Chiming in to say that I also think getting a competent and competitive AI is the top priority, far above getting a narrative AI.



In fact, if a faction's strengths in game match with their lore, an AI that plays competently and to win will most likely act "in character" anyway, and at that point the narrative/role-playing AI is more a matter of presentation. The Roving Clans might cite economic disputes when they declare war on you, the Drakken might present themselves as peacekeeper, and the necrophages just think you're crunchy and taste good with ketchup, but under the hood they all just want what you have in order to win.

Granted, the "good" factions might be more reluctant to stab a long-term ally in the back just because he's getting ahead in points or an easy target, but other than that all AI should always aggressively try to win (not necessarily by military might).



That said, making the AI more communicative could help by reducing the number of reports and complaints about dumb AI when the AI is actually working fine. They might just have a legitimate reason for that unexpected marketban...

I'll chime in because i feel it needs to be said.

On the point of pshychic ai, I feel that that may be a step too far. I am however in favor of information hoarding ai.



In my opinion, the AI NEEDS to have all the information the player has. Not simply respond to what goes on in the game.



A few things I feel the ai should know:



1. Any time an army gets within its line of sight, it should look at and remember that army's makeup. From what units are in it to what they appear to be equipped with. This information should be remembered for later reference so it has some idea of what the other empire's army structure looks like (and can plan a possible counter set up).



2. AI should at all times be aware of the game score and know which empires are doing well of which fronts. This can serve as a way of guessing an empire's military strength even if it hasnt seen any of the units they use.



3. AI should have hard coded into memory the special abilities and unique units of every empire along with the general stats of every minor faction and minor faction unit.

(For example, if it sees that there is a roving clans empire playing, it should also know what it's basic military units look like, it should know the clan CANNOT declare war and so are more likely to be friendly (whether they are looking for a friend or just want to know the clans can be bullied with little fear of military retaliation is up to the ai). They should know that the clans will LIKELY shoot for an economic victory and so will want to be cooperative with mutually beneficial trade routes. Ai should know all of the clans unique tech and what that makes their cities good at and what abilities that tech gives them and bla bla bla)

ALL THIS INFORMATION should be know to the AI from the BEGINNING and it should be able to respond to it. (For example, if everyone is in the 2nd age, and it sees a mercenary army running around, it should know this likely belongs to a roving clan empire because they alone can create these this early in the game.)



4. AI should be aware of what empires are on the map (like the player is assuming nothing was left random) and, with that information, be able to do some slight planning from the very beginning of the game. (Information like: on this map, there is an orange necrophage empire. Necrophages are generally very war crazy. Though I have not made contact with them yet, I do see on the edge of my fog of war an orange border. Therefore there is a necrophage city nearby, they may not yet be aware of my presence, I should plan accordingly.)



5. AI should have its quest line generally memorized. (It should know that, after I do X, i will have to do Y, I should take some steps now to ensure that, upon doing X, I can quickly do Y without much time passing by. After I do that, i will then be asked to do Z. how long could it take me to do that? Should I begin preparing for Z now if I can?) And it should have memorized what sort of rewards it will get for doing each step of its quest line so it can prioritize what to get done even if it isnt going for a quest victory.



6. AI should be aware of upcoming winter or the approaching spring so it can prepare a bit ahead of time instead of only reacting as the season changes



7. Ai should have memorized the branches for hero advancement and so know what sort of heroes it wants to buy when looking for more. (and know what strengths enemy heroes may have)



8. Ai should know at all times the diplomatic relations of all empires and be able to react accordingly. Two nations at war that you dont like? Are there any shenanigans you can pull to ensure they kill off eachother's armies to allow you to conquer both of them in one fell swoop?



9. AI should be aware of what other empires know. (This empire has seen my armies, they now know my general army structure, they may plan accordingly should i go to war with them in the near future. I should consider modifying my army in a manner that counters what they my do If I want to go to war with them. Or, I made a trade request with another empire asking for strategic resources, knowing what can be made with those resources, they may know what I am planning to do with those resources and plan accordingly.)





... i could think of more, but i think this gets my point across. Basically, AI needs to know everything about every faction in terms of special abilities, military units and general play style and should be well aware of any information the player has access to and can respond accordingly.

BPrado wrote:
why give it knowledge that the players doesn´t have?

That is the question...



It's to provide the human mind with an interesting challenge.



AIs can beat Chess grandmasters, yes, but EL is more like Go than Chess, and the finest academic minds on the planet are still struggling to build AIs that can win at Go. Struggling with super-computers. And don't forget that in Go you can see the entire state of the board at all times, whereas in EL you can't. You might be thinking "but an AI won at Jeopardy recently!" and you'd be right. But said computer was a cluster of ninety 3.5 GHz eight core processors. The average end-user doesn't have that kind of hardware, and while we have some truly brilliant programmers on staff we're not 50 years ahead of the state of the art.



The games industry has a long tradition of pretending that we're magic and this is toxic, I think, both for developers and for the players. It is toxic for the games industry as a whole. Hence G2G. G2G is about trying to break this cycle of hype and disappointment by including players in the process and treating them with respect and honesty.



And so the honest truth is that you just can't have your cake and eat it too: just about all strategy game AIs cheat in one way or another, the better ones are just better at hiding the fact that they cheat.

just about all strategy game AIs cheat in one way or another, the better ones are just better at hiding the fact that they cheat.

Which is why telling everyone you're going to cheat isn't a popular idea



What about amping up aggression against players to raise difficulty?



The more "difficult" the AI is the more aggressive they behave toward the human empire(s). At the highest difficulty, the AIs are out to get you from turn 1...

And so the honest truth is that you just can't have your cake and eat it too: just about all strategy game AIs cheat in one way or another, the better ones are just better at hiding the fact that they cheat.

Propbuddha wrote:
Which is why telling everyone you're going to cheat isn't a popular idea



What about amping up aggression against players to raise difficulty?



The more "difficult" the AI is the more aggressive they behave toward the human empire(s). At the highest difficulty, the AIs are out to get you from turn 1...

I think realistically we understand that some cheating by the AI is required (resource boosts etc). To me, however, that shouldn't mean endless stacks of units marching my way everyturn. Ideally you give the AI enough "cheats" to make better decisions, to make better armies, to be harder to beat without being unrealistic (example below)



One example: The army fighting issue could all come down to how the AI builds armies and handles them in battle. Giving the ENDLESS AI resources bonuses (cheating) allows the AI to build 4-5+ stacks for each 1 of mine (maybe more). But if my 1 stack can beat those 5 AI stacks, then the higher difficulty becomes nothing more then a time sink and irritant. I would like to see the AI with 2-3 extremely well equipped well designed armies (with heros) instead of the endless stacks of fodder that get thrown at me.



What I would like to see is an incrementally scaling difficulty in AI unit design, battles, etc with difficulty (instead of just more units at higher difficulties), an example below...

(Forgive me, I don't completely remember difficulty scales..., I just wanted to give an example of how scaling across difficulties could be implemented)



Noob: 0.5 Stack Count, Weaker Weapons

Easy: Normal Stack Count, Weaker Weapons, Heros

Normal: Normal Stack Count, Highest Unlocked Weapons, Heros with Random Accessory Boosts

Serious:1-2X Stack Count, Highest Unlocked Weapons, Damage/Initiative Rings, Heros with Damage/Initiative Boost Accessories in one stack

Impossible: 2-3x Stack Count, Highest Unlocked Weapons, Damage/Initiative Rings, Heros with Damage/Initiative Boost Accessories in one stack

Endless: 2-3x Stack Count, Highest Unlocked Weapons, Damage/Initiative Rings, Heros with Damage/Initiative Boost Accessories in EACH stack

wilbefast wrote:
And so the honest truth is that you just can't have your cake and eat it too: just about all strategy game AIs cheat in one way or another, the better ones are just better at hiding the fact that they cheat.

Can you please tell me the cheats that the AI in AoW 3 uses on Knight difficulty? Or the ones used by Fallen Enchantress: LH on the equivalent setting? Because if, in direct opposition to what the devs claim, those games cheat at the difficulty where they are not supposed to cheat, then they do it in a much more clever and believable way than EL does. And I think we can all see it.



Endless stacks of poorly designed armies. Bad tactical choices and behavior. Poor army composition even when they have assimilated useful minor factions. Those things are still present in the game. Giving huge bonuses to the AI just makes beating them a slog, not a strategic or tactical endeavor. So even without looking at cheats, there are still many areas that the game needs to be tightened up.



Most of us are not simply saying never make it cheat. We're saying tighten up the obvious areas of weakness first and then look at where cheats are unavoidable. Because currently, the bonuses on higher levels don't enhance the AI, they just make it able to do impossible stuff in spite of its failings. That is extremely jarring for a player. It makes the AI look really silly.

@Wilbefast



Can you give us any idea of what can actually be changed with the Endless Legend AI? We talked about it not being possible (I know everything is possible, but I take it you don't really want to put too much effort in without getting payed. I fully understand that this is business) to actually change the AI from diplomatic to dictatorial just by patching.



Which parameters are you willing/able to change?

Slashman wrote:
Can you please tell me the cheats that the AI in AoW 3 uses on Knight difficulty? Or the ones used by Fallen Enchantress: LH on the equivalent setting? Because if, in direct opposition to what the devs claim, those games cheat at the difficulty where they are not supposed to cheat, then they do it in a much more clever and believable way than EL does. And I think we can all see it.



Endless stacks of poorly designed armies. Bad tactical choices and behavior. Poor army composition even when they have assimilated useful minor factions. Those things are still present in the game. Giving huge bonuses to the AI just makes beating them a slog, not a strategic or tactical endeavor. So even without looking at cheats, there are still many areas that the game needs to be tightened up.



Most of us are not simply saying never make it cheat. We're saying tighten up the obvious areas of weakness first and then look at where cheats are unavoidable. Because currently, the bonuses on higher levels don't enhance the AI, they just make it able to do impossible stuff in spite of its failings. That is extremely jarring for a player. It makes the AI look really silly.

Hmm... well I wanted to refer you to Metalynx's post but it seems you replied similarly to that one too, and I can't find any citation which shows cheating in those games on normal difficulty. So ok, it's a fair cop - you got me



There is room for improvement in EL's AI before we start considering cheating more or differently and yes, some games out there have managed better performance without or with little cheating.



I suppose I'm getting a bit ahead of myself, thinking about what far-reaching structural improvements could really make the AI pose a serious challenge. There are plenty of little things to consider before we get there though.



edit:

clykke wrote:
@Wilbefast



Can you give us any idea of what can actually be changed with the Endless Legend AI? We talked about it not being possible (I know everything is possible, but I take it you don't really want to put too much effort in without getting payed. I fully understand that this is business) to actually change the AI from diplomatic to dictatorial just by patching.



Which parameters are you willing/able to change?

Well it was mentioned here, here and here that the AI is exploitable in diplomacy so improving this subsystem has pretty high-priority. The AI should at least check the market to calculate the base value of the strategics/luxuries it is buying/selling. Ideally it should modify it's price based on who it's selling to, but that would require structural change. Some structural changes may be possible though, provided they're small and not too numerous!



Second it would be good to take a look at the how military strength is calculated - this is used both for the battle result prediction (shown to the player) and by the AI, and it's sometimes off. Improving this would therefore mean killing two birds with one stone. Likewise better AI target-prioritisation would be good both for players (as it determines how default targets are assigned) and AIs, though this point is a lot trickier than it sounds - the rabbit-hole goes pretty deep.



Finally there have been a number of requests to make the AI aware of impending victory conditions, so I'll see what I can do about that - the issue is of course that if the AIs start trying to stop the winning empire then diplomatic victories might be made impossible.

wilbefast wrote:
Hmm... well I wanted to refer you to Metalynx's post but it seems you replied similarly to that one too, and I can't find any citation which shows cheating in those games on normal difficulty. So ok, it's a fair cop - you got me

Well I wasn't trying to 'get' you. If it came across that way, my apologies. I just wanted to be sure we were on the same page in terms of the whole cheating thing. I'm not opposed to bonuses either. I just don't want to see bonuses be the universally applied band aid for AI performance failings.

There is room for improvement in EL's AI before we start considering cheating more or differently and yes, some games out there have managed better performance without or with little cheating.



I suppose I'm getting a bit ahead of myself, thinking about what far-reaching structural improvements could really make the AI pose a serious challenge. There are plenty of little things to consider before we get there though.

I am not sure how much you may have considered what I proposed a while back with giving the AI access to the army loadouts used by players and making those templates for better design once they have similar access to the resources required. That happens in Legendary Heroes and it can be pretty effective. This would mean the game would have to keep a database of the loadouts that the player has used in past games.

wilbefast wrote:
Hmm... well I wanted to refer you to Metalynx's post but it seems you replied similarly to that one too, and I can't find any citation which shows cheating in those games on normal difficulty. So ok, it's a fair cop - you got me



There is room for improvement in EL's AI before we start considering cheating more or differently and yes, some games out there have managed better performance without or with little cheating.



I suppose I'm getting a bit ahead of myself, thinking about what far-reaching structural improvements could really make the AI pose a serious challenge. There are plenty of little things to consider before we get there though.



edit:





Well it was mentioned here, here and here that the AI is exploitable in diplomacy so improving this subsystem has pretty high-priority. The AI should at least check the market to calculate the base value of the strategics/luxuries it is buying/selling. Ideally it should modify it's price based on who it's selling to, but that would require structural change. Some structural changes may be possible though, provided they're small and not too numerous!



Second it would be good to take a look at the how military strength is calculated - this is used both for the battle result prediction (shown to the player) and by the AI, and it's sometimes off. Improving this would therefore mean killing two birds with one stone. Likewise better AI target-prioritisation would be good both for players (as it determines how default targets are assigned) and AIs, though this point is a lot trickier than it sounds - the rabbit-hole goes pretty deep.



Finally there have been a number of requests to make the AI aware of impending victory conditions, so I'll see what I can do about that - the issue is of course that if the AIs start trying to stop the winning empire then diplomatic victories might be made impossible.

This is a good beginning in my opinion. I understand the problem with diplomatic victories, and I don't really have a good solution. I've never liked that kind of victory. It differs from other victory conditions because it relies on your opponents cooperating - something they should not do when when you are about to win. Frankly I'd rather you remove this type of victory if it is required to make the AI more competitive, but I am also aware that that will most likely never happen.



You didn't mention changing overall priorities of the AI factions (more focus on safety/military strength). Is that because it is not doable, or because you don't feel it is necessary? I think my main problem with Endless Legend is how military conquest is basically always the fastest/easiest way to success. The thing about military might in most 4x games, is that it basically nullifies everything else. It doesn't matter if you have the most pop, dust, science, diplomatic points or are almost through the quest line, if someone can take it all away from you in a few turns. Constructing anything within your cities without having the tech, strategic resources or units to defend it, means you are basically building it for your neighbor, who will just claim it whenever he wants.



Building units does two things:

It defends your territory, and it allows you to go on the offensive.



It doesn't really weaken my economy compared to my opponents, because if they do not match my military power, I will defeat them in short order.



EDIT:

Also, the AI should, if possible, be more focused on having very strong units than alot of units. It is very easy to win even if you are outnumbered 10 to 1 as long as you units are much stronger individually. This is mainly due to the limited number of units in armies and the slow reinforcement. Once the initial army is defeated, every single reinforcement coming joining the battle is going to die basically without moving.

clykke wrote:
This is a good beginning in my opinion. I understand the problem with diplomatic victories, and I don't really have a good solution. I've never liked that kind of victory. It differs from other victory conditions because it relies on your opponents cooperating - something they should not do when when you are about to win. Frankly I'd rather you remove this type of victory if it is required to make the AI more competitive, but I am also aware that that will most likely never happen.

A solution might be to allow an Allied Victory condition. This is an option in AoW 3.



However, you are right. Diplomatic victories are very unrealistic when you have factions competing. I think that diplomatic victories as a mechanic might need some serious reworking. If not in EL, then in future titles.



A culture/religious victory makes more sense than a diplomatic one. But we don't have that option in the game either.



If we had the internal faction system from ES 2 in EL, I could see diplomatic victory being a really powerful threat. If you can convince the peaceful internal factions of another empire that it is entirely in their best interest to be friends with you, you'd have a convincing means to dominate, whether the race leader wanted war or not.

Slashman wrote:
A solution might be to allow an Allied Victory condition. This is an option in AoW 3.



However, you are right. Diplomatic victories are very unrealistic when you have factions competing. I think that diplomatic victories as a mechanic might need some serious reworking. If not in EL, then in future titles.



A culture/religious victory makes more sense than a diplomatic one. But we don't have that option in the game either.



If we had the internal faction system from ES 2 in EL, I could see diplomatic victory being a really powerful threat. If you can convince the peaceful internal factions of another empire that it is entirely in their best interest to be friends with you, you'd have a convincing means to dominate, whether the race leader wanted war or not.

Allied victories should be available as an option. I don't really know how I feel about them. They don't always make sense (in a game of 8 players if two players are far ahead they can just ally and basically ensure victory) but in other cases (when AI stomping with a friend) it would be nice to have.



The diplomatic victory should in my opinion 1) not rely on cooperation of opponents and 2) make some sense outside the game. It should symbolize a point where you are both very far ahead or at least strong, and at the same time "liked" by - if not all the other leaders - then at least the general population of the world.



I am thinking it should be a form of high score (time) victory hastened by upholding peace and sharing wealth/knowledge.



My idea is to give diplomatic points for handing out gifts in form of techs/dust/resources/units to the weakest empires and by the ability to enforce peace by military might. Make it possible to demand a truce, and if not agreed upon, you can enter the war on the non aggressors side. If you win, truce is signed and you get diplomatic points.



Winning diplomatically would require you to be ahead or at least in a good position (needs to be tested and balanced), and have earned enough diplomatic points by sharing your wealth and securing peace (going to war against anyone as the aggressor should subtract points).



This way you do not rely on rivals to cooperate and the victory makes more sense in a context outside the game. Basically you become the police of Auriga, taking care of those in need, and secures a balance of power between the different factions, striking hard against any leader who tries to prey on the weak. At the same time you run the risk of enemies using your gifts against you, and must try to stay relevant/strong despite handing out techs/dust/units.



Obviously this is just a rough sketch, but I believe it could work very well if balanced and tweaked correctly, and it would be a fun and different style to play once in a while.

Slashman wrote:
A solution might be to allow an Allied Victory condition. This is an option in AoW 3.

I suggested a similar concept a long time ago...



Since it is possible for multiple players to acheive victory on the same turn, this probably isn't that difficult to implement. It's basically a group Expansion victory.







This of course can be disabled (just like the others) for player who are uncomfortable with the concept...

wilbefast wrote:
I'll see what I can do about that - the issue is of course that if the AIs start trying to stop the winning empire then diplomatic victories might be made impossible.

Yeah, but as you've said elsewhere (of something I can't remember), that's a game design issue, not an AI issue.



Note that in multiplayer games, diplomatic victory is just a way for players to give the win to somebody in lieu of somebody they really, really hate-- it's saying, "Okay, I won't win, but neither will you." And it's limited in the speed with which it can give that win away, so that's not even an unreasonable behavior for the AI, at least on high difficulty.



In practice, I've never seen diplomatic victories really enter play in multiplayer games except via spam exploits or, in mixed games, AI exploits.



Which kind of reminds me of something that's easy to forget, so I'm going to bold it:



Most multiplayer games end up with an AI at some point due to drops, and some players consider games ruined by this because AI are so exploitable. AI affects multiplayer games and it's a good reason to focus on a competent, self-interested AI over a story-based AI.

Here's AI problem.



That's the deal I'm having with Cultists.



Let's sweeten it with some glassteel.



It seems unreasonable. Let's try giving dust itself.



The queen doesn't seem to be thinking straight. Check out market price. She values 101 glassteel the same as 15 dust even though she could take 101 glassteel, sell 1 for 15 dust and have 100 glassteel more.



I understand you can't calculate the price easily as it fluctuates when you sell or buy it but come on. You can have rough estimation, one of those things is clearly better than the other. And just in case you're wondering Roving Clans are not in the game so she has access to the market.
Trading Lords - Turn 104.zip

wilbefast wrote:
That is the question...



It's to provide the human mind with an interesting challenge.

...

And so the honest truth is that you just can't have your cake and eat it too: just about all strategy game AIs cheat in one way or another, the better ones are just better at hiding the fact that they cheat.

Ok, I agree. Presented with the same amount of information, it´s only natural that a human mind will find "better", or shorter logical connections between all that than a game AI - we´re supposed to both because of sheer processing power and because we´re bound to a much broader and more complex external design than algorithms.



So I guess any feasible solution for the short term in order to improve the AI would have to go through forms of cheating; and then we should maybe start discussing which of these logical routes that make our game more efficient can be simulated by an algorithm to look like the AI is doing the same thing, when it´s actualy mildly cheating (just in order to look human, since it´s fundamentally not and we want it to be).

BPrado wrote:


So I guess any feasible solution for the short term in order to improve the AI would have to go through forms of cheating; and then we should maybe start discussing which of these logical routes that make our game more efficient can be simulated by an algorithm to look like the AI is doing the same thing, when it´s actualy mildly cheating (just in order to look human, since it´s fundamentally not and we want it to be).

Perhaps one partial solution in giving the ai an edge would be to make it inherently more lucky when it comes to what it finds in ruins?

Would making roving armies slightly more antagonistic towards human players help? (Acting as unofficial and unreliable body guards for ai cities if the ai has no need of killing them for exp?)

What if the AI were more willing to make fair trades with fellow AI if they are at peace or allied (Ai is willing to help build up other AI for the sake of making things harder on the player, kind of like how friends will often unofficially team up to wipe out a third party while simultaneously trying to outdo each other)?

HappyHead wrote:
Perhaps one partial solution in giving the ai an edge would be to make it inherently more lucky when it comes to what it finds in ruins?

I don't think this will help. The reason being that when given the large bonuses of higher difficulty levels, the AI still just swarms you with endless stacks of bad armies. It doesn't play smarter the more stuff it gets, it just gets to skip playing the same game as the player.

Would making roving armies slightly more antagonistic towards human players help? (Acting as unofficial and unreliable body guards for ai cities if the ai has no need of killing them for exp?)

I think the Roving Clans needs some attention across the board. It is a difficult faction for an AI to play. It has a huge hurdle in not being able to go on the offensive when it wants.

What if the AI were more willing to make fair trades with fellow AI if they are at peace or allied (Ai is willing to help build up other AI for the sake of making things harder on the player, kind of like how friends will often unofficially team up to wipe out a third party while simultaneously trying to outdo each other)?

This sort of thing should be in the game already IMO. AI factions should interact with each other to further their own goals. For instance, in Fallen Enchantress Legendary Heroes, I have had an AI faction pay off another to eliminate me. The same sort of things should happen in EL. The Cravers would be a perfect faction to hire to kill another faction. They can't have peace anyway.

ilitarist wrote:
Here's AI problem.



That's the deal I'm having with Cultists.



Let's sweeten it with some glassteel.



It seems unreasonable. Let's try giving dust itself.



The queen doesn't seem to be thinking straight. Check out market price. She values 101 glassteel the same as 15 dust even though she could take 101 glassteel, sell 1 for 15 dust and have 100 glassteel more.



I understand you can't calculate the price easily as it fluctuates when you sell or buy it but come on. You can have rough estimation, one of those things is clearly better than the other. And just in case you're wondering Roving Clans are not in the game so she has access to the market.

The AI doesn't yet consider the market value of resources-

wilbefast wrote:


Well it was mentioned here, here and here that the AI is exploitable in diplomacy so improving this subsystem has pretty high-priority. The AI should at least check the market to calculate the base value of the strategics/luxuries it is buying/selling. Ideally it should modify it's price based on who it's selling to, but that would require structural change. Some structural changes may be possible though, provided they're small and not too numerous!

In general making the AI less susceptible to this kind of exploit is something we're very keen on doing!

Slashman wrote:




I think the Roving Clans needs some attention across the board. It is a difficult faction for an AI to play. It has a huge hurdle in not being able to go on the offensive when it wants.

Sorry, I meant the wandering armies, y'know, from the minor factions? What if they were just inherently more nice to AI?