BPrado wrote:
I´ve always wondered (on my throne on top of the 0 knowledge I have on the subject) why not write this one module that´s dependent on difficulty level and which controls the AI´s decision to deny Victory Conditions, making it progressively more aware and reactive. I have a feeling that´s kind of what everyone expects out of difficulty levels.

Yes but flat stat bonuses are so much easier to do. That's basically what it boils down to.



AI is often the last thing considered when building a 4x regardless of the lip service given to it in the pre-release sales pitch.

Slashman wrote:
First, I'm not sure why Total War: Shogun 2 isn't a 4x. I'm sure there's some vague reason somewhere.

Well, I don't know either, but I wanted to head off any "what is/isn't a 4x" discussion off at the pass. I read the Isn't Endless Legend supposed to be a military game thread before this one, and I was unsure whether it was a touchy topic around here.

Second, you are simultaneously asking for a less passive AI and an AI that passively lets you win because you think you should have a smooth, non-violent victory once you get to a certain level of progress down a victory path near the end game. I don't think you can have both.



One of the problems with the AI is exactly that it isn't compelled to act to stop other players from winning when the end game nears because it's not aware of the flow of the game. AoW 3 does exactly the opposite and the AI there is much more effective and believable. The difference being that the AI tends to directly tell the player why its moving against him/her wrt to nearing victory conditions. "Hey...you're nearly done breaking all the seals. We can't have you rule the world. We're going to stop you!" is feedback that he player can understand. That's missing from a lot of 4x games.

I shouldn't have talked about a non-violent win, really. I made it sound as though I want to win by making daisy chains and singing songs. I'm fine with war, but only when it makes sense for the AI to do so. An AI that's forced to subdue its logic and declare a hopeless war isn't really understandable.



Also, the AI might want to take "the weakness of its armies" more seriously. I've only ever seen the cultists be nervous. Everyone else seems to get envious, even when I vastly outmatch them. They don't understand how precarious their situation is, sometimes.

The problem as it stands now is that there really isn't any way the AI can stop you from victory without declaring a hopeless war. There needs to be more diplomatic and espionage options to make it viable for a weaker AI to try and stop you without declaring war. If you have a good standing army and are well on the way to a science victory then the AI's should be actively trying to slow your tech pace through espionage actions. If you are close to Wonder, or Quest then they should be lowering your production, attempting to assassinate your high level hero that kind of things. You need to give AI (and players) the ability to disrupt the leader without declaring that hopeless war otherwise that is the only option it has.

Darkscis wrote:
The problem as it stands now is that there really isn't any way the AI can stop you from victory without declaring a hopeless war. There needs to be more diplomatic and espionage options to make it viable for a weaker AI to try and stop you without declaring war. If you have a good standing army and are well on the way to a science victory then the AI's should be actively trying to slow your tech pace through espionage actions. If you are close to Wonder, or Quest then they should be lowering your production, attempting to assassinate your high level hero that kind of things. You need to give AI (and players) the ability to disrupt the leader without declaring that hopeless war otherwise that is the only option it has.

Yes, and even in the case of military options, it is important for the AI to grasp the necessity of alliances. If you (or another empire) becomes powerful and threatening, it makes sense for a coalition to be formed against them (like in Gal Civ 2). So not only does the AI need to understand and use the military and non-military tools at its disposal, but it also needs to grasp the balance of power, which it does not.

Stinky wrote:


1) Have AI empires research imperial highways after the tier 2 weapons/armour and meritocratic promotion. Diplomat's manse should get a look-in too. The AI is determined to get the fortification/militia techs and I think I understand why: the AI gets boosts to FIDS at least, and it doesn't feel the need to research public libraries for beakers or imperial highways for dust. Anything that improves their military score is probably considered the number one priority. Unfortunately, the AI doesn't know how dull that is. I also feel miffed because in order to get the basic, mutually beneficial trade routes, I have to trade them Imperial Highways, usually to my disadvantage.

There is currently an over-emphasis on military technology - it's something we're working on at the moment

Stinky wrote:


2) Empires with weak military power/research/commerce could propose peace and commerce/research treaties of their own free will. Empires that are worse off should probably not race back to cold war when the honeymoon's over; you don't want your necromage neighbour to create yet another battleborn army, guys. Trust me on this. Cold war is not safe if you move your armies outside your territory!

Something else we're working on is the mechanism by which the AI chooses to propose diplomatic treaties - we'd like to make it a little more "rational" and a little less "emotional". Bear in mind however that the reason the AI acts this way at the moment is to avoid its being too predictable

Stinky wrote:


3) Have warring AIs bribe neighbouring empires to join in wars. At the moment, the only time you get multi-empire battles is when an opportunist jumps into the fray.

BPrado wrote:
I´ve always wondered (on my throne on top of the 0 knowledge I have on the subject) why not write this one module that´s dependent on difficulty level and which controls the AI´s decision to deny Victory Conditions, making it progressively more aware and reactive. I have a feeling that´s kind of what everyone expects out of difficulty levels.

KnightofPhoenix wrote:
Yes, and even in the case of military options, it is important for the AI to grasp the necessity of alliances. If you (or another empire) becomes powerful and threatening, it makes sense for a coalition to be formed against them (like in Gal Civ 2). So not only does the AI need to understand and use the military and non-military tools at its disposal, but it also needs to grasp the balance of power, which it does not.

Yeah, this would be really cool - pursuit of victory conditions and coordination between empires are things that have been on the wish-list for a long time, and don't doubt we're very keen on doing something about them. I'm afraid though that this isn't planned for the current pass - we're looking at the behaviour of each individual empire (technology, diplomatic exchange, choice of improvements, armies, tactics, etc) first. This might seem less exciting but it's the foundation that everything else stands on, so it needs to be as stable as possible.

Stinky wrote:


4) Regardless of whether #3's possible or acceptable, could the game please tell us if an AI declares war on another AI? It'd be nice. It's useful info because it's alerting me to a possible land-grab. It's also good flavour. If people don't like it, they're free to turn off the notification from the options.

Interesting idea - perhaps the AI could send you a message if your relations with them are sufficiently good? Something like "be advised: I am declaring war on X, pick a side!"

Stinky wrote:


This next paragraph is just in case Amplitude considers having a 100% guaranteed scripted brawl happen at the end of the game.

We don't, players tend to find this kind of thing contrived and frustrating - but I will read on anyway

Stinky wrote:


I was going to bring up Shogun 2, but it might not be a legitimate example as it's not a 4x, so let's go with a game of vanilla Civ V.

No need to let legitimacy get in the way of a good example - looking beyond 4X (or even beyond video games) can be a good way of bringing new ideas to the table

Stinky wrote:


Those AIs had a diplomatic malus with the player that increased per turn, so that war was inevitable. If I remember right, the idea was to always have a dramatic clash in the end game. The trouble was that until you found out what was going on, you'd see long-time allies turn on you for no apparent reason. Now, if AIs turn on their former allies because they have 3/5 science victory techs, or they're building the game-winning wonder, well, maybe there's a case for the backstab. But I don't think even that would be very satisfying, as it disregards diplomacy up to that point. Is the ability to set up a non-violent win such a terrible thing? Either way, I'd rather mash "next turn" waiting for the wonder to be built than go back to Civ V's insane AI leaders.

The issue of how to make the end-game interesting without a contrived AI "stacks on!" is a tricky one, and one we're really thinking about for ES2. In any case we're not overly keen on introducing an artificial aggro-per-turn to the AI.

Slashman wrote:
One of the problems with the AI is exactly that it isn't compelled to act to stop other players from winning when the end game nears because it's not aware of the flow of the game. AoW 3 does exactly the opposite and the AI there is much more effective and believable. The difference being that the AI tends to directly tell the player why its moving against him/her wrt to nearing victory conditions. "Hey...you're nearly done breaking all the seals. We can't have you rule the world. We're going to stop you!" is feedback that he player can understand. That's missing from a lot of 4x games.

Yeah, for long-term friendships between empires to really make sense though there would need to be some kind of allied victory - or at least a system of vassalage à la Solium Infernum. If allies have a reason to betray you late-game it's because there can only be one in most 4X games. AIs could of course explain why, and this would go some way to making the betrayal feel less contrived. Thing is most victories aren't like AoW3's - it probably wouldn't feel satisfying to have an AI declare war because you're about to win a diplomatic victory.

wilbefast wrote:


Interesting idea - perhaps the AI could send you a message if your relations with them are sufficiently good? Something like "be advised: I am declaring war on X, pick a side!"

And friendly ones under attack could ask for help.



I think these thing should be done via Diplomat´s Manse, it could use more value.



A sort of Denounciation can also help the AIs react better, and form sides and backstab other players in a more subtle way than open war.

BPrado wrote:


A sort of Denounciation can also help the AIs react better, and form sides and backstab other players in a more subtle way than open war.

Denouncing has to be one of the worst implemented parts in Civ 5, and I think it would be very tricky to implement correctly from an AI perspective. Most empires are going to denounce someone they dislike (obviously). As you said, since it is more subtle, it is something that weaker players would use against stronger. The issues arise because as soon as you 'have mercy' and don't destroy that minor empire you are fighting, then sign a truce at the players behest, he is going to denounce you.



This leads to a slippery slope full of denouncement chains. I would literally force myself to finish off empires (or simply stay in a war state) instead of signing truces because I knew they would denounce me as soon as I made peace.

willbefast wrote:


Yeah, this would be really cool - pursuit of victory conditions and coordination between empires are things that have been on the wish-list for a long time, and don't doubt we're very keen on doing something about them. I'm afraid though that this isn't planned for the current pass - we're looking at the behaviour of each individual empire (technology, diplomatic exchange, choice of improvements, armies, tactics, etc) first. This might seem less exciting but it's the foundation that everything else stands on, so it needs to be as stable as possible.

I would love to see this. Hopefully ES2?

Caotico09 wrote:
Denouncing has to be one of the worst implemented parts in Civ 5, and I think it would be very tricky to implement correctly from an AI perspective. Most empires are going to denounce someone they dislike (obviously). As you said, since it is more subtle, it is something that weaker players would use against stronger. The issues arise because as soon as you 'have mercy' and don't destroy that minor empire you are fighting, then sign a truce at the players behest, he is going to denounce you.



This leads to a slippery slope full of denouncement chains. I would literally force myself to finish off empires (or simply stay in a war state) instead of signing truces because I knew they would denounce me as soon as I made peace.

Denounciations work perfectly in single player in Civ5, I really don´t understand your comment.



Of course a nation you just conquered bits of will Denounce you if they can, that´s the intended function.





edit - The dillema between finishing someone off because it´s diplomaticly "easier" and signing peace because you simply can´t go on is supposed to be there - and in civ5 it is heavily influenced by the warmongering penalty that increases with each city you conquer + the huge amount of penalty you get from removing a player from the game, ignoring all happiness issues. Unless you have incredibly solid relations with someone previous to that, you can say goobye to diplomatic trade - which in turn means even less happiness.

wilbefast wrote:
Yeah, for long-term friendships between empires to really make sense though there would need to be some kind of allied victory - or at least a system of vassalage à la Solium Infernum. If allies have a reason to betray you late-game it's because there can only be one in most 4X games. AIs could of course explain why, and this would go some way to making the betrayal feel less contrived. Thing is most victories aren't like AoW3's - it probably wouldn't feel satisfying to have an AI declare war because you're about to win a diplomatic victory.

Well first of all, even when I turn off allied victories in AoW3, I still get empires forming and breaking alliances based on nearing victory conditions. I really like that feel to the game.



In terms of diplomacy and diplomatic victories, I believe that these are done incorrectly in most 4xs. A diplomatic victory should be as a result of other nations citizens liking you, not the leader itself. Actually most things should flow from that. I suspect that this is how you're approaching ES 2 and I wish it was already in EL.



Stardrive 2 doesn't just let you make unfavorable trades to empires you want to get friendly with. The people's tolerance level and disposition toward the other empire is a weighting factor. You can't just add more stuff to the deal willy nilly. It also acts to stop AI players from making bad deals and being exploited. There are diplomatic actions you need to undertake to build tolerance in the other races' population like trade treaties (tolerance is raised at the expiration point of the treaty), citizen exchanges and open borders etc.



I have always felt that most diplomatic victories are kind of hollow. No leader would simply want to see another empire, no matter how friendly, become the dominant force in the world. But if that other faction was so liked by your own people, you couldn't just declare war on them. You might be able to undermine them in other ways though. Like spreading false propaganda to other factions or even by making their people trust yours less so you could convince them to go to war.



Like I said, I got a few of these ideas from Stardrive2's diplomacy and espionage system which are probably the first ones I have really liked in a 4x.

Slashman wrote:




In terms of diplomacy and diplomatic victories, I believe that these are done incorrectly in most 4xs. A diplomatic victory should be as a result of other nations citizens liking you, not the leader itself. Actually most things should flow from that.

This. This right here.



Someone once made the suggestion that a future dlc for EL could/should include a "council or Auriga" system somewhat analogous to civ 5's world congress but different in mechanics and purpose. I sort of brushed the notion off as something that doesn't fit within EL's theme, but this idea right here both legitimizes this idea as well as making a sudden allowance for end game hostile AI.

If there was some sort of... "culture war" silently being waged at all times throughout EL, this would make an end game hostile A.I. totally ok. If, say there were some sort of drawback to declairing war, like angering the people or losing their faith in some way, an A.I. AND player would have their hands tied in a way. They'd have to at least get all their ducks in a row before they could really declare war to stop a diplomtic victory (they'd have to somehow lower the standing of the enemy empire in the eyes of the people before they could declare war without some sort of moral drop?)



I dunno, im just spouting off half baked ideas at this point... but the main idea im trying to get across here is MAYBE, through the addition of a whole new branch of mechanics (and consequent A.I. coding), the elusive endgame issue with 4x games in general could be solved in Endless Legend.

HappyHead wrote:
I dunno, im just spouting off half baked ideas at this point... but the main idea im trying to get across here is MAYBE, through the addition of a whole new branch of mechanics (and consequent A.I. coding), the elusive endgame issue with 4x games in general could be solved in Endless Legend.

I think you're on the right track. Being the ally of all races doesn't mean you're the beloved of all races' citizens.



There are countries who are allies of every UN nation, but that doesn't make those countries well liked by every nation. They can't claim control of the world like that.



Consequently, just maintaining an alliance with 4 or 5 races doesn't mean you have a say/influence in their governance. That's a big reason why I never bother to try to play the diplomatic angle in 4xs, and why I prefer diplomacy systems like SotS. Gunboat diplomacy at least makes sense.

Interesting idea



Not sure we can change this in EL at this stage though. I'd recommend having your voices heard over in ES2-land though, as things are not yet completely finalised where diplomacy is concerned. Here's a thread that was started on the subject.

Hi



The game AI is overall very weak, both in development as in combat, but it shows more on combat on targeting selection.



An improved AI would benefit the game imensly. (Better AI not give the computer faction +X production/combat/whatever modifiers)



Thanks

How is the ai re-coding coming along?

Any interesting quirks that have come up as the project has progressed?

HappyHead wrote:
How is the ai re-coding coming along?

Any interesting quirks that have come up as the project has progressed?

+1 I.m also looking forward to an update on this ...

HappyHead wrote:
How is the ai re-coding coming along?

Any interesting quirks that have come up as the project has progressed?

We're still on it but things are starting to fall into place now - the VIPs have been able to try a few builds and they seem to be fairly satisfied, I hope you folks will be too



As for quirks... hmm... I might try to write up a big post about all the stuff we've done to explain the various steps we went through. It'd be interesting to have a nice recap. But that'll be when we're actually finished and confident that everything is stable - it's at that point too that we'll start talking about release dates. Thanks for your patience on this - it's coming, and soon!

Keep the good stuff for the stream©

Frogsquadron wrote:
Keep the good stuff for the stream©

Yep - we'll be discussing the new AI in tonight's stream, please do tune in if you're interested

Hi all!



Thanks you for all your feedback on the AI over the last months - we've worked hard on improving it in accordance with your suggestions, and the results were released yesterday as part of version 1.3.0, "Forges of Creation"



You can find the full patch-notes here - below are the broad strokes.









[HR][/HR]













Equipment choices

abmpicoli wrote:


No accessories whatsoever! Not even basic iron ones (...) Not a single equipment with titanium or glassteel, even with the empire having a good quantity of it...

• Previous-tier iron items are phased-out
• Veteran units (with a level higher than the current era) are retrofitted first
• Dust equipment is used in late-game
• General improvements to the use of strategic equipment
• Better handling of accessories: governors use tomes, generals use insignias, etc
  
  

Managing settlers

Maverik wrote:


I can assume that this settler was preparing to colonize a region in the north, but another faction was faster and could not colonize on time. What was your reaction? It stood there 20-30 turns, even when loading game was still there. This does not make much sense. I do not remember when or why just leaving (I did not attack him).

Primus9 wrote:
I saw a settler just off the land looking like it was heading towards a region I had just colonised. The AI left it there.

• Change target region if an opposing settler claims the region first
• Pull back to a friendly city if attacked or if there is nowhere to colonise
  
  

Choosing fights better

abmpicoli wrote:


There are situations where the defending army only have one reiforcement tile and one single tile where to place their units... This makes me take an entire army of no matter how many opponents they have, I can simply take them out, one unit per time...

• Improved the calculation of military power slightly
• Take deployment zones into account: don't fight if you won't have enough room to field your forces
• Take potential enemy reinforcements into account
• Take high-ground and tile bonuses into account: avoid fighting uphill into a fortified area if possible
• Don't reinforce hopeless battles
• Don't retreat when under half hit-points
  
  

Army composition

Slashman wrote:
Poor army composition even when they have assimilated useful minor factions.

• Try to have balanced proportions of available unit models in armies
  
  

Siege warfare

Fuu wrote:
In a fairly close siege for the AI Cultist city, the AI chose turn after turn to embark and disembark its troops over the adjacent coastline for no reason. When they tried to break my siege, a significant portion of their army was on a boat and couldn't help.

• Better handling of armies when attempting to breaking a siege
• Better handling of armies when attempting to mount a siege
• Notably: disembark from ships more intelligently in order to mount or break a siege
  
  

Battle targeting

Caotico09 wrote:
AI still way over values wounded units without looking at terrain or unit location.

• A large number of small fixes: general improvements to default unit battle targeting
• Fixed an issue where units would target nobody and so do nothing under certain circumstances
• Improved accuracy in battle auto-resolution
  
  

[HR][/HR]













Overall

natev wrote:
The AI tends to make unintelligent choices about which techs to research, which leads to unintelligent decisions about which improvements to build.

Overlord_Yuri wrote:
The AI needs to make some sensible assessments in choosing research.

Slashman wrote:
The lack of 'common sense' in AI basic structure and research choices is also a big factor but this stuff has been covered already in the thread.

• Decisions are based on trends rather than current state
• Take into account the cost in strategic resources of an improvement when choosing whether to unlock it
• Generally improved estimations of when we need technologies: minor faction slots are unlocked when minors are available to assimilate, peace is unlocked when we want peace, etc
  
  

Military research

• Don't research so much military technology before it is needed: focus on having a strong economy instead (especially industry)
• Better evaluation of unit models: unlock cavalry only if we don't already have access to cavalry
• Take into account the cost discounts provided by strategic equipment technologies when choosing whether to unlock them
  
  

Naval research

• Research boats when and only when it is useful to do so ie. empires or resources on other continents
• Research cargo docks when and only when we have cities with tiles where docks and be built
  
  

Faction-specific

• Ardent Mages should research pillars and arcana much more readily
• Cultists research their stockpile generation abilities when they plan to attack a city
• Necrophages should now more readily research faction technologies
• Better handling of Vaulters/Mezari technologies with unlocks that require Holy Resources
  
  

[HR][/HR]













Colonisation

Maverik wrote:


AI founds a city on a border, this is almost always a bad idea, because you do not take the whole territory, also does not try to develop a sense for Level 2 districts.

The city of the image its probably not better example, I see worst places every time.

HappyHead wrote:
I suspect that the AI may not consider room for expansion when settling their cities

Andy06r wrote:
City Placement snowballs into the other economic issues. A popular civ4 mod improved the AI a full difficulty just by optimizing city placement.

• Colonise much more quickly and efficiently
• Don't build colonists when at war
• Choose safer regions than previously
• Consider the amount of free space (potential for higher-level districts) when choosing a spot for colonisation
• Better patterns for city expansion
  
  

Trade

Maverik wrote:
I always build roads in the conquered cities because they never have.

• build roads a lot more often
• build cargo-docks in order to establish sea trade
  
  

Resource management

Rustam wrote:
Another thing: the AI very rarely uses the market. Especially rarely buys something, even when he has a lot of dust, and a strategic resource or luxury very useful.

norph wrote:
AI just barely use the market. Sometimes he sell a few items of some resource, and from time to time buy a unit or a hero, and that's all.

• Manage dust better to keep the empire in the black (rather than selling off units in winter)
• New system for selling surplus strategic and luxury resources on the marketplace
• Improved Holy Resource selection for Vaulters/Mezari
  
  

Overall strategy

• Change primary focus based on other empires' current scores: try to be the best at something
• Cities, once mature, will specialise based on their strengths
• Invest significantly more money into heroes, try to match the hero count of other empires
  
  

[HR][/HR]













Asymmetrical evaluations

Caotico09 wrote:


This shows me purchasing luxuries for 100 gold, and then selling them for 700 gold or a technology.

• Be less naïve, less easily manipulable: evaluate terms differently depending on who we are dealing with
• Bargain much more harshly with empires that are leading than empires that pose no threat
• Correctly take the market value of resources into account when buying and selling commodities
• Military clout now plays as much larger role in diplomatic negotiations
  
  

[HR][/HR]













Factions

hera35 wrote:
To me it seems that most, if not all AI are running on some sort of generic personality and playstyle script

Enlil wrote:
TL;DR: Each faction needs a default research tree with priority values assigned to each tech.

• each faction will tend to want war, cold war, peace and alliances to a varying degree
• different factions will tend to propose different terms to the player
• different factions will tend to prefer...
  
  
  • ... different kinds of diplomatic terms
  • ... different kinds of technologies
  • ... different kinds of improvements