An argument in favor of certain scripted tendencies.

[Dalwin]

I think it is fairly well accepted at this point that the AI would do better in its production and research etc. if some of its tendencies were scripted. I think there is no strong argument in favor of the current one size fits all approach to such things.

The Devs strongest and almost only argument against that is their perception that a scripted AI is more predictable and therefor in the long run a less challenging opponent.

The solution seems relatively obvious. For each of the majors make 4 or 5 scripts for an overall production and research strategy (including some template preferences). Have the AI choose one semi-randomly at the start of the game. I say semi-random since some would have a higher weight and be more likely to be chosen than others.

The other argument against is the time it would take. In my opinion this would not take as long as many other changes would. It is also not the highly difficult type of AI programming and would likely not even require @SteelVolt 's direct attention.

You could take that a step further by making say 3 scripts for the minors and depending on the nature of the country assign them to one or the other (no randomness). Countries like China for example should be taking a far different approach to their overall scheme than should a country like Hungary.

 

[permanently_afk]

I think it is fairly well accepted at this point that the AI would do better in its production and research etc. if some of its tendencies were scripted. I think there is no strong argument in favor of the current one size fits all approach to such things.

The Devs strongest and almost only argument against that is their perception that a scripted AI is more predictable and therefor in the long run a less challenging opponent.

The solution seems relatively obvious. For each of the majors make 4 or 5 scripts for an overall production and research strategy (including some template preferences). Have the AI choose one semi-randomly at the start of the game. I say semi-random since some would have a higher weight and be more likely to be chosen than others.

The other argument against is the time it would take. In my opinion this would not take as long as many other changes would. It is also not the highly difficult type of AI programming and would likely not even require @SteelVolt 's direct attention.

You could take that a step further by making say 3 scripts for the minors and depending on the nature of the country assign them to one or the other (no randomness). Countries like China for example should be taking a far different approach to their overall scheme than should a country like Hungary.

Hi Dalwin,

I'm currently in a hurry, but the counter-argument boils down to:

1. Maintainability
2. Going around the problem instead of through it (though, there is nothing inherently bad in this approach)
   
3. Narrow-band solution, providing only for a set parameter of circumstances. Once you leave these circumstances, the thing breaks completely
4. Exploitability

That said, partially (extremely narrow) scripting the AI as a stop-gap measure would be something I am in favor of. Although, given this is software, we would run the risk of the band-aids to become permanent additions.

I'll expound if I find more time. Questions?

 

[Alex_brunius]

The solution seems relatively obvious. For each of the majors make 4 or 5 scripts for an overall production and research strategy (including some template preferences). Have the AI choose one semi-randomly at the start of the game. I say semi-random since some would have a higher weight and be more likely to be chosen than others.

The main issue I see with these scripts that is unless they impact and are tied into the overall strategy of the AI and pretty much all parts of the game, they wouldn't be very useful, or might even make the AI significantly worse.

Let's for example say that one of the German production, research and "template preference" scripts is preparing for Seelöwe. ( A classic alternate history what if ).

So the AI will research marines, airforce and naval techs, and produce a stronger airforce with naval bombers, more surface fleet as well as make a marine division template.


But unless the AI also changes it's construction strategy in 1936 to build more dockyards, and also changes it's NF strategy to not attack Soviet, and also changes it's battleplans to invade UK, all these other changes would result in nothing except a weaker Germany which due to it's lack of tanks and appropriate forces get steamrolled when it reach the NFs and is making it's historical Barbarossa...

It might see some initial success in the pripyat marshes due to it's Marines, but as soon as the German marines fight Soviet Armor on the plains they fold like a house of cards, and naval bombers sure as hell won't help them.



Now if your proposing making a script which ties into all parts of the game and all aspects of the AI I say that might start to get interesting, but can it be done and is it easier to do then making an AI using proper weights instead to evaluate each situation? I don't know enough about AI development to say either way honestly.

 

[Misaka_Complex]

So the AI will research marines, airforce and naval techs, and produce a stronger airforce with naval bombers, more surface fleet as well as make a marine division template.

Or maybe the German AI should just learn to mass fighters, CAS and paratroopers and take a British port properly, and then send the rest of their army in. Air superiority is also gonna help them later against the Soviets.

 

[Alex_brunius]

Or maybe the German AI should just learn to mass fighters, CAS and paratroopers and take a British port properly, and then send the rest of their army in. Air superiority is also gonna help them later against the Soviets.

The only reason this exploit works is because Navy can neither blockade supply to a port nor blockade divisions transporting when they are spread out since the entire Royal navy get stuck in weeks of combat with a single convoy or division and let all other convoys and divisions happily sail by ignoring them. Once devs fix these issues such a strategy would fail as miserably as it would historically.


Regardless your missing my point by getting hung up on the specific example I used...

My point was that scripted AI priorities in production, research and division design needs to be connected to all other parts of the game or they can end up making the AI weaker.

 

[jpd]

And, of course, there is the little elephant in the room called "other nations". It's very well for your germnay Ai to say "today I go for alternate history variant seelowe", and then follow the script to accommodate that. But what if other nations around you won't play ball, like the Soviet AI refusing a molotov-ribbentrop pact? If you're railroaded on a script you can't properly react to this.

 

[Alex_brunius]

And, of course, there is the little elephant in the room called "other nations". It's very well for your germnay Ai to say "today I go for alternate history variant seelowe", and then follow the script to accommodate that. But what if other nations around you won't play ball, like the Soviet AI refusing a molotov-ribbentrop pact? If you're railroaded on a script you can't properly react to this.

Very well put. Or even worse, if the Soviet isn't an AI at all, but a player that anxiously awaits the historically scary German invasion...


Another important thing that is needed here is to display and communicate the AIs intention somehow ( when appropriate ).

As a player without understanding that the German AI aims for Seelöwe I would say that it's just a very bad AI when I fight it's weak land army ( regardless of how well prepared for Seelöwe it was ).

If my Espionage efforts instead told me that they suspect Germany is preparing to invade UK my perception of the AI becomes totally different. Instead of cursing the bad AI doing things I don't understand I make plans to send Lend Lease in the other direction to help UK hold while I plan to stab Germany in the back when I see that they attempt to land and gain a foothold in UK.

By such a simple thing as informing the player about AI intentions my perception of the AI went from "very bad AI" to interesting AI which does alternative things that I can adapt to and use to win. In this case it's perfectly fine that the German AI was weak on land because I understood it's intention to be strong at sea and in the air instead.

 

[Misaka_Complex]

My point was that scripted AI priorities in production, research and division design needs to be connected to all other parts of the game or they can end up making the AI weaker.

And as you said so yourself, an AI Germany scripted to train more marines and focus on a navy is going to get rekt hard by Soviet tanks, not to mention that its going to take years for Germany to even have a chance against the Royal navy if we were to put it your way and assume a historical scenario.

As of current mechanics the easiest method is the paratrooper method for the AI because due to how the game works right now, you can't blockade supplies or entire armies from landing even if you max out the convoy interdiction doctrine. Whether if its a " bug" is another question, but that is how the game works right now. Also if you split air wings into groups of 20 or less you will find that they hit ships much more efficiently than groups of 200 or bigger. Thus this eliminated the need for Germany to make naval bombers because CAS does just as good as a job and can also be used later against the Soviet Union.

 

[Alex_brunius]

Since you done it twice yourself now, I will do the same and split out a small part of your post and make a response which has nothing to do with the topic at all...
( Maybe it will teach you not to do this yourself in the future ).

Also if you split air wings into groups of 20 or less you will find that they hit ships much more efficiently than groups of 200 or bigger. Thus this eliminated the need for Germany to make naval bombers because CAS does just as good as a job

In one of my MP games my wing of 300 naval bomber just oneshot an enemy Carrier in the first hour of navalcombat. CAS can not do the job anywhere near close to as good as NAV can.

 

[Misaka_Complex]

I actually very much enjoy it when you take a small part of my post and talk about specific sentences which I mentioned, so thank you and please keep doing that since I appreciate it.

In one of my MP games my wing of 300 naval bomber just oneshot an enemy Carrier in the first hour of navalcombat. CAS can not do the job anywhere near close to as good as NAV can.

When I start the battle of Britain in my MP games and split by CAS into groups of 25 they actually have multiple engagements sinking several capital ships in the first hours of the battle of Britain. But hey I'm sure that how smaller wings work better than larger wings of like 300 are certainly another bug in the game, right?

 

[Alex_brunius]

Yes let's have Paradox teach the AI even more exploits that needs to be fixed ASAP anyways so they break the AI again... What could possibly go wrong?

I'm also sure that an AI using exploits only the top 1% of MP players abuse won't be frustrating for new players at all...


If an airwing of 25 planes can sink a capital ship in 1 hour something is obviously broken, can you point to a bug report with savegame so they can fix it?

 

[Misaka_Complex]

Yes let's have Paradox teach the AI even more exploits that needs to be fixed ASAP anyways so they break the AI again... What could possibly go wrong?

The only part that could go wrong is that the AI is gonna be too strong as it adapts to how it maximizes efficiency, but obviously it can't go wrong since the AI is gonna get a lot smarter meaning that a lot less people are going to complain about the AI being "too weaK' or their SP games being "too easy" .

 

[Alex_brunius]

The only part that could go wrong is that the AI is gonna be too strong as it adapts to how it maximizes efficiency, but obviously it can't go wrong since the AI is gonna get a lot smarter meaning that a lot less people are going to complain about the AI being "too weaK' or their SP games being "too easy" .

No...

What happens when the AI is "stronger" is that people complain that the mechanics are broken instead.

Remember for example when the launch AI port striked all poor players German starting fleet into oblivion? Mechanics were changed so it's no longer possible to do ( and the mechanics were changed in a rushed hacky bad kind of way as well with a hard blocker to port strikes which is just bad ).

Thus teaching the AI exploits would just be counterproductive since it would make people complain the mechanics are broken and demand they are instantly fixed ( with rushed fixes that also break the AI ).

 

[Misaka_Complex]

What happens when the AI is "stronger" is that people complain that the mechanics are broken instead.

Ah yes, you make a fair point here as I remember how powerful port strikes were back in vanilla. So what do you think would be the best way to change port strikes? Also this means that people will always have something to complain about regardless of whether if the AI is "strong" or "weak." I agree that there are a lot of "exploits" to be fixed in the game to make things balanced, but then again its almost an year after release and the game still has all these kinds of problems...

 

[Alex_brunius]

Ah yes, you make a fair point here as I remember how powerful port strikes were back in vanilla. So what do you think would be the best way to change port strikes?

The way all bomber missions should work.

Strong fighter, radar and AAA presence should reduce their efficiency / increase disruption down to reducing their damage with almost -99% ( and this feedback should be shown in a good way to the player ).

That way the bombers still fly and try to bomb, but do almost no damage and get shot down in droves as a thanks.

This change would also fix the big issue with CAS we have in multiplayer and improve Strategic bombing balance at the same time.

 

[jpd]

well, radar presence in and of itself should not reduce enemy bomber efficiency. Radar only detects. It does not shoot at anything. What it instead should do is increase the own fighter (and possibly AAA) efficiency against said bombers .)

 

[Alex_brunius]

well, radar presence in and of itself should not reduce enemy bomber efficiency. Radar only detects. It does not shoot at anything. What it instead should do is increase the own fighter (and possibly AAA) efficiency against said bombers .)

Yes that's what I meant

 

[Misaka_Complex]

The way all bomber missions should work.

Strong fighter, radar and AAA presence should reduce their efficiency / increase disruption down to reducing their damage with almost -99% ( and this feedback should be shown in a good way to the player ).

That way the bombers still fly and try to bomb, but do almost no damage and get shot down in droves as a thanks.

This change would also fix the big issue with CAS we have in multiplayer and improve Strategic bombing balance at the same time.

That is indeed a good suggestion, as I have also wondered why CAS port strikes don't work the same way as regular bombing where they can be intercepted and is effected from enemy AA. Speaking of vanilla port strikes I have experienced the issue of ships going missing from those strikes, and that problem is still in the game today where my ships simply go "missing" and their loss is reported nowhere in the naval UI which I find very disappointing considering that this problem existed from vanilla but was never fixed. Also I think that radar should also reveal the number of enemy divisions in its range which increases the significance of this building.

 

[jpd]

Erm, no. It really should not. Radar beams in a straight line from the emitter to whatever target it hits that reflects the beam back. It's that reflections that gives you the blip on your scope. However, the earth is curved, and the ground is full of clutter (trees, houses and whatever else is positioned on the surface) so that radar reflections will not work to detect ground forces.

You can circumvent this by deploying AWACS aircraft (that look down from a position in the sky), but those didn't exist in the WW2 era.

 

[Misaka_Complex]

Erm, no. It really should not. Radar beams in a straight line from the emitter to whatever target it hits that reflects the beam back. It's that reflections that gives you the blip on your scope. However, the earth is curved, and the ground is full of clutter (trees, houses and whatever else is positioned on the surface) so that radar reflections will not work to detect ground forces.

I thought it would've been cool if the numbers showed if radar was maxed out in a state with the last tech as a bonus. Otherwise maybe a "?" with on the number for enemy divisions would be fair maybe? Since Radar can also detect signals and gather intelligence.