Solid suggestion but one thing undermines it. Meta. Whatever AI will research, it won't matter much when players have meta templates with best division width and stuff.
An argument in favor of certain scripted tendencies.
- Thread starter Dalwin
- Start date
-
We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
IMO, Dalwin is correct, in that scripting production, techs and division/naval buildup would be much better than rely on weights that can apparently wildly alternate, which creates lots of wierdness, switching and weak armies/navies.
That said, scripts could be more general, not country specific. For example:
-build strong land army - will ignore navy-related stuff, focusing entirely on army, composition depending on amount of factories/resources.
-build navy - same as army, but for navy.
-homeland defence - focus on defencive units
-homeland defence for island nations
could mean there are only 5-7 general scripts to maintain, while each country can have dynamic weighting on what strategy it wants to employ.
After all, Germany does need to know how to sealion. Even if it doesn`t rush for it, if there is human player in axis, Soviets can be defeated, and then sealion and naval expancion are a must.
Just weights seem to insufficent/hard to make consistent.
That said, scripts could be more general, not country specific. For example:
-build strong land army - will ignore navy-related stuff, focusing entirely on army, composition depending on amount of factories/resources.
-build navy - same as army, but for navy.
-homeland defence - focus on defencive units
-homeland defence for island nations
could mean there are only 5-7 general scripts to maintain, while each country can have dynamic weighting on what strategy it wants to employ.
After all, Germany does need to know how to sealion. Even if it doesn`t rush for it, if there is human player in axis, Soviets can be defeated, and then sealion and naval expancion are a must.
Just weights seem to insufficent/hard to make consistent.
Transit complete ;-) If you have any questions on the terminology I use, feel free to ask. I know I tend to use strange words. To quote myself, the main counterarguments for scripting the AI wholesale are:
Problem 1 is not so much about "If you introduce new mechanics, simply write new scripts for them." but rather the fact that old code remains vaild (the buzzword here would be "reuse" or "modularisation"). If you have a working unit controller (ha!) or naval controller, you can simply add new mechanics and you may need to extend the contollers, but you won't need to retouch them completely for every balance change and new mechanic. Another aspect is the complexity explosion especially ahistorical HOI games may have. This ties in to point 3 and refers to the fact that the game tree (that is, a graph made from game states as nodes and decisions as nodes) only has one actual path but quite a few potentials.
If you observe the interaction of France and Germany, you'll get quite a few nodes (e.g. France goes communist, calls USSR. Germany goes democratic, France fascist, WW2 driver should switch to France.) Add in the other majors and you'll get a tree with (no guarantees on correctness, but the range counts) sum of([number of states for a country] to the power of number of countries). Scripting this is hell, because you have to touch or retouch every node. If you solve it by agent or algorithm (e.g. [potentially hostile country] has gotten new allies, we have to do something!) it gets far easier, if you can manage the corner cases. Of course, this means you have to define for each country:
Problem 2 carries over from problem 1, but is also kind of a human factor. The team may be proud and able. They may have weighted all factors and have come to their conclusion. We will never know, unless we get their meeting minutes.
Problem 3 sets up Problem 4. Since humans are crazy and rarely do the programmer the favor of behaving in a predictable manner, you'll leave some cases or comination of cases unsolved. There may be fallback routines, but they can be counterproductive. Moreover you may enter a dead state (i.e. hangup, freeze) from which the AI can't recover. In games where there are only a few states, this is an acceptable drawback (e.g. SC I, version 1.0. Later they got better because the scripts got more varied) because it saves time, money and nerves. In the case of HOI...well HOI3 tended to break irrevocably if you went strongly off-script. And you still have point 1, the complexity. This means that an observer-game can freeze, despite no human factor being involved.
Problem 4 follows from problem 3 if a human uses it to gain an advantage over other humans or the AI.
To this, you can add the "usual problems" of limited time, money, RAM memory, computation time and the need of the game to run at an acceptable speed.
Were I agree with you is that scripting may be beneficial for the game when employed in moderation and with great restraint. When going by the standard Input(state of the game world)-Processing(the "AI")-Output(actions of the AI in the game world) model, afaik the team is working to improve the "P". For me, a permissible use of scripts would be to narrow down the amount of input available. The use of scripts to "corset" the AI would be of much benefit, since it would reduce the complexity of the calculations necessary. Of course, you should aim for eventually removing the corset bones.
Problem with these is, that they are to some degree interdependent. Especially 3 and 4. Shall we?
Problem 1 is not so much about "If you introduce new mechanics, simply write new scripts for them." but rather the fact that old code remains vaild (the buzzword here would be "reuse" or "modularisation"). If you have a working unit controller (ha!) or naval controller, you can simply add new mechanics and you may need to extend the contollers, but you won't need to retouch them completely for every balance change and new mechanic. Another aspect is the complexity explosion especially ahistorical HOI games may have. This ties in to point 3 and refers to the fact that the game tree (that is, a graph made from game states as nodes and decisions as nodes) only has one actual path but quite a few potentials.
If you observe the interaction of France and Germany, you'll get quite a few nodes (e.g. France goes communist, calls USSR. Germany goes democratic, France fascist, WW2 driver should switch to France.) Add in the other majors and you'll get a tree with (no guarantees on correctness, but the range counts) sum of([number of states for a country] to the power of number of countries). Scripting this is hell, because you have to touch or retouch every node. If you solve it by agent or algorithm (e.g. [potentially hostile country] has gotten new allies, we have to do something!) it gets far easier, if you can manage the corner cases. Of course, this means you have to define for each country:
- What is a hostile country?
- What is the "something appropriate" in this instance?
Problem 2 carries over from problem 1, but is also kind of a human factor. The team may be proud and able. They may have weighted all factors and have come to their conclusion. We will never know, unless we get their meeting minutes.
Problem 3 sets up Problem 4. Since humans are crazy and rarely do the programmer the favor of behaving in a predictable manner, you'll leave some cases or comination of cases unsolved. There may be fallback routines, but they can be counterproductive. Moreover you may enter a dead state (i.e. hangup, freeze) from which the AI can't recover. In games where there are only a few states, this is an acceptable drawback (e.g. SC I, version 1.0. Later they got better because the scripts got more varied) because it saves time, money and nerves. In the case of HOI...well HOI3 tended to break irrevocably if you went strongly off-script. And you still have point 1, the complexity. This means that an observer-game can freeze, despite no human factor being involved.
Problem 4 follows from problem 3 if a human uses it to gain an advantage over other humans or the AI.
To this, you can add the "usual problems" of limited time, money, RAM memory, computation time and the need of the game to run at an acceptable speed.
Were I agree with you is that scripting may be beneficial for the game when employed in moderation and with great restraint. When going by the standard Input(state of the game world)-Processing(the "AI")-Output(actions of the AI in the game world) model, afaik the team is working to improve the "P". For me, a permissible use of scripts would be to narrow down the amount of input available. The use of scripts to "corset" the AI would be of much benefit, since it would reduce the complexity of the calculations necessary. Of course, you should aim for eventually removing the corset bones.
That scripted AI should work if you check the 'Historical AI preferences' seems fine. But you cannot possibly script a Seelowe in 1936, it is absurd. Unless Hitler had some sort of crystal ball no German would ever think about planning an invasion of the UK in 1936. The AI should be confronted with choices that the leaders at the time had to make under the circumstances and the information that was then available. So, the choice between Seelowe, the Mediterranean strategy and Barbarossa is a choice facing the German AI after the fall of France.
That the NF trees are scripted for all countries to a greater or lesser degree aligns with this. Where it all goes wrong is when a player starts in 1936, knowing that war is going to break out in 1939, and goes for a totally a-historical building strategy: exactly like a German player starting to plan for Seelowe in 1936. As an aside to that, the Anglo-German naval convention of 1935 limited German surface and U-boat numbers, and so there would or should be an option for the Germans to abrogate that treaty, but they should start the game limited by it. It is hard to hide building large warships, and they tended to be propaganda events. Abrogation or a direct breach would certainly rattle Britain.
One way, and at the moment the most effective way, of stopping this sort of crystal ball gazing is to start in 1939.
K
That the NF trees are scripted for all countries to a greater or lesser degree aligns with this. Where it all goes wrong is when a player starts in 1936, knowing that war is going to break out in 1939, and goes for a totally a-historical building strategy: exactly like a German player starting to plan for Seelowe in 1936. As an aside to that, the Anglo-German naval convention of 1935 limited German surface and U-boat numbers, and so there would or should be an option for the Germans to abrogate that treaty, but they should start the game limited by it. It is hard to hide building large warships, and they tended to be propaganda events. Abrogation or a direct breach would certainly rattle Britain.
One way, and at the moment the most effective way, of stopping this sort of crystal ball gazing is to start in 1939.
K
That has less to do with "strong AI" and more to do with the function/ability of port strikes being imbalanced to the overall gameplay. You even state it yourself, the "mechanics were changed" as being the primary factor here, not the AI that was changed. Obviously the AI might have been changed, but it was the effectiveness of the port strike missions that was the problem, not the AI's determinations on when and where to use them.
On the discussion of AI and scripting in general, people should be aware that all AI is scripted. Data IN > Process > Data OUT > Determination. So really this discussion is semantics and degrees of flexibility to various aspects of the logic/formula behind the AI.
I agree that the mechanics were not working optimally. My point was that it was the AI which exploited this weakness in the mechanics that highlighted the need for a rushed change, instead of a change which fit better with the overall air warfare model.
The point I was making was that it's fruitless to script/code the AI to be using known exploits which needs to be removed / fixed / rebalanced anyways just for the AI to become short term "stronger", because all it will do is highlight the need of making a quick and dirty fix of said exploit.
For example if Paradox were to see the vulnerability of ports before the release of the game and changed the AI to use all TAC bombers on port strike missions instead of ground attack such a change would need to be reverted back after changes have been done to mechanics which make port strikes less effective. ( This was probably not the case at all in this specific case, just using it as an example )
Not really.
Paradox games use both script and code. Script ( as used by Paradox ) is what can be changed via the text files and easily modded. Code cannot, but requires changing the exe.
Most of the AI functions is in code/exe, so it cannot be scripted at all when it comes to HoI4.
At first I was leaning towards agreeing, but after reflection I have the same concerns @jpd and @permanently_afk brought up, maintenance of said system, exploitation, and covering all of the if-thens brought up by all the possibilities throughout such a system.
Right now the game is too "unstable" for such a system. PDS still hasn't fixed basic sub-systems and settled on finished focus trees for all the nations. Trying to maintain a large script system over every update would likely be a manpower challenge.
Exploitation might be easy, in that as soon as a nation makes a certain choice you know where it is going. I will admit the current focus tree system is just as exploitable and this is always an ai problem with almost any gsg game now.
Catching and coding all of the possible conflicts of multiple script trees for each nations is my biggest concern. PDS admittedly missed many of these conflicts in the initial release of HOI4 and has been playing catch up ever since. And then missed some in TvF where they only added a few focus trees. If I understand your suggestion the complexity of linking multiple scripts for every nation and how each nation reacts to each script of each other nation is frankly boggling.
Maybe after another year or two when PDS has revamped the air and naval system, added espionage and diplomacy systems, and upgraded all the focus trees I could see this for the major and mid majors. Right now i wouldn't even bother to mod it for majors, there are just too many announced changes coming.
Just quickly:
I (and a few of my peers) call a narrow, problem-specific and (often) non-compiled solution "script". It's bascially shorthand for "this solves a specific problem in the field". Whipping them up takes litte time, and they often lack the i/o checks of a "proper" solution. Scripting a problem to death therefore implies solving each case "by hand" instead of thinking of a general solution (which often is complicated, complex and prone to errors). Giving the computer specific instructions for specific situations is - probably - the most popular meaning of the term "script something" on gaming forums.
Instead of directly quoting any individuals at this point, I will reply in general. I do this because less than half the thread is on topic and what is managing to stick to the subject seems to share a theme.
To paraphrase the main objection to having each major use targeted scripting in support of a high level strategy, it seems to be, "What if this sets the course of the AI upon a bad path?"
My response in a nutshell is one of my favorite quotes from Teddy Roosevelt, "In any moment of decision the best thing one can do is the right thing, next best is the wrong thing, the worst is nothing."
Right now the AI drifts more or less aimlessly with no real plan at all. Its strategy is whatever the NFs tell it to be for the moment. I suppose that is better than nothing at all, but its approach is certainly not integrated. One of the biggest conceptual flaws to the processes used by the AI is that it lacks a sense of future. Its decisions are all based in the current situation. Having overarching scripts for those limited areas to which I referred; research, production and templates; would at least give it some sense of direction. This would be much more efficient than how it currently handles things.
It seems completely obvious that this has to be done separately for each major nation. Japan has a greater need to research naval advancements than does Germany, for example. How does one get the AI to play effectively without tailoring tendencies? One could argue that if the AI is smart enough and takes enough factors into consideration, then it will come to the right conclusions regardless of which country it is controlling. That philosophy requires technology beyond what we have on our desks and programming which is beyond the capabilities of Paradox. They have neither the manpower nor the capital to produce something so complete. Individually scripting research tendencies is well within the scope of this project and their capabilities.
Would this approach produce a perfect AI? Of course not, but then neither will the current course of action that they are following. Would my suggestion produce a significantly better AI? I firmly believe that it would.
To paraphrase the main objection to having each major use targeted scripting in support of a high level strategy, it seems to be, "What if this sets the course of the AI upon a bad path?"
My response in a nutshell is one of my favorite quotes from Teddy Roosevelt, "In any moment of decision the best thing one can do is the right thing, next best is the wrong thing, the worst is nothing."
Right now the AI drifts more or less aimlessly with no real plan at all. Its strategy is whatever the NFs tell it to be for the moment. I suppose that is better than nothing at all, but its approach is certainly not integrated. One of the biggest conceptual flaws to the processes used by the AI is that it lacks a sense of future. Its decisions are all based in the current situation. Having overarching scripts for those limited areas to which I referred; research, production and templates; would at least give it some sense of direction. This would be much more efficient than how it currently handles things.
It seems completely obvious that this has to be done separately for each major nation. Japan has a greater need to research naval advancements than does Germany, for example. How does one get the AI to play effectively without tailoring tendencies? One could argue that if the AI is smart enough and takes enough factors into consideration, then it will come to the right conclusions regardless of which country it is controlling. That philosophy requires technology beyond what we have on our desks and programming which is beyond the capabilities of Paradox. They have neither the manpower nor the capital to produce something so complete. Individually scripting research tendencies is well within the scope of this project and their capabilities.
Would this approach produce a perfect AI? Of course not, but then neither will the current course of action that they are following. Would my suggestion produce a significantly better AI? I firmly believe that it would.
You see this in the production screen for nations under AI control.
When I took over the Raj in our last MP game, not one of her production lines had more than 20% of the production bonus. It was clear the AI had been swapping around lines for years. But it's the freaking Raj. At least the infantry kits production line should have been humming along for some time without getting switched around.
I do hope that I stay on topic ;-)
True enough. My personal reservations against a band-aid via scripting are that those things have the tendency to become fixtures. I am supported by this by a bit of computer-science/programming literature (f. ex. Clean Code [the one with the galaxy on the cover]). If you introduce something, make sure you can build on it. Else, throwing it out will be a fight. And yes, I am arguing to authority here.
Yes, this is something I agree on. Although, I would first recheck the focus trees and their handling, since those are imho the "driver" for the game.
Thing is, you could simply add weight to the relevant AI subroutines. Problem is, they don't work. Currently. I'd take this opportunity again to pitch the suggestion of reducing the amount of input. The grumbling introduction of impassable borders is a step into this direction.
As has been pointed out in this thread, the issue isn't that it doesn't solve the problem. The issue is that it is hard to change (that includes removal. Have you ever put of work for so long that at the end it became the mountain of doom?) afterwards.
Programmers often are reluctant to introduce that kind of stuff, if they are good. I think the problem in this case is a disconnect between the thinking of programmers/developers and non-devs. It kinda makes these kind of things hard to explain because they rely on a lot of implicit knowledge and experience. Now that I think about it, you could construe software and hardware as some kind of mystery cult. Long live the byte, as it is revealed to us!
May I ask how much programming experience do you have, Dalwin?
Edit: I ask because that isn't tattoded on your forehead (on the forums) and you demostrate competence and well-reasoned arguments (and yes, I am really bad at these "social interaction" things).
Last edited:
I do not disagree with the basis of this argument, not even a little. I do, however, disagree with your conclusion and interpretation of where this tendency takes us in the future.
Yes, anything seen as an interim solution, or bandaid in the common vernacular, runs a risk of becoming a permanent fixture. I have already successfully pushed for one of those early in the game and the community successfully did again just now in the Sahara.
Before deciding if that is a risk worth taking or if it is even really a risk at all one must ask one question. Is their currently stated plan at all realistic? Is their a significant likelihood that they will succeed in the next six months or even in the next two years? My opinion is emphatically that I believe they will not. The AI, in HOI4, will never reach the lofty plateau set by their current design philosophy. Perhaps in HOI5 but not in 4. My suggestion is realistic and attainable even if it is less satisfying for @podcat and company on a philosophical level.
My first programming classes were in 1976. I have no degree in the field and my only professional experience is to have developed some business applications using visual basic. I do, however, have some experience in project management and I think there is a fair amount of overlap in this context.
Point. In this case, I see no further point in arguing this considering your greater amount of experience. Although I'll maintain my stance of cautioning against simply scripting it wholesale. Why is the script of fourth semester software engineering "team roles" flashing before my eyes?
External/internal POV perhaps? Idk. And honestly, I do not feel qualified to render judgement on this, given my lack of information..
Don't take me there. I've had some bad experiences in both directions with "it's attainable" and "that's too much effort". Unfortunately this was in the course of working for a person who rapidly oscillated between these positions. *gah* Further, the focus on the thing shifted from week to week, and I forgot to take notes (and was far too intimidated by the guy). Oh well, lessons learned.
Still, I'd maintain that limiting the amount of input for the agent would be the better strategy. This would be applicable to more than just the majors. Being more flexible, you could either retain it (save CPU) or remove it when the processing step is competent enough.
This feels weird. Extremly so. For me, the experiences are in reverse, since I hold a degree and have a bit of experience in programming. And yes, you are right, I too think there is some serious overlap in this. Problem is, programmers don't grow on trees. Although I should be happy because of this, since it increases my salary ;-) Let's flash-clone @podcat , that'll save some effort. And yes, I think the HOI team needs more members. Six is the number going round, parts of the orthodoxy suggests the maximum is eight to twelve before you have to increase the overhead for a second unit.
I wanted to interject.
You all's debate is great and all, but it seems as this has become a case of arguing over a tree, and failing to see the forest.
The AI is already scripted. Some of you have alot of familiarity with mods, so this kind of weirds me out to tell you all, but, you already know the AI is scripted. What Dalwin seems to have originally been saying before he got side tracked is that the one ai file, plus the basic overall general ai file for specific countries, that each country needs an AI file with depth, or even MULTIPLE AI files, that could be loaded due to game start conditions, events in game, or semi-randomly, or even purely randomly, to further increase the amount of options and tailored strategies that actually work, ones a player might pull off in MP, to any specific AI, RATHER than having a general AI that is trying to do the right thing for dozens of countries. Thats more complexity not less complexity. Which in that light seems impossible. How can one size fit all? How can the general AI control both Argentina and Germany and be challenging? It cant. Thats not a complicated issue or hard to arrive at. That level of AI takes a long time and alot of resources to develop.
While there is a counter argument elegantly pointed out by the first responder, that a really good general AI would actually work for all situations regardless of nations or focii or historical accuracy, and you could push through a long period of developing it, there is also an argument to be made that perhaps that journey is for naught and that we could pursue something more like HOI3 and mods like Black Ice had, where AIs could pick from "profiles" and events, now "focuses" could provide more of the things that make an AI succeed rather than fail. That maybe making the old system better, and pushing through there, is better than continuing down this same path?
Or maybe im getting it all entirely wrong.
You all's debate is great and all, but it seems as this has become a case of arguing over a tree, and failing to see the forest.
The AI is already scripted. Some of you have alot of familiarity with mods, so this kind of weirds me out to tell you all, but, you already know the AI is scripted. What Dalwin seems to have originally been saying before he got side tracked is that the one ai file, plus the basic overall general ai file for specific countries, that each country needs an AI file with depth, or even MULTIPLE AI files, that could be loaded due to game start conditions, events in game, or semi-randomly, or even purely randomly, to further increase the amount of options and tailored strategies that actually work, ones a player might pull off in MP, to any specific AI, RATHER than having a general AI that is trying to do the right thing for dozens of countries. Thats more complexity not less complexity. Which in that light seems impossible. How can one size fit all? How can the general AI control both Argentina and Germany and be challenging? It cant. Thats not a complicated issue or hard to arrive at. That level of AI takes a long time and alot of resources to develop.
While there is a counter argument elegantly pointed out by the first responder, that a really good general AI would actually work for all situations regardless of nations or focii or historical accuracy, and you could push through a long period of developing it, there is also an argument to be made that perhaps that journey is for naught and that we could pursue something more like HOI3 and mods like Black Ice had, where AIs could pick from "profiles" and events, now "focuses" could provide more of the things that make an AI succeed rather than fail. That maybe making the old system better, and pushing through there, is better than continuing down this same path?
Or maybe im getting it all entirely wrong.
I think there's a bunch of things to keep in mind here. There's the angle put forward by Krafty above that, from a certain perspective, all of the AI's behaviour can be seen as scripted (ie, at the end of the day the AI is following a set of decision rules). Then there's the 'leave the core decision rules unchanged, but have some high-level rules on top that can be scripted to mix things up'.
In my modding, I've mused of randomly setting flags at the start of the game that do something like that, with AI strategies driving geographic and build preference, and the flags being used to link everything together. I haven't done anything (I want to give Paradox a lot more time before I get into that - my AI strategies at the moment are just along the lines of "no silly early-game invasions") but a lot of the tools are available, and triggers for various strategies and for setting flags to guide those strategies (or events that introduce templates) are in place. Flags can be used for AI technology preference modifiers as well. So something like this could be done.
However, it would be a lot of work. Lots and lots and lots, just for one country, done well, that gets exponentially complicated the more countries are treated this way.
Personally, I'm hoping the 'core' AI decision rules become sound, and the main benefits of scripting are small 's' scripting. Say, for example, having 'triggered as appropriate' strategies for Barbarossa that are at an operational level, with a choice, so that the AI has a general preference to push north, or south, or straight for Moscow, or what-have-you. Germany attacking France is an even easier example. There's really only so many ways to slice it, so a few 'weighted random selection' strategies that get Germany over the line would help, if it's too hard to get the unit controller to manage the kind of operational drives that we're all familiar with in history.
One of the benefits of starting operationally and working up would be that it's manageable (one set of strategies can be added at a time, and evaluated, and if it works well, well and good, and if it doesn't, throw the idea out).
Template wise, whether this driven by either core AI decision rules or scripted template changes, it'd be great if this was managed in a way that nations would make choices sensible to their circumstances (and, as Dalwin well says, with a degree of variability). I haven't looked at it closely (dirty land stuff) but I'd imagine while it might be a bit clunky, it'd be possible now to have some basic template variation scripting, possibly even with some 'game start' OOB loading for AI nations that gives them starting OOBs to suit their starting division template choices. That's a lot of work though, and something that was a bit more organic (driven by decision rules rather than static scripts) would be preferable, if it was achievable.
In my modding, I've mused of randomly setting flags at the start of the game that do something like that, with AI strategies driving geographic and build preference, and the flags being used to link everything together. I haven't done anything (I want to give Paradox a lot more time before I get into that - my AI strategies at the moment are just along the lines of "no silly early-game invasions") but a lot of the tools are available, and triggers for various strategies and for setting flags to guide those strategies (or events that introduce templates) are in place. Flags can be used for AI technology preference modifiers as well. So something like this could be done.
However, it would be a lot of work. Lots and lots and lots, just for one country, done well, that gets exponentially complicated the more countries are treated this way.
Personally, I'm hoping the 'core' AI decision rules become sound, and the main benefits of scripting are small 's' scripting. Say, for example, having 'triggered as appropriate' strategies for Barbarossa that are at an operational level, with a choice, so that the AI has a general preference to push north, or south, or straight for Moscow, or what-have-you. Germany attacking France is an even easier example. There's really only so many ways to slice it, so a few 'weighted random selection' strategies that get Germany over the line would help, if it's too hard to get the unit controller to manage the kind of operational drives that we're all familiar with in history.
One of the benefits of starting operationally and working up would be that it's manageable (one set of strategies can be added at a time, and evaluated, and if it works well, well and good, and if it doesn't, throw the idea out).
Template wise, whether this driven by either core AI decision rules or scripted template changes, it'd be great if this was managed in a way that nations would make choices sensible to their circumstances (and, as Dalwin well says, with a degree of variability). I haven't looked at it closely (dirty land stuff
) but I'd imagine while it might be a bit clunky, it'd be possible now to have some basic template variation scripting, possibly even with some 'game start' OOB loading for AI nations that gives them starting OOBs to suit their starting division template choices. That's a lot of work though, and something that was a bit more organic (driven by decision rules rather than static scripts) would be preferable, if it was achievable.
I don't think that the scripting is necessarily that straightforward or potentially less effort than just small basic research and production algorithms that could have weights per country. As permanently_afk points out, there are also some drawbacks especially around maintenance if you do it by country.
I think the better solution would be to evaluate the research and production of the major nations in the current version. Determine what they do poorly and determine at least a base algorithm that could handle it well. In comparison to army management, research and production are way easier to have the AI do decently.
I think the better solution would be to evaluate the research and production of the major nations in the current version. Determine what they do poorly and determine at least a base algorithm that could handle it well. In comparison to army management, research and production are way easier to have the AI do decently.
By the time you are doing "different weights per country," You are already on board with what I am seeking. The important part is not the semantics of how the pieces are labeled. Nor is it one of the degree of separation between the pieces when it comes to different nations. The main point, in my eyes, is that they need to recognize from day one that the different nations should be taking different approaches to strengthening their positions and achieving their goals. Just the simple tendency at the moment of every AI driven nation from Germany to France to China to Peru to strive for nearly identical 7+2 infantry templates even if producing those is unrealistic for many nations, points to the serious need for more customization.
I don't know precisely how the AI works, but I'd suggest instead of scripting, a 'strategic level AI' is added. That is, an AI that recognizes what position it is in and what long term goals it needs to accomplish to do well. The added benefit to this approach is that the AI will still work with total-conversion mods, while scripting pretty much breaks the AI for them.
Am I missing something or isn't this already done in common/ai_strategy/<country>.txt? (To some degree at least)
They have paid a bit of lip service to it, though it is my understanding that is mostly concerned with evaluating the importance to different theaters. I do not think it does much with research priorities, or production decisions, or template preferences. Now I realize that I am getting repetitive with those three items, but by now some might have noticed that those are also the most central things focused on by mods such as Expert AI 2.0.
Now I think that mod goes too far in areas like giving free templates to various countries and so on. I still want the AI to have to work for what it gets. I would simply prefer that it has more of a plan while doing so instead of stumbling around blindly.
Now if there are more specifics listed in that particular set of files, I would not mind hearing some details.