Concerns about the AI

[akinata]

After looking at some feedback from whom already played AOW4, my nb1 concern is the AI seemingly passive even when at war with the player.
which i am getting planetfall vibes and that's not good.
i don't want to see armies standoffs instead of engaging each others and factions declaring wars on you but never attacks.
and when they do so they send one pitifull stack while keeping all their armies around its main cities...
If an Ai declare war that should means it has what it takes to attack you and will commit to it. not cower in fear until you get powerfull enough to stomp it without much challenge.

i understand that the AI check our powerlevel and do not try to engage us if we have armies equivalent or superior to it's army power. but then it shouldn't declare war out of nowhere.
especially if we are both very far from each other without it having any allies at war with us. (that happened in planetfall)
maybe the AI should still try to commit to an engagement by attacking where most of our troops are not. or still attack us even if the power balance isn't in their favour.

no unit mod system here so we wont have an AI not knowing how to mod its troops correctly. i still hope it will know how to use the new systems in AOW4 and still be a challenge even in late game. snowballing the Ai was too easy whn reaching late game in planetfall.

EDIT: i forgot another important point.
in planetfall the Ai tends to separate its forces to attack weak settlements with bare minimumu forces giving you a high chance of fending them off instead of attacking with a strogn force to ensure victory EVEN when taregting weak settlements.

 

[nvc for the soul]

I share these concerns. Actually it is the only real concern I have. All other things look very, very good (and there are some interesting discussions in the forums to make the game even better). But too passive AI-leaders would be a real gamebreaker for me (and indeed it gives me AOW-PF vibes).

In other threads I shared a few ideas to counter passive AI-leaders. I share one here.

Suggestion: consider Relative Total army-strengths instead of Absolute Individual army strengths.
The main idea here is: the AI should not consider winning a specific battle as an individual leader. Instead AI should consider winning the war, and if allied, it should consider winning the war as a team of allies. This leads to more aggressive behavior, especially on higher difficulty levels. To be specific: it engages in more battles instead of not engaging or fleeing.

When a AI-leader is considering to attack with an army it (most probably) compares its own army strength with that of the opponent. These comparisons can be made in at least two ways:

1. Consider the strength of individual armies
For example:

a. AI-army has 1000 and Human-army has 1500 strength.​

b. Predict the outcome of battle. e.g. AI-army loses entire army (1000) and Human-army loses 2 units with strenght 500.​

c. Outcome: AI doesn't attack because it will (probably) lose.​

I suspect that AI atm compares in a similar way.

2. Compare the ratio of army strength of all armies (in other words: consider relative total army-strengths)
Take the same example as in 1, but slightly change it from step c.

d. Consider the ratio-strength. Eg. before the battle, AI-total army has 10000 and Human-army has 2000. Ratio army-strength is: 5 vs 1 in favor of AI.​

e. Based on the prediction in step b, after the battle AI-total army strength is 9000 and Human-army has 1500. Ratio army-strength is: 6 vs 1.​

f. Outcome: AI attacks because it will be become relatively stronger (the ratio has improved for the AI). Imo this outcome makes more sense than to flee even if the AI loses the battle in this example.​

To simplify 2d&e you can also calculate predicted relative losses: AI loses 1000 of total 10000 (10%). Human loses 500 of total 2000 (25%). Relative losses are lower for AI, so outcome is to attack.

(*edited* to ensure that this way is more or equally aggressive compared to way 1 in all situations, simply add a condition that these additional steps are only taken in case the outcome of step 1c was not to attack. Then only 'non attack' outcomes can change in 'attack' outcomes.)

I suggest to compare in the 2nd way. Some advantages of this type of comparison:

• It gives an opportunity to add Allies (incl. Teamed AI's) into the consideration. Namely in step 2d&2e: calculate the total AI-army strength as the strength of the individual AI plus all its Allies! This makes sense because it is not about winning or loosing for 1 individual AI but about winning or loosing as a team of allies. Based on my experience with AOW3 and PF, the AI-leaders do not take allies/teams into account. This lowers the challenge to the point that there is not a real challenge fighting against a team of 7 AI-emperors, which is unfortunate.
• AI at higher difficulty levels become more aggressive, which make sense. If AI at higher difficulties get more resources, it should be relatively easy for them to have stronger armies overall (at least to the mid-game), as they can produce units all the time in their separate unit queue. Using way 2 of comparison, this should make AI more aggressive at higher difficulties, which make sense. (and it prevents too much aggresiveness on lower difficulty levels)
• I don't think the extra steps are so difficult to implement; these are steps on top of the existing comparison. Ofc there are other considerations that AI has to make, like defending a city; but if you quantify that, you can also quantify it in a relative way instead of an absolute way (similar as explained above in way 2).
• Overall it makes AI-leaders more aggressive on the strategic map (i.e. engage in more battles and not flee) which seems to be needed atm.

 

[TulajMnie]

I wouldn't be too concerned with AI behaviour YET.
The version of the game given for the review was early one, most likely fresh from development. If the Devs got a feedback that AI seems passive I'm sure it will be one of the things to focus during testing and polishing phase.
Only really meaningful thing to do right now (in my opinion) is to hold on theorycrafting untill we see a game version close to release one.
Making assumptions like that may very well hurt the game sales and are simply based on early iteration of the issue. And that's not helping anything. On the contrary.
If the problems remain on release thats a good time to fix them if they are still in the game.

 

[nvc for the soul]

I wouldn't be too concerned with AI behaviour YET.
The version of the game given for the review was early one, most likely fresh from development. If the Devs got a feedback that AI seems passive I'm sure it will be one of the things to focus during testing and polishing phase.
Only really meaningful thing to do right now (in my opinion) is to hold on theorycrafting untill we see a game version close to release one.
Making assumptions like that may very well hurt the game sales and are simply based on early iteration of the issue. And that's not helping anything. On the contrary.
If the problems remain on release thats a good time to fix them if they are still in the game.

I am afraid it is not only theory crafting, because this passiveness was apparant in AOW-PF and partly present in AOW3 as well. So, it seems to be a trend and imo it is the major weakness of the game for Single Player (and still the AOW-series is my favorite of all time).
I share the concern not to hurt sales but because I want the game to succeed badly. And if I can help that happen, I gladly do. That is the intention of my post.

The suggestion I made will hopefully also help to mitigate the team-issue in previous AOW's: AI-teams seem to make no impact on AI-behavior at all. I think that is unfortunate. AI's in teams can and should be more risk-taking imo. I imagine this issue can be easily overlooked by the devs. So my intention is also to get attention about that issue and to give a possible solution.

Ofc we can only really know about AI after we have tried it. So, yes, there are assumptions being made, but that is no reason not to post this concern imo.

 

[TulajMnie]

I am afraid it is not only theory crafting, because this passiveness was apparant in AOW-PF and partly present in AOW3 as well. So, it seems to be a trend and imo it is the major weakness of the game for Single Player (and still the AOW-series is my favorite of all time).

I very much believe in your good intentions pet me start with that.
And I agree AI wasn't the best in previous games, hope it will be better this time around.
Also couldn't agree more that bad AI makes solo play extremely frustrating - make it passive all challenge is gone, make it stomp player and there is no fun in even trying.

I share the concern not to hurt sales but because I want the game to succeed badly. And if I can help that happen, I gladly do. That is the intention of my post.

My comment wasn't meant as a way to shut the conversation nor invalid your ideas. It was a general thought towards similar thought experiments that may be less gentle towards the balance of the earlier game versions.

The suggestion I made will hopefully also help to mitigate the team-issue in previous AOW's: AI-teams seem to make no impact on AI-behavior at all. I think that is unfortunate. AI's in teams can and should be more risk-taking imo. I imagine this issue can be easily overlooked by the devs. So my intention is also to get attention about that issue and to give a possible solution.

Thats a valid point!

Ofc we can only really know about AI after we have tried it. So, yes, there are assumptions being made, but that is no reason not to post this concern imo.

Couldn't agree more!

 

[Ferrus Animus]

So I consider the AI's aggressiveness based on relative unit strength based on observations. Like losing units and having the AI suddenly move from its capital, or seeing the AI cross water with a big army, only to turn right back as new units join my army.

 

[nvc for the soul]

So I consider the AI's aggressiveness based on relative unit strength based on observations. Like losing units and having the AI suddenly move from its capital, or seeing the AI cross water with a big army, only to turn right back as new units join my army.

I am not sure what message you are trying to convey. Can you explain a bit more?
If you are suggesting that aggressiveness is also about movement over 2+ turns (and not only about decisions to start a battle) then that is certainly a valid point. Or are you suggesting that my suggestion leads to more strange AI-movement? If so, can you explain why a bit more?

My comment wasn't meant as a way to shut the conversation nor invalid your ideas. It was a general thought towards similar thought experiments that may be less gentle towards the balance of the earlier game versions.

Thanks for clarifying! I partly misinterpreted your comment. (btw my suggestion may get invalidated; I am fine with that as long as AI-passiveness is gone)

 

[Ferrus Animus]

I am not sure what message you are trying to convey. Can you explain a bit more?
If you are suggesting that aggressiveness is also about movement over 2+ turns (and not only about decisions to start a battle) then that is certainly a valid point. Or are you suggesting that my suggestion leads to more strange AI-movement? If so, can you explain why a bit more?

Yes.
In Planetfall, if the AI is hostile, but doesn't want to attack they keep their units back at their cities/capital.
If they do want to attack they start moving them (usually multiple stacks near each other) toward their attack target.

If you scout them well (or play certain kinds of MP so someone else can see them), you can observe them.
If that calculation changes, the AI can then move back, or start moving.

 

[Xerberus]

in the underground dwarf video - which was after or during the playtime of the youtubers / previewers - the devs called the AI "ropey AI" ... its just not finished yet, which is normal.

i think we will get servicable AI at release and with updates getting improved AI. during the dlc phase getting more polished i bet too.

i hope we will see a more stellaris-like dlc model, would be good for the AI developement too.

 

[nvc for the soul]

Yes.
In Planetfall, if the AI is hostile, but doesn't want to attack they keep their units back at their cities/capital.
If they do want to attack they start moving them (usually multiple stacks near each other) toward their attack target.

If you scout them well (or play certain kinds of MP so someone else can see them), you can observe them.
If that calculation changes, the AI can then move back, or start moving.

Ok, so during a journey towards a human target (movement of 2+ turns) the AI can change its mind if the situation changes. That can be sound behaviour, right?
To make a connection with strength comparison: let's assume that during a journey towards a human target the AI evaluates the situation each turn. Comparing army strengths is most probably part of this evaluation. Then you can compare these strengths in different ways (as explained above):

1. Compare the strengths of the individual AI army with the target army. If during the journey the target's army strength changes (eg. target army becomes stronger), it may reconsider and return home or choose another target.
2. Compare the ratio of army strength, or simpler, compare the predicted relative losses. If during the journey these predicted relative losses changes (eg. the predicted relative losses of AI become greater than the human's), it may reconsider and return home or choose another target.

So, yes the AI can change its mind and flee back home in both types of comparison. But, in case of big relative differences in total army strength (AI and its allies have far more army strength than the human) it seems to be more difficult to change AI's evaluation to attack when using way 2; thus making AI more willing to proceed on the journey to the human target. Or simply put: AI is more willing to move to human targets, because it is more willing to accept losses (because losses are relatively less impactful for a strong AI).

As an addition: to ensure that AI-behavior is more or equally aggressive compared to way 1 in all situations, simply add a condition that the additional comparison of predicted relative losses is only done in case the outcome of way 1 was not to attack. Then only 'non attack'-outcomes can change in 'attack'-outcomes.

 

[Xerberus]

i don't think the Ai in planetfall is bad per se, i do think though they need more buffs and be more agressive. but of course that is personal preference.

we have to consider though, that good AI doesn't necessary mean fun AI. a good AI might hold back and not attack because it doesn't see itself winning, but that won't make for a fun game would it. the AI should pose somewhat of a thread and make it entertaining to play against.

when i was younger i wanted the AI in the games to stand on equal grounds because i wanted it to be a challenge of wits and not cheating buffs. nowadays i actually prefer the AI gets some preferential treatment so that i have a better challenge. i am fine if the AI gets free resource and sometimes some units for free and has less problems with its economy and maybe slight movement bufs or whatever. as long as the gameplay is believable and the "cheats" not so obvious i am fine with it because the thing i want most is a fun experience.

its a game and in singleplayer you are playing with yourself basically and literally. yes the program gives you so called "AI" to fight against but in essence those are parameters you try to fight, it doesn't have to be on equal grounds with you. i see the features and mechanics of a game more for the player than for the AI. because again i want somewhat of a challenge and be needed to use all mechanics, if the AI opponent isn't good enough then the mechanics don't matter in a strategy game because you don't need any strategy against bad opponents.

 

[nvc for the soul]

i don't think the Ai in planetfall is bad per se, i do think though they need more buffs and be more agressive. but of course that is personal preference.

I think AI-leaders are way too passive on the strategic map in Planetfall. I also don't see AI play as a team when it is allied to others. Now it is of course extremely difficult to make AI really play as a team, but the general principle is rather simple:

If an AI-leader plays in a team, it can take more risks, because it is about winning the war as a team of allies, and not about preventing loosing individual armies.​

And this can be implemented (I think) with a different army comparison where it is about weighted losses (as explained above).

we have to consider though, that good AI doesn't necessary mean fun AI.

True. Actually I think that fun is the most crucial thing in the game.
Fun has of course a subjective notion:

• For me, it is not fun when AI-leaders are too passive and (almost) never attack my armies or my cities.
• For me, it is fun to have a decent challenge in the game.
  • For me, it seems logical that the challenge is higher on higher difficulties and when playing against AI-leaders in teams.
  • I think there is room for improvement in AI-leader behavior on the strategic map to increase the challenge, in particular: be more aggressive for higher diffculties and when they have allies.

a good AI might hold back and not attack because it doesn't see itself winning, but that won't make for a fun game would it. the AI should pose somewhat of a thread and make it entertaining to play against.

Exactly, it should be a thread. And this thread can be higher with higher difficulties, and when AI-leaders are in teams/allied.

when i was younger i wanted the AI in the games to stand on equal grounds because i wanted it to be a challenge of wits and not cheating buffs. nowadays i actually prefer the AI gets some preferential treatment so that i have a better challenge. i am fine if the AI gets free resource and sometimes some units for free and has less problems with its economy and maybe slight movement bufs or whatever. as long as the gameplay is believable and the "cheats" not so obvious i am fine with it because the thing i want most is a fun experience.

I am totaly fine with buffs for AI as well, in all kinds of ways!

 

[Sifer2]

It seems to be a hard thing to get right. Lot's of games in the genre suffer from either AI being too passive or too anti-player with them all dog piling on you. I feel part of the problem is the AI in these games declare war because they want to raid you, and try to take your undefended cities. Then when you suddenly build defensive armies, and rally your main army home to defend the AI realizes it can't win so just camps in it's own cities. This then gives the player the impression the AI is broken declaring war for no reason then never attacking. AI in Total War Warhammer is especially bad about that. I noticed in Planetfall at least a lot of time the AI immediately attacks with a big force not long after the war declaration though so i'm not as worried.

 

[akinata]

It seems to be a hard thing to get right. Lot's of games in the genre suffer from either AI being too passive or too anti-player with them all dog piling on you. I feel part of the problem is the AI in these games declare war because they want to raid you, and try to take your undefended cities. Then when you suddenly build defensive armies, and rally your main army home to defend the AI realizes it can't win so just camps in it's own cities. This then gives the player the impression the AI is broken declaring war for no reason then never attacking. AI in Total War Warhammer is especially bad about that. I noticed in Planetfall at least a lot of time the AI immediately attacks with a big force not long after the war declaration though so i'm not as worried.

more like, a big force arrive nearby your borders, but stays there and never attack. except for a few isolated low tier stacks trying to raid your undefended cities you will easily kill because the AI tried to attack with the bare minimum instead of sending strong forces to take weak settlements.
thus making the Ai unable to effectively take anything from you.

 

[nvc for the soul]

To quote an experience from someone in a Steam-discussion:

TungstenCarbide 8 Mar @ 5:11am

I mean tbf the AI on Planetfall was pretty darn simple to beat haha. Going 1 on 7 (extreme difficult) was not an overly difficult task because it never really wanted to attack (I think it had something with wanting to both defend itself too or something). But seeing if there are any improvements would be nice!

I have the same experience with 'going 1 on 7' (meaning 1 against a 7 teamed AI on highest difficulty) i.e. it never really wanted to attack. Personally I think it has to do with only taking the individual AI into account and not the allies/team (see previous posts).

It seems to be a hard thing to get right. Lot's of games in the genre suffer from either AI being too passive or too anti-player with them all dog piling on you.

Yes, solid AI-behavior on the strategic map is probably the most difficult thing to make in the entire game. I am already satisfied when it is 'decent' in the same sense that AI is decent in the tactical-combats. And decent includes taking allies into account imo.

I feel part of the problem is the AI in these games declare war because they want to raid you, and try to take your undefended cities. Then when you suddenly build defensive armies, and rally your main army home to defend the AI realizes it can't win so just camps in it's own cities. This then gives the player the impression the AI is broken declaring war for no reason then never attacking. AI in Total War Warhammer is especially bad about that. I noticed in Planetfall at least a lot of time the AI immediately attacks with a big force not long after the war declaration though so i'm not as worried.

Ok, but when you play against a team of 7 AI's you start with war already declared and this doesnot change during the game. Even in these situations the AI almost never attack, which is illogical and should be solvable (as I tried to explain in an earlier post).

 

[Mechalibur]

That's interesting, the Planetfall AI would beeline for my capital pretty frequently after declaring war on me in my experience. Maybe something about being in a team of 7 caused their behavior to act up in weird ways or something?