Improved AI

uriak

(This is a repost of my reddit thread)

Celtik has released his latest (and final) AI mod for Warhammer 1, Advanced AI
As a tester, I believe you may wish to give it a spin, especially if you like to keep the game rules almost unmodified.

This mod includes previous work on the AI recruitement and balancing of AI factions in campaign but also extends the AI tweaking to new features :

• diplomatic relationships (AI ignore less the player at low difficulty, only attacks when it's logical at high) diplomatic relationships (AI ignore less the player at low difficulty, only attacks when it's logical at high)
  
• skill and spell usage in battles have been reworked and are more coherent
  
• horde factions are more dangerous
  
• skills are no longer picked randomly by the AI when leveling up its characters.
  

A few words about AI limitations and the scope of this mod

• The AI in warhammer is still unable to do a few key things and no mod (without heavy scripting ) could change this.
  
• it won't ever disband units that are no longer needed. So early game armies will stay until destroyed
  
• the Ai never destroy buildings and will always recruit a stack from a single province. This severely limits the army compositions
  
• there is no trigger for some battle informations and the generic battle AI itself can't be modded
  
• The Ai can't adapt its armies to fight a specific culture.
  

The mod itself changes so many parameters that it's not compatible with many large scope mods, especially overhauls. It can't be expected to be, because what it does is trying to make the AI more believable and fun within the set of the vanilla rules. It could be extended to mods that add new factions and units, however, though it's no little work.

Even if you've already played with several mods, you may still enjoy this one for a fresh take on the game. The armies you'll meet and they way the sandbox evolves can be quite different from the base game. Many mods already change some aspect such as faction balance or diplomacy and they are not compatible but they most often try to solve the same set of issues at their core.

In any case feedback will be interesting for the possible adaptation of this work to the next game. There is still a lot of unknown parameters of course and we have yet to discover the flaws and new advantage of the vortex and mega campaigns. We can hope CA will steer things in the right direction and mods such as this one can surely lead by example.

There is a lot they can do to improve the base features of their Campaign and Battle AI. Some elements are subjective and by design, such as the general level of agression or shyness of the AI armies, but others are really in the realm of the perfectible.
Even if I do agree with some of their solutions, most of the time when a feature is simplified, this is because the AI can't nicely handle it.

Overlord87#6613

His mods have always been in my "must use" list of mods. Excellent stuff all around.

Needless to say, the release of his latest mod pushed me towards immediately starting a new campaign

danielcoh92

Lads, don't forget that this mod is just a band-aid.

CA HAS to acknowledge the issues with it's BASIC non functioning AI and improve it significantly so the game will be fun and challenging again.

giving the AI bonuses on higher difficulties is not the solution, they have to give it objectives just like the player and make decisions (right now the AI is absolutely horrible). the campaign map should be living and the player should see the AI trying to achieve it's goals and get a sense of competitive spirits when you play, having to actually fight for rich regions and control of key points, and make the game more about "good versus evil" and not the nonsense it is right now.

as long as CA don't acknowledge the issue with it's AI, the game won't get any better, regardless to what content they release. They are very good at designing new content, but the foundations are SO bad and broken that it ruins the experience as a whole.

regardless to what I said above, CeltiK's mods are amazing and I use them since the day I realized how stupid and not competitive the AI is.

HighRoller64#8823

Overlord87 said:

His mods have always been in my "must use" list of mods. Excellent stuff all around.

Needless to say, the release of his latest mod pushed me towards immediately starting a new campaign

Same here, I'm loving this one. Very well done.

Beermachine

Just started a new game with this mod.

AI development has been getting progressively worse in computer games for years, with less focus and budget devoted given that generally it doesn't in itself sell more copies (just vastly impacts the longevity of a game).

Throw in the fact that AI development is a specialist field in high demand that commands premium salaries (which game companies won't pay), it's also not surprising.

Obviously it's most apparent in the strategy genre given that AI programming is more complex, and have a much higher need for at least a semi-competent AI.

Luckily there are still talented amateurs who spend countless hours modding AI behaviour to improve games, despite the games themselves often making it more difficult with archaic non modular design and hard locked code.

At the very least companies should make AI modding much easier and far reaching, especially if they can't be bothered to devote any resources to doing it themselves.

Axikus

uriak said:

the Ai never destroy buildings and will always recruit a stack from a single province. This severely limits the army compositions

Actually, it does.

https://forums.totalwar.com/discussion/200045/ai-cant-demolish-buildings-apparently-it-can

But probably only does so in desperate situations.

danielcoh92

Axikus said:

uriak said:

the Ai never destroy buildings and will always recruit a stack from a single province. This severely limits the army compositions

Actually, it does.

https://forums.totalwar.com/discussion/200045/ai-cant-demolish-buildings-apparently-it-can

But probably only does so in desperate situations.

the AI can't demolish buildings.

they probably got sacked.

Axikus

danielcoh92 said:

Axikus said:

uriak said:

the Ai never destroy buildings and will always recruit a stack from a single province. This severely limits the army compositions

Actually, it does.

https://forums.totalwar.com/discussion/200045/ai-cant-demolish-buildings-apparently-it-can

But probably only does so in desperate situations.

the AI can't demolish buildings.

they probably got sacked.

Six turns apart, they already have tier 2 barracks on the second screenshot, and there were no wars between them and any other faction.

uriak

This may be possible in case of fully blocked faction. Alas, that don't mean the AI may do more mundane and regular "pruning". The most egregious cases are those T3 max building made in capital slots or those "early building" taking the place of important buildings needed after the T4 level.

@Beermachine : I partially agree but not on all point.

First, even if it seems weird, the current string of deep learning development don't really apply to games for a host of reasons. They are the domain of large scale operation with computer farms, they require important case sets for the learning parts and in the specific domain of videogames not based on really fixed rules, this can't be the case, and finally, their benefits are not evident in games in need of "opposition" not a winning partner. In strategy game most often the AI handles a large subset of the game, larger than just being a bot.

I agree that it looks to be at best a second level priority, which seems strange because a bad AI can be gamebreaking, but lett's take one of the most recent examples. Civ VI has been shipped witth (and still has for the most part) an AI unable to play the basic level of the tactical layer of the game. Did this hamper the game success? Maybe, but it's clearly a long running issue, not one that appears on early reviews.

That is to say the "it's good enough" attitude can stay.
In the case of TW, the CAI itself doesn't have an agenda most of the time but does a job good enough to survive and pose a certain challenge. There are already quite many factors allowing for specific intereactions, but they have to be manually tweaked, which is a great part of the job done by CeltiK here. It's like a functional engine, with parts that aren't well oiled if you wish.

There are missing elements though, but I wouldn't call them mandatory, at least all.

The AI needs to be able to disband units when they are obsolete or worse when they are hindering the economy. I've tried to see if I could kill units selectively myself, but couldn't find a function that allows it (killing units, not an army) I don't even know if it exists.

The planning phase for an invasion, is not mandatory : the AI knows to recruit, can be told what (once again, the quantity/quality have been tweaked in this mod) since the way to conduct war is pretty direct in TW, it works.

The devlopement itself, can be tweaked by modding too, but there is a need to tell the AI when a building is not desired.

I think it's important to ask CA to at least communicate on the topic. If they aknowledge something it's a good step.

Beermachine

@uriak

I agree, and realise the vast differences in AI development in different fields of IT. The thing is, speech recognition, optical character recognition, vehicle automation, robotics etc have all made significant advances in the last 10 years because companies have devoted the money to the research, and are now reaping the rewards.

In computer gaming (outside of Chess and Go), no real advances have been made, and in some ways have regressed. If the gaming industry was in dire straights (like mid 80's), this is understandable, but given that many major companies (especially in the strategy genre, Paradox is a good example) are recording record profits this is less acceptable.

This is exasperated greatly by the fact that game developers (C, C++, C#, etc), are generally the worst paid in the IT industry. Many of the best talented, even if they'd prefer to be in games development, instead opt for enterprise software, banking, infrastructure, etc. Basically anything but game development where salaries are very low and the workload is high.

As for Civilisation 6, while it may not have impacted their short term fiscal projections, Civ 7 and any Civ 6 expansions will sell markedly less, in large part due to the extremely poor AI. The fact that a lot more people are currently playing Civ 5 than Civ 6 is a sure indication of this. Given how profitable the franchise has been to date, this will be a major concern to their marketing department and upper management.

uriak

We can wish the sales at laast tell them something indeed; and yes the most talented aren't in VG dev, or at least don't stay. The pay, the hours, the cycle of delivering unfinesh products with a focus on the next iteration are not attractive.


As for the game AI, it' not fully true. The thing is what advances have been made are often offset by the increased complexity of the tasks. Pathfinding etc are quite good and when you see the state of the battle AI, I'd say it can hold its own, albeit not in all scenarios. (there are thousands of entity in play)

For instance, CiV 4 AI was more competitive but moving stacks and making them attack with the best odds was way easier than the current UPT system. THAT said, the experience of CIV V should have lead to something one order of magnitude above, and they should have strived for something "good enough"

Some older games with a "good" AI were the result of time spent honing the parameters perfectly, alas, this work can't be applied easily to iterations of games.
One example I remember was HoMM 3, considered to have had the best AI in the serie. They did find the best empiric weighting for it's pathing algorithm, letting it do a maximum of task within each turn.
But with the change o rules in the next game, their tweaking wouldn't cut it and was too costly. Thus the Ai was "dumbed down"...

The core issue is still that graphics = gameplay > stuff works > performance > AI is the most common in order of priority

Beermachine

uriak said:

The core issue is still that graphics = gameplay > stuff works > performance > AI is the most common in order of priority

Absolutely. Though in modern gaming I think it's now more like:

Graphics > gameplay > performance > non critical bugs > AI.

While I'd love to blame the companies, in many cases this change is consumer driven.

uriak

Well, there is a need to really ask them about some AI improvements. Blaming them can lead to topic avoidance, but we should be encouraging progress too.

But at the same time be wary of vague stuff like "recruitement improved" in the past patches that is not actually true.

Horus168

Beermachine said:

uriak said:

The core issue is still that graphics = gameplay > stuff works > performance > AI is the most common in order of priority

Absolutely. Though in modern gaming I think it's now more like:

Graphics > gameplay > performance > non critical bugs > AI.

While I'd love to blame the companies, in many cases this change is consumer driven.

Those first three effect short term sales, where's how good the ai is is a much longer term thing. I'm sure a better ai would improve player satisfaction and lead to more repeat purchases in the series and further sales by word of mouth but it seems like it'd be a much slower gain of customers than thhe other three have the potential to provide.


Don't get me wrong, improved ai is the feature i most want in TW. Doing code development myself, it doesn't seem line it would be hard for devs to improve from the current algorithms if it was given even moderate resources.

uriak

That's why I maintain it should be a community concern raised as often as others. Of course it's not as trivial as changing shields, or what's not, but they did work to answer this.