How to improve the AI?

[FOARP]

It strikes me that much of what the player is doing when they look at the map and create a plan of attack is shape-recognition - they identify salients, and opportunities for encirclement, based on the shape of the front-line.

Couldn't the AI in a strategy game like HOI identify salients and the opportunities to encircle the enemy by analysing the shape of the front line? And then deploy armour/mobile forces to the places from which it wishes to launch a pincer attack?

I have to admit that I am far from an expert on AI, but it strikes me that if software can easily identify faces in a picture (OK, most of the time at least) then detecting salients, and the location of their "roots" would be far more easy. Applying a template plan-of-attack to the situation would then also be relatively simple (by which I mean doable with a research team and enough time/money, a not necessarily doable by Paradox).

 

[Nasr]

The only real way to "improve" the AI is to level the playing field.

That would be achieved by removing the inherent advantages a human has - such as foresight, situational awareness, AI predictability.

You would do that by reducing the amount of input you can have when moving your forces around for battle - such as eliminating the ability to move around single divisions, corps e.t.c., and instead automating the way battles are fought.

I think this is what they are doing (from watching videos), although I don't know if they are fully automating things.

 

[Axe99]

The only real way to "improve" the AI is to level the playing field.

That would be achieved by removing the inherent advantages a human has - such as foresight, situational awareness, AI predictability.

An AI can have foresight, situational awareness, and need not be predictable.

Situational awareness is the easiest - the AI has access to the same data a human has, and is able to process more of it more quickly. It's just a case of the AI systems used to process the data and how well they work together. I've seen AI in games where their situational awareness has been excellent when compared with a human (for example, Gal Civ II) and atrocious.

Predictability - just have the AI have a number of possible approaches, and have them decided by a random number generator, with percentages based on the commander's characteristics, force mix and doctrine. Then, with a large enough front, you'll probably have more than one option to apply a particular approach, and the human player is left guessing. Predictability should be one of the easier problems to solve.

Foresight - this is the toughest, but hardly insurmountable. At any time, the AI can calculate (with better precision than the vast majority of humans) where enemy forces can be at any point in time. The AI has just as much data as the human has, it again comes down to the systems to process it. However, given the time dimension, this is generally a good deal more complex than straight-out situational awareness.

Of course, all of this takes a lot of work and effort, and there'll be limits on CPU cycles available. But the place where AI can be improved is in the systems managing these things (and others, like production and what-have-you). For me, the #1 thing I'm looking for in HoI4 is improved AI, so have my fingers crossed they've been beavering away at it to make it the best it can be .

 

[Secret Master]

It strikes me that much of what the player is doing when they look at the map and create a plan of attack is shape-recognition - they identify salients, and opportunities for encirclement, based on the shape of the front-line.

Couldn't the AI in a strategy game like HOI identify salients and the opportunities to encircle the enemy by analysing the shape of the front line? And then deploy armour/mobile forces to the places from which it wishes to launch a pincer attack?

I have to admit that I am far from an expert on AI, but it strikes me that if software can easily identify faces in a picture (OK, most of the time at least) then detecting salients, and the location of their "roots" would be far more easy. Applying a template plan-of-attack to the situation would then also be relatively simple (by which I mean doable with a research team and enough time/money, a not necessarily doable by Paradox).

I used to think this, but my analysis of AI behavior when on the attack (and looking at what theater AI requests for assets) in HOI3 keeps coming up with a different problem: the AI can "see" holes and pincers. The problem seems to be that it has really poor force recognition.

The AI will see a hole in the line and strive to exploit it. The catch is that it doesn't understand what these holes mean in the context of surrounding enemy units. The AI is so easy to sucker into traps in HOI3.

I think that the AI in HOI4 will be better in regards to recognizing the threat posed by enemy forces. EU4 and CK2 are better about this.

I probably should also point out that I'm not entirely sure why human players are better at recognizing traps. It may have something to do with experienced human players understanding what a country could produce in time for a certain point in the war, and just assuming that "I don't see it, but it must be there waiting for me."

Of course, it may also be because human players are better about getting intel in the first place than the AI.

 

[winterwatchers]

I wouldn't mind if the AI has perfect Intel on human players to compensate for lack of tactical understanding. I mean most nations ddid have a pretty decent estimate of enemy forces.

 

[Cybvep]

What always bothered me was the fact that the AI was very poor at using special units like MTNs in proper terrain. Terrain modifiers are just simple numbers and terrain is always static. It shouldn't be very hard to make the necessary calculations.

 

[wingo]

AI can be actually pretty clever compared to average human, it can see and tries to exploit weaknessess, even on strategic levels. In my last Italy game I saw the British do amphibious invasions of poorly defended areas and paradrops on undefended islands (Crete, Cyprus), after breaking through their lines Greeks even evacuated almost all their remaining troops to Crete and it was pretty hard to capture it with 10 full divisions guarding it (of course capitol was moved there so no supply shortages). And later when Crete rebelled (did not garrison it, my bad), US even paradropped reinforcements there. While I was distracted by these events and Barbarossa in full swing, UK tried to open the Med by doing simultaneous invasions of North Africa and Arabian Peninsula (storming with MAR and exploiting with MOT), later trying to storm the Gibraltar, which failed because I was ready there.
On the other hand, when UK defended, Gibraltar fell too easily, AI should put higher priority on it. Generally, AI is better at attacking than defending, usually because of lack of foresight (of course you should have reserves ready to defend strategic locations like Gibraltar and Suez, but maybe the AI thought "I have bigger navy than Italy, no worry" just like an inexperienced human would think) and players taking advantage of game design while AI is hardcoded to do certain things, like sending small fleets around Sicily so I can defeat them piecemeal with my deathstacks.

The thing is, AI is not meant to be the godly unbeatable adversary, but a fun challenge but ultimately beatable for 95% of players. That it looks stupid to 5% of players that have hundreds or thousands of hours into the game is only inevitable, those players would beat the average HOI player as easily as the AI, especially when they know exactly when and how the war will unfold and what makes the AI "tick", because otherwise other people would complain if it was too ahistorical. Of course there are weak areas, like choosing the right army composition and using it for its intended purpose (like using arm/mot to create pockets instead of trying to do it with leg inf) or better reaction to new threats (it tends to either overreact or react too late, but to be honest, humans often do that too when surprised). And priorities - do not pull troops from the defense of Moscow because the front in Caucasus is not fully manned north/south, just cut the losses and defend the much shorter horizontal line. To be fair, it almost cut my troops in Caucasus and was only stopped because of lack of troops, and historically if the Italians completely manned the southern front and do as well as the Germans did there (as I did in my game) freeing german troops for the central and northern fronts, historically USSR would probably lose the war too.


To sum it up, I think the biggest reason why AI sucks is game design, if HOI4 will be designed in such a way that ahistorical builds will not be completely overpowered compared to roughly historical ones that will go a long way to help the AI stand a chance. Also, when AI makes a mistake, try to think "Could a human make a similar mistake?" instead of "AI made a mistake, it is stupid!". Because if you have only one AI (the devs said they will not make different ones for different difficulties), making an AI that makes no mistakes is even worse game design than making a stupid one. Because the AI is there to make the game fun, not to beat you senseless repeatedly. It is also pretty historical, everyone made mistakes in war, and some mistakes were really huge which even HOI3 AI would not do.

 

[Laffertytig]

What always bothered me was the fact that the AI was very poor at using special units like MTNs in proper terrain. Terrain modifiers are just simple numbers and terrain is always static. It shouldn't be very hard to make the necessary calculations.

same here. the AI in previous games never factors in terrain penalties which is why it routinely send tanks into cities, mountains etc. it surely wouldn't to be hard to code.

 

[unmerged(748229)]

I would argue that for the HOI series the AI is the most important part of the game. And yet it never seems to work right.

It strikes me that much of what the player is doing when they look at the map and create a plan of attack is shape-recognition - they identify salients, and opportunities for encirclement, based on the shape of the front-line.

Couldn't the AI in a strategy game like HOI identify salients and the opportunities to encircle the enemy by analysing the shape of the front line? And then deploy armour/mobile forces to the places from which it wishes to launch a pincer attack?

Well, first we should talk about the more glaring problems with previous AI:

1) AI tries to stabilize a front by redeploying units... from 1000km away. AI does this repeatedly across a front and weakens it significantly. AI will also send (many) units from the front lines to deal with partisans, instead of using a closer unit that is already 'garrisoning' a location behind the lines.

2) AI will continue sending unnecessary resource/supply convoys despite huge losses.

3) AI does not react to enemy bombing and will send air units to completely useless bases.

I used to think this, but my analysis of AI behavior when on the attack (and looking at what theater AI requests for assets) in HOI3 keeps coming up with a different problem: the AI can "see" holes and pincers. The problem seems to be that it has really poor force recognition.

The AI will see a hole in the line and strive to exploit it. The catch is that it doesn't understand what these holes mean in the context of surrounding enemy units. The AI is so easy to sucker into traps in HOI3.

I think that the AI in HOI4 will be better in regards to recognizing the threat posed by enemy forces. EU4 and CK2 are better about this.

I probably should also point out that I'm not entirely sure why human players are better at recognizing traps. It may have something to do with experienced human players understanding what a country could produce in time for a certain point in the war, and just assuming that "I don't see it, but it must be there waiting for me."

You're right. I think the real issue is recognizing supply routes. If you have 20 divisions surrounded by the enemy and only 1 province through which supply enters, it should give the AI a red flag to either enlarge the entryway or withdraw the units.

 

[Smileyou]

in other words:no real AI exists, only scripted computer response..what i mean is a real artificial intelligence that
would use pattern recognition software and compare it to previous rounds of fighting to analyse and recognize
movements of units, number, strength, IC etc, and come up with an answer to counter the defeat more
effectively..this has sadly not been created in any game..

so all we can hope is for a better scripted so called "ai"..

 

[Karaya 1]

Of course a good human player will be better than the average AI. There are however multiple instances in the paradox games where the AI has ridiculous advantages because it really is too stupid. Like in EU4 where it has intel on everything, but simply pretends not to know most of the time. Or in HoI2 when AI planes have unlimited range (at least DH) because otherwise the AI wouldnt use them correctly. Or HoI 3 AI sending 100 ships to a lvl 1 or 2 port because...well there is no good reason for that.

 

[FOARP]

in other words:no real AI exists, only scripted computer response..what i mean is a real artificial intelligence that
would use pattern recognition software and compare it to previous rounds of fighting to analyse and recognize
movements of units, number, strength, IC etc, and come up with an answer to counter the defeat more
effectively..this has sadly not been created in any game..

so all we can hope is for a better scripted so called "ai"..

To be fair, I think it would probably require a research grant sufficient to commit a development team for a few years to work on it to create something like that. Established companies who can afford to look forward to how things are going to be 5+ years from now can do things like this, and maybe Paradox is getting to that point, but none of the big companies have done anything like that yet.

 

[Laffertytig]

in other words:no real AI exists, only scripted computer response..what i mean is a real artificial intelligence that
would use pattern recognition software and compare it to previous rounds of fighting to analyse and recognize
movements of units, number, strength, IC etc, and come up with an answer to counter the defeat more
effectively..this has sadly not been created in any game..

so all we can hope is for a better scripted so called "ai"..

sounds like skynet to me!

once AI is developed to do the above it will no longer be called AI and we will all be doomed.:blink:

 

[ringhloth]

The best way to build a "better" computer player is, rather than programming the game to be played by a human, program it to be played by the computer. After all, computers can beat humans in simple games. So, the problem with making a computer play a game like HoI well is that every layer of complexity will decrease the ability of the computer to compete with humans. There's a balance, then, between trying to make an interesting game, and trying to make a interesting opponent. Especially since consumers so rarely comprehend that interesting games make for boring opponents.

 

[Axe99]

The best way to build a "better" computer player is, rather than programming the game to be played by a human, program it to be played by the computer. After all, computers can beat humans in simple games. So, the problem with making a computer play a game like HoI well is that every layer of complexity will decrease the ability of the computer to compete with humans. There's a balance, then, between trying to make an interesting game, and trying to make a interesting opponent. Especially since consumers so rarely comprehend that interesting games make for boring opponents.

I'd actually argue that it's more the case that the more complex the game, the harder it is for the humans designing the AI to do it in such a way that it can provide an interesting, plausible opponent, because instead of emulating a simple system (say checkers), the code needs to reflect a whole bunch of interconnecting systems, with a whole bunch of variables. I'm not saying anything about the capabilities of the dev team here (from what I can see they're doing an excellent job) - but it's not the issue that the computer can't cope, but that as humans it's a whole other magnitude of issue to design AI that deals with so many complex systems. Of course, at the end of the day, it's the same result, but I'd bet good money that as AI design continues to improve, things will continue to get better .

 

[haardfyr]

If they would introduce command cycles as seen in command ops games, the ai would improve drastically as the human player wont be able to constant react. That and a deeper fog of war.

 

[No idea]

If they would introduce command cycles as seen in command ops games, the ai would improve drastically as the human player wont be able to constant react. That and a deeper fog of war.

Agreed. The order delay works wonders to lever things between the ai and the player (unless you are really good at the game, and, even then you fail from time to time). Thta and the fact that your units are not, in general terms, above those of the ai because you research better and make better builds.

 

[Secret Master]

Agreed. The order delay works wonders to lever things between the ai and the player (unless you are really good at the game, and, even then you fail from time to time). Thta and the fact that your units are not, in general terms, above those of the ai because you research better and make better builds.

Isn't this partially what battleplans are meant to address?

You can react instantly if you want, but the bonuses from the battleplan mean you really need to weigh your options.

 

[No idea]

Isn't this partially what battleplans are meant to address?

You can react instantly if you want, but the bonuses from the battleplan mean you really need to weigh your options.

Yes, if i understood battle plans correctly they will stop players from reacting instantly after they are aware they have commited a mistake or lose the battleplan bonus. It is not exactly as orders delay, but it is better than nothing.

 

[BarrosRodrigues]

It strikes me that much of what the player is doing when they look at the map and create a plan of attack is shape-recognition - they identify salients, and opportunities for encirclement, based on the shape of the front-line.

Couldn't the AI in a strategy game like HOI identify salients and the opportunities to encircle the enemy by analysing the shape of the front line? And then deploy armour/mobile forces to the places from which it wishes to launch a pincer attack?

I have to admit that I am far from an expert on AI, but it strikes me that if software can easily identify faces in a picture (OK, most of the time at least) then detecting salients, and the location of their "roots" would be far more easy. Applying a template plan-of-attack to the situation would then also be relatively simple (by which I mean doable with a research team and enough time/money, a not necessarily doable by Paradox).

The AI can´t handle terrain in HOI III and is barely functional in EU IV with this in mind what do you think are the odds of something like that happening in HOI IV? It is not that PDS does not know how to do it, they can do it if the people in charge let them.

The only real way to "improve" the AI is to level the playing field.

That would be achieved by removing the inherent advantages a human has - such as foresight, situational awareness, AI predictability.

You would do that by reducing the amount of input you can have when moving your forces around for battle - such as eliminating the ability to move around single divisions, corps e.t.c., and instead automating the way battles are fought.

I think this is what they are doing (from watching videos), although I don't know if they are fully automating things.

As EU IV clearly shows that is not enough; the AI needs to receive much more love to be more than a brain-dead pushover.

EU4 and CK2 are better about this.

Yes but only because 99,9% of the situations it is impossible to encircle and destroy in those games. Whenever there is a chance to trap AI regiments then it always falls for it because the AI in EU IV is as brain-dead as in HOI III TFH.
The AI in HOI IV could be much better if it receives the necessary love from PDS, at least it should be easy to make it better than in EU IV because the AI is barely functional in that game; it has so many flaws that it falls apart almost instantly in most situations.


Sorry if I sound too negative but this games AI needs some serious love from the devs because eye candy and features that only the player can use does not cut it and is becoming old rather quickly.