. It's not thinking at all and any time it accidentally does something that we think is clever, it's just a random event. Like all RTS AIs. once you know how it works, it's a joke to play against.

And yet, funnily enough, they've made it good enough to challenge folks who are new to the game. There's often a thread on the first page here created by someone who says that the AI beats them whatever they do. Then they get advice from folks who pull back the curtain to reveal the reality of the Wizard of Oz and then the AI is easy to beat. The AI does actually manage many of the systems well enough to teach us the game.

[Sverd]

Man people got mad at this post. I like paradox and their games a lot, but I guess the paradrones meme is real. When I've paid nearly a hundred dollars for a product that's been in development for nearly half a decade, I expect the AI to be capable of not consistently making basic, massive tactical mistakes like being unable to comprehend that units attached to their armies will get encircled if they decide to refocus their entire offensive effort with no indication to their allies.

[Ryan]

Consider the level of effort and years spent on making a good chess AI. It still it can lose. How much more complicated is HOI4 to chess?

[The Doctor]

Postat inițial de Turnop:

Man people got mad at this post. I like paradox and their games a lot, but I guess the paradrones meme is real. When I've paid nearly a hundred dollars for a product that's been in development for nearly half a decade, I expect the AI to be capable of not consistently making basic, massive tactical mistakes like being unable to comprehend that units attached to their armies will get encircled if they decide to refocus their entire offensive effort with no indication to their allies.

Have you played either of Gary Grigsby's 'War in the East/West wargames? They have good AI opponents that can sometimes recognise these tactical mistakes. Of course they are turn based and focus on one theatre of operations and the AI has just one side to manage. Even so, the AI takes a few minutes between turns to resolve its turn.

We all WANT Paradox AI to be as good as the AI in those games. However, it's a much bigger challenge for the AI programmer to manage. I want them to improve the AI as much as the next guy. However, my expectations are tempered by my admittedly limited knowledge of strategy gaming AI programming.

[The Doctor]

Postat inițial de Ryan:

Consider the level of effort and years spent on making a good chess AI. It still it can lose. How much more complicated is HOI4 to chess?

Using Chess AI as an example will usually earn you ridicule when you do so but we seem to be a more forgiving crowd here. Chess has 64 squares and 32 pieces that move in a fixed manner and the rules are ridiculously simple compared to a simple computer strategy game. No diplomacy, production, unit design etc, etc, etc. It's so bleeding obvious that it's bizarre that we need to say it. Plus, chess is not a RTS game either. It's turn based.

BTW, the chess program AI I play against from time to time hasn't lost a game against me for quite a while. :D It's good enough for me.

[flyboymb]

I used chess as an example as a way to 'teach' a program basic rules of a game. If HOI4 were running a chess program, it might randomly charge the queen into a position where it could be taken by 6-8 pieces the next turn with no other gain made. The AI needs to learn the 'common sense' rules of the game in order to better utilize the complicated aspects.

The volunteer issue is something that could be fixed for all factions at once. If there is a way to write code that is universal to the game, put it there. Otherwise copypasta to each faction. Again, 'thow shalt not' rules are much less likely to have a butterfly effect compared to rules that entice the AI to behave in a particular way. You're merely closing one door and keeping the others open.

What is likely at issue here is that this is considered to be an extremely minor issue compared to the other issues that need to be fixed in this game. Considering the patch release rate, I doubt they will ever get to polish issues like this when it is much more profitable to fix the squeekiest wheels.

[AlexMBrennan]

Because it was not AI at all, just prescripted algorithm

That's sophistry - the process programmers use to make computer controlled players act with something approximating intelligence is called "artificial intelligence"[en.wikipedia.org].

Whether that happens to consist of a couple if-then-else statements or a deep neural network is immaterial - what matters is that the result can pass for something a human player might do. HOI4 does not succeed.

[comsubpac]

As far as we know they have one guy working on the AI part time. So we get what we pay for I guess.

[markuslarsson81]

Not to put words in people's mouths but I think many people are frustrated about the obvious fact that Paradox puts more resources into creating new content/dlc so they can sell more instead of improving the game which people expect nowadays.

I mean I saw all the dev diaries for Man the guns and it locked like massive changes were made and I pre-ordered it, I just now completed a full campaign and I gotta say the feels mostly the same as 1.5. I don't think the fuel mechanic changes much in terms of approaching the game nor the other features.

[Bored Peon]

Postat inițial de markuslarsson81:

Not to put words in people's mouths but I think many people are frustrated about the obvious fact that Paradox puts more resources into creating new content/dlc so they can sell more instead of improving the game which people expect nowadays.

Every DLC came with improvements to the game, and about half the improvements were also included in the base version.

Postat inițial de markuslarsson81:

I don't think the fuel mechanic changes much in terms of approaching the game nor the other features.

It makes a huge difference for some countries. Some countries need to invest in refineries and oil prospecting just to keep the fleets floating and their planes flying. Take away their fuel reserves and watch em crumble.

Also with the removal of oil to build ships, tanks, planes, etc it allows a much more flexability in what you can build.

[GoldenTalon]

Postat inițial de comsubpac:

As far as we know they have one guy working on the AI part time. So we get what we pay for I guess.

No more. Podcat admitted on their Reddit event last week that they have NO AI programmers anymore.

[Fatbill]

A better question, is why does the AI in HOI4 suck giant balls, when the AI in HOI3 and HOI2 were not nearly as bad?
Yes, different game, different programming langue and all that, but the decision trees at the core of AI gameplay should not have been much different.
Or at least not the extent of how terrible and stupid the AI in HOI4 is.