I think that a lot of people have made a wide variety of good points. I think that we're each going to end up focusing on different aspects because we have different playstyles.



My style is toward a military, steamroller Broken Lords that rolls over the opposition with a single, strategic-heavy army starting in middle/late tier 2, around normal turn 40.



I'm reasonably confident that I could make a simple flowchart that describes my play that, with the bonuses given Endless difficulty AI, would be unbeatable. At least, unbeatable by me.



In order to defend against my style of play, high difficulty AI needs to:

• Research and build from a relatively fixed list of high value technologies and improvements
• Retrofit units to full strategic weapons and accessories
• Stop fighting battles it can't win (and remember when it loses a battle it thought it would win)
• Group its armies in order to bring maximum number of reinforcements into battle
• Anticipate military action on its own capital
• Focus on taking capitals
• Gang up on players approaching a victory
  
  

If high difficulty AI did all of the above, I might stop winning my high difficulty games. Some of these things are less appropriate to easier difficulties (being ganged up on is frustrating, fixed lists are effective but boring).

The biggest issue I have with the AI is they leave a lot of units garrisoned in their cities when they really should. I was playing online with two other people and one player was snowballing out of control with units. So naturally I bribed all the AI to attack this player, but when they all went to war they did nothing but keep like 18-30 units garrisoned across their empire. Also another thing I noticed, I'm sure it was said already though I did not read all the post, is that they make some really strange choices before attacking a target. In the same game mention just before the snowballing player dc'ed when he was attacking a city with 3 stacks of units and a hero, but rather then attack with all three stacks the AI from what I could tell made his hero a governor and broke 2 stacks away from the attack. There are others, but I sure a lot of those issues where brought up already.

I've had a read through of some of the suggestions and I think the community has done a very good job of identifying the main problems.



My #1 suggestion would be to have separate AI behavior for each faction so that the AI can take advantage of the faction's unique abilities. I understand that you have randomized the AI personalities which means that I might face a peaceful Necrophage faction or some other behavior that is actually a massive handicap for that side. I appreciate that crafting a separate AI for each faction would be a lot of work but you only have eight factions so it might be doable within the length of this game.



My #2 suggestion would be to have the AI upgrade all its units at every opportunity. For example, one of the very first things I do when I start a new game is to add Iron accessories, armour and weapons to all my starting units. What a difference that makes! The AI should do the same. You only have three unique units per faction so creating templates for each unit doesn't seem like too much work, something doable within the lifetime of this game. Yes, there are assimilated factions but I would be happy if the AI could upgrade its unique units within the game. One standard template with all the Iron accessories tiers 1-3 and then Dust from tiers 1-3 for the standard units, and one with special weapons for titanium or glassteel, one with armour and one with both etc.



My #3 would be improving the tech picks the AI makes. This could be tied into #1. If it picks military techs, I will have to do the same and so those AI picks are not wasted.



Those are just my picks, already all mentioned.



One thing, not directly related to the AI is that your battle maps are very, very small indeed. The battle maps in AoW3, for example are absolutely massive by comparison. It's a cool idea to use the strategic map as your tactical map but there are too many bottlenecks on your average battlefield and the AI just can't handle bottlenecks. It's not just this game that can't handle bottlenecks, it's any where 1UPT is used for combat and EL has 1UPT in combat. One way to improve the situation might be to allow units to move/attack up 'cliffs' but at an increasingly severe penalty so that the AI doesn't send its units off on a long useless jaunt and remove itself from the battle for 2-3 impulses. I would also ensure that there is plenty of space for the AI to deploy all of its units at the start of a combat.



And thanks for eliciting the feedback of the community on this matter.

Rustam wrote:
Please, read this thread.

The main thing: AI should place independently units before the beginning of fight. Otherwise the army significantly loses in efficiency. And is worse than that: the part of army can remain locked, without a possibility to approach to the opponent - that is visible on my sreen.

This is a bug but the good news is we have already encountered it I hope it will be fixed soon.

abmpicoli wrote:
I've said this:

/#/endless-legend/forum/5-general/thread/2890-your-feedback-on-ai



To which you have answered this:



I think you are missing the point... I'm talking about being able to outsource the AI, by exposing an interface to it, and let interested third parties develop alternate AIs... I don't know how difficult that is , and it will basically depend greatly on how you have designed the AI's architecture... If you already have a well encapsulated AI, then it will be only a matter of exposing the interface and allow some communication protocols to hook into it (and yeah, I know, it's easy, any MIT Ph.D team can do that... )...

Well, outsourcing the AI would be a huge amount of work. I’m afraid this will not be possible on Endless Legend.



So, I tried to sum up the different points you noticed. Let’s say the AI should:

• Plan movement of the whole army instead of each unit during the battle
• Move out of a reinforcement tile when reinforcements are still waiting during the battle
• Retrofit its units faster/better
• Take into consideration the best angles to attack (reinforcements, deployment zone, etc.)
• Focus more on strategic resources (on this point, the last update improved a lot how AI prioritizes strategic resources)
• Being able to not include reinforcements if the battle seems to be doomed
• Use the market smarter and more often
• Use diplomacy to stop winning empires; for instance, by organizing a common war against the growing empire with the weaker empires as allies, breaking a peace with a player close to a diplomatic victory, or not trading techs from the 6th era
• Be less exploitable when trading via the diplomacy
• Should not target first the militia during battles
• Have better army mixes
• Make better hero assignations
• Be able to equip accessories
• Use faction abilities such as Force Truce or Vaulters’ teleportation
  
  

And of course, thank you for your feedback and keep it coming!

Manu wrote:


So, I tried to sum up the different points you noticed. Let’s say the AI should:

And of course, thank you for your feedback and keep it coming!

One thing that a couple people have posted about that isn't on your list:

• Better research selection/prioritization by the AI.
• Better handling of units right before siege and during siege (pulling units out of city before siege and not leaving units to die in cities after fortification is gone)
  
  
  
  

To me, would be a important point improve their awareness of borders, they seem not have in mind the places where they should attack, or where they will be attacked more easily. You can attack a border town and maybe this empty without defense.. and sometimes they will attack a distant city without sense.



PD: I must say though that for me the problems of AI should be accompanied with significant changes in the overall balance of the game.

Don't know, if it has been said here, sorry if it has.

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.

It wouldn't bother me, but there is a faction, which's advantages are COMPLETELY depends from market usage of other factions. Yes, Clans may be interesting for multiplayer, but in a game against AI their only feature is double health for mercinaries.

One more: during combat, consider height more carefully: the AI should use as a tie breaker between two equally good tiles the height of the tile: the units should prefer high ground, choose where to deal the most damage.

abmpicoli wrote:
One more: during combat, consider height more carefully: the AI should use as a tie breaker between two equally good tiles the height of the tile: the units should prefer high ground, choose where to deal the most damage.

I watch a lot of auto-resolve battles to understand how the AI would possibly manage to tie a battle that is surely won for my side. In my experience, the AI already puts very high priority on higher ground, only below morale i´d say, and a lot more than I often would have preffered. I think it is, along with its fixation for morale, one of the main factors why it can´t ever have efficient formations during an entire battle. I think the way it prioritizes high ground should perhaps be different, but not more intense.

One of the biggest weaknesses of the combat AI seems to be that it can be lured, easily. If the enemy has only infantry and I've got mostly nameless and one preacher, they'll always go for the preacher, first. So if I put him in the corner of the combat map such that he can be reached by going through an arduous roundabout course and put my nameless on the cliffs to the side, the infantry goes straight into a traffic jam for fighting the preacher in the corner and I get to shoot at them.

In the end, it is the already mentioned point that they don't consider formations + the point that they do not seem to weight the target priorities with the expected time until they can actually do damage on the target.

Folks have covered a lot of ground here already but I want to emphasize something that I feel is very important.



The fact that the AI seems not to understand it's own strengths or weaknesses based on its faction is always going to be problematic.



Necros should prefer war to peaceful co-existence. The Drakken need to use their Force ability to stop a war they are losing or force an alliance they would find advantageous.



It personally makes me worried when you say the AI is more role-playing than pursuing clear objectives based on its faction. What was basis for doing this? Why do you think it makes for more effective AI?

Something just happened that made me remember this thread.



TL/DR I´m a fairly strong Cult on an island and have taken a Roving Clan under my wings. After around 70 turns of friendship, diplomatic trade, compliments exchanged and military assistance from my part, they market ban me, and some 10 turns later go into Cold War with little to no possibility of negotiating Peace again.



They were the first faction I met, and I instantly made peace to complete a quest. Not long after that, they settled lands i had been protecting from necrofages´ incursions. We exchanged vision and I saw he was not really such a threat, so I figured I rather have them there than constantly fending off the damn bugs. By this point I already had a couple of nice armies stationed in critical regions, and the blue roving clans settled another one of those. The bugs started harassing them, so I kept the armies where they were for safety.



I made sure I constantly complimented them from then on, figuring the hero-led armies would probably count negatively in whatever diplomatic count there might be. I even kept them away from the city capitals to make sure I didn´t get in the way of some district, I don´t know if that´s possible. When they Market Banned me, I assumed they were after some luxury or tech, so I tried trading my freedom to no use. I simply nullified it then and went on. They then warned me about my armies in their region, and I proceeded to move them just away to the borders. Soon, they declared cold war and no reasonable trade would convince them otherwise.



My armies can get both their cities in the same turn we go cold war. I have all the villages converted, so if it wanted any of them it would need to do that. But there´s plenty space on the map to settle yet, it doesn´t have any military advantage or even condition to stand the consequences of its decision and it´s actively fighting a major part of its own defenses so far in the game. It doesn´t make any sense. It´s as if it did not consider its experience at all, or as if it could not tell apart friendly military presence from non-friendly.



I´ll up saves and link them here.



edit: - Turn 108 / Turn 109

I didn´t honestly believe it would repeat it, I thought it would make me look like a fool

endless/standard speed/random factions/many continents

The files are on dropbox

BPrado wrote:
I watch a lot of auto-resolve battles to understand how the AI would possibly manage to tie a battle that is surely won for my side. In my experience, the AI already puts very high priority on higher ground, only below morale i´d say, and a lot more than I often would have preffered. I think it is, along with its fixation for morale, one of the main factors why it can´t ever have efficient formations during an entire battle. I think the way it prioritizes high ground should perhaps be different, but not more intense.

OK, concrete case: I had this situation with the Vaulters, last day. The marine had an order to attack a demon, and was 4 tiles away. There was 2 possible tiles, both with exact the same terrain, the same distance to run, and the same end morale bonus, but one was on high ground, and one was on lower... It chose the lower ground tile...

BPrado wrote:




My armies can get both their cities in the same turn we go cold war. I have all the villages converted, so if it wanted any of them it would need to do that. But there´s plenty space on the map to settle yet, it doesn´t have any military advantage or even condition to stand the consequences of its decision and it´s actively fighting a major part of its own defenses so far in the game. It doesn´t make any sense. It´s as if it did not consider its experience at all, or as if it could not tell apart friendly military presence from non-friendly.

The fact that you use the phrase 'did not consider its experience' wrt an AI is the biggest glaring issue with your post. AI doesn't have feelings and does not consider past events in the same way that humans do. In all likelihood, the AI is simply seeing you excel and doesn't want you to get a victory so it is doing the only thing it can do since it can't declare war as Roving Clans.



Now that doesn't mean that other bad factors of the AI aren't in play in your scenario, but you are looking at the AI more as a person, which is wrong. AIs don't have 'regrets' and they can't analyze a situation in the same terms that you can. Before we even start going down the AI long view road, I sincerely believe that there is a lot of ground that needs to be covered regarding the basics of gameplay (both strategic and tactical) for the current AI. A lot of which is being covered up by the larger bonuses of the higher difficulty AI.

(: I´m not saying it should be my bff, give me stuff for free for all the help, not even not Market Ban me. I´m saying it should be able to recognize cooperative behavior and act accordingly; maybe even more than that, it should be able to not suicide.



I´m actively keeping away armies that are already at war with it, factions become "Blood-Brothers" and send alliance invites for muh much less. I haven´t sent a single threat, I´m not the weakest or the closest or the most convenient faction for it to decide to have attrition with. I was not the first on anything by that point. I can only assume its motivation might have been its quest, and if that´s the case, there should be a whole lot of work on those priorities there, because it shouldn´t be chasing victory conditions it has no chance whatsoever to achieve. And honestly, saying I look at the AI as if it were a person is a bit insulting.







The game is over now, the RC in question eventually attacked one of the villages in "their" region with a ridiculous army, i got both cities, a couple more he settled in the exact same spot despite our current war and then some 30 turns later he offered 200 gold for a truce.

Now, this game also gave pictures of the AI trading victory techs for ridiculous things. Steam won´t let me upload the first one for some mysterious reason, but here´s the >second and >third techs I got from trade. I would eventually win 22 turns after trading the third. I researched two techs and got a science victory.

BPrado wrote:
(: I´m not saying it should be my bff, give me stuff for free for all the help, not even not Market Ban me. I´m saying it should be able to recognize cooperative behavior and act accordingly; maybe even more than that, it should be able to not suicide.



I´m actively keeping away armies that are already at war with it, factions become "Blood-Brothers" and send alliance invites for muh much less. I haven´t sent a single threat, I´m not the weakest or the closest or the most convenient faction for it to decide to have attrition with. I was not the first on anything by that point. I can only assume its motivation might have been its quest, and if that´s the case, there should be a whole lot of work on those priorities there, because it shouldn´t be chasing victory conditions it has no chance whatsoever to achieve. And honestly, saying I look at the AI as if it were a person is a bit insulting.

You are taking insult where none is intended. My aim wasn't to single you out, it was to demonstrate that when we talk about the AI and what it should or shouldn't do, we need to also understand that it isn't seeing the same situations or options like we are.



I agree that in your scenario, antagonizing you is a bad idea...if I were a human player. There are things you mention though, which I am very uncertain about the AI seeing as a factor. You keeping other armies away may not be a deciding factor in AI behaviour at all. You also say you are not the weakest faction for it to fight with. It can't declare war at all. Maybe you are simply the biggest threat to its own victory that it can perceive. I have no idea if it has contact with all the other factions or not, I'm just saying that you need to be careful of applying human logic to AI in terms of long term thinking. Most game AIs cannot do things that way.



It is for that very reason that I am way more concerned about getting the basics of empire building worked on for the AI. A dev admitted in this thread that the AI works off personalty for role-play reasons. That, to me is a much larger basic problem. What is the point of Necrophages having a 'nice' or 'peaceful' personality?



In your case, what is the basic premise (from an AI POV) of Roving Clans being hostile? I'm still unsure whether a faction that CANNOT declare war is a good idea at all. I personally think a penalty for declaring war is much more sane than never being able to, but that's just me. I've done a lot of talking to developers about AI recently because I've always been interested in why it is so hard to get right. Most people have absolutely no idea how limited a game AI really is. Because of that, they ask for things that may simply not be feasible.



It's entirely possible that Amplitude may have done the most damage to the AI performance by both the heavy asymmetrical nature of the factions, combined insufficient focus on the factions to pursuing things that make sense specifically for them. 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.

Slashman wrote:
In your case, what is the basic premise (from an AI POV) of Roving Clans being hostile? I'm still unsure whether a faction that CANNOT declare war is a good idea at all. I personally think a penalty for declaring war is much more sane than never being able to, but that's just me. I've done a lot of talking to developers about AI recently because I've always been interested in why it is so hard to get right. Most people have absolutely no idea how limited a game AI really is. Because of that, they ask for things that may simply not be feasible.

If the AI could handle being sneaky and started burning down their enemies with mercenary armies or goading hostile empires into declaring war on them, I'd say such a faction is a good idea. But for that to work, this AI would need to understand that it needs to mercilessly apply rogues to its perceived threats, instead of being peaceful just because there's a peace or cold war state going on. And that requires a faction-based AI, so I'm on your side with respect to that request for that very reason.

Nosferatiel wrote:
If the AI could handle being sneaky and started burning down their enemies with mercenary armies or goading hostile empires into declaring war on them, I'd say such a faction is a good idea. But for that to work, this AI would need to understand that it needs to mercilessly apply rogues to its perceived threats, instead of being peaceful just because there's a peace or cold war state going on. And that requires a faction-based AI, so I'm on your side with respect to that request for that very reason.

Especially for a game like EL it seems like that would be the only way to go about things, but evidence points to the contrary. Early on when the game first launched, there was a patch that supposedly added random personalities to the AI factions. I immediately wanted to understand why that would even be desirable but never got any answer. Why would you give a chance for a Roving Clans AI to get stuck with an 'aggressive' or 'warlike' personality or for a Necro to get a 'passive' or even 'co-operative' one? But before we even got to that point, factions still didn't even know what special abilities they had as evidenced by the Drakken only getting the ability to use its Force power months later. And it still doesn't use it very well.

Slashman wrote:
I agree that in your scenario, antagonizing you is a bad idea...if I were a human player. There are things you mention though, which I am very uncertain about the AI seeing as a factor. You keeping other armies away may not be a deciding factor in AI behaviour at all. You also say you are not the weakest faction for it to fight with. It can't declare war at all. Maybe you are simply the biggest threat to its own victory that it can perceive. I have no idea if it has contact with all the other factions or not, I'm just saying that you need to be careful of applying human logic to AI in terms of long term thinking. Most game AIs cannot do things that way.



...



In your case, what is the basic premise (from an AI POV) of Roving Clans being hostile? I'm still unsure whether a faction that CANNOT declare war is a good idea at all.

If the AI is unable to recognize why its cities still stand, it´s hopeless AI. If the AI is unable to recognize that cold war means instant defeat, it´s hopeless AI. It doesn´t matter if I´m the biggest threat to some of its many possible victory conditions (which couldn´t be the case unless quest), if it´s unable to recalculate its priorities based on available data, it´s worthless AI.



I don´t think I understand the question. But not being able to declare war only makes it the more important for the AI to be able to pursuit peace when it´s highly, extremely convenient.





edit:

abmpicoli wrote:
OK, concrete case: I had this situation with the Vaulters, last day. The marine had an order to attack a demon, and was 4 tiles away. There was 2 possible tiles, both with exact the same terrain, the same distance to run, and the same end morale bonus, but one was on high ground, and one was on lower... It chose the lower ground tile...

Yes, it just happened to me. My telsem warrior was one tile away from an eyeless one unit, there was a plains on a lower level and a forest on the same level, and it went to the lower one to attack without anyone else around or even close. So weird.

BPrado wrote:
If the AI is unable to recognize why its cities still stand, it´s hopeless AI. If the AI is unable to recognize that cold war means instant defeat, it´s hopeless AI. It doesn´t matter if I´m the biggest threat to some of its many possible victory conditions (which couldn´t be the case unless quest), if it´s unable to recalculate its priorities based on available data, it´s worthless AI.



I don´t think I understand the question. But not being able to declare war only makes it the more important for the AI to be able to pursuit peace when it´s highly, extremely convenient.

I've been saying the AI is hopeless from day 1. Everyone else just said 'turn up the difficulty, you're not playing it at the right level'. Turns out the right level was to give the AI so much of everything that it doesn't have to play the same game you do.



My question aimed to highlight how great some of the factions sound on paper but fail in execution. Nosferatiel summed it up. If the AI cannot handle a faction based around subterfuge and subtlety, why are you expecting it to react and respond intelligently to you doing things like clearing out armies for it?