Awesome! I hope to see some changes to diplomatic AI decisions and players impressions. Not every faction needs to play like a vulture.

Indeed you lose the flavour and theme with all factions using the same generic diplomacy A.I.I am surprised as GC2 did such a better job.

That's really great news. I've put suggestions at other points in the forum, but to summarise them all in one place:

1) Please have standard skill upgrade paths for AI heroes to use so that it builds its heroes more effectively. E.g. death magic summoner, bow using assassin, club using warrior, etc. etc.

2) If the AI is not always building in all its cities (as is apparently the case from the Fog Of War thread) it certainly should be! It should also be using all its Essence slots all the time if possible, with Meditation as a fallback option if nothing else is suitable. The Fog Of War thread suggested that it seemed to only be adding one city spell at a time, there isn't really any reason to do this.

3) The AI should put units within one move of each other into maximum size stacks if it can. This is most obvious after Tornado, but it also seems to happen when it's sending reinforcements.

4) The AI should target Outposts and loot more. In general, if it has a target of opportunity (e.g. it has a more important target but comes within one move of something it can take) it should get the thing within one move. If it gets to its goal it should immediately get another goal (so it can use up movement points). Settlers who arrive at their desired spot should immediately build, and the AI should then decide what it can do with that city or Outpost.

5) The AI should prioritise production and unrest production buildings more, especially on higher difficulty levels. I understand that different AI personalities may have different priorities, but in general they don't seem to prioritise production as much as they should. I capture AI cities after 100 turns and they still haven't built workshops.

6) In normal circumstances the AI should only train units in fortresses.

7) If the AI has just lost a lot of units it should add unit training to the front of its queues.

8) If its in defensive mode (it has a large enemy stack in its borders) in an ideal world it would use a different flavour of strategic AI, where it focuses entirely on defending its cities, training more units, building defensive buildings, using strategic spells to immobilize the danger stack, and try to build up a powerful enough stack or stacks to repel the invaders. In general it shouldn't move its units from its cities unless they're helping to defend against the danger.

9) Is the problem with creating mounted troops simply that it doesn't have unit designs for the mounted troops? In general I can't see any reason not to build its normal units with wargs or horses if it has wargs or horses available.

10) Not really an AI issue but it does seem to be quite a serious exploit; giving a hero a horse and then waiting for monsters to leave their lairs so you can steal the loot is just too powerful. Either a: the main monster group should never leave its lair or b: the main monster group should never leave its lair on higher difficulty levels or c: it should be an option to say whether the main monster group ever leaves its lair.

Thanks again for creating such a great game, I haven't enjoyed a game this much in a long time.

I think FE will be more fun with an AI that is unpredictable rather than optimal

it isn't a great symmetrical strategy game, so I'd prefer the rogue-like approach where luck matters more and excitement is created by not being able to predict the outcome of battles

As GC2 was brought up, I thought I'd chime in too.  The behavioral AI in GC2 was top notch... each race felt different and felt like it had a personality.  I didn't feel like I was playing against a generic AI but against that actual race.  And each race had a different feel to it.

Any move to make the AI like it was in GC2 is a move in the right direction.

I'd like it if Kraxis would stop paying small, almost dead factions to declare war on me, when it's quite obvious who's doing the paying, I've never even seen said faction, and am more than capable of crushing him.

nice

I hope to post some private builds this weekend. These would just be EXEs but the forums are a treasure trove of ideas.

Alstein pointed out the main thign which is that the AI, at this state, plays the game like I would. In GalCiv, I had the advantage of seeing lots of people play the game (and being able to spend crazy time on the AI).  LH/FE have always changed so much during dev that I was playing a game of catch up.

This is going to be fun.

Frogboy, I'm having some issues getting the AI to use custom sovereigns and custom factions properly it seems.

Also it was mentioned earlier in post #14 about AI becoming Vassals. That would be an awesome feature from my perspective (sometimes if you want a buffer zone where you aren't committed to dealing with enemy units, plus for any role-play purposes for a story you're playing it would increase immersion).

So far I'm having a blast!

I like the idea of the AI playing differently depending on the sovereign's personality, although I assume it would be a lot of work because you'd have to come up with several different viable strategies.

The AI rushing the player as soon as it meets you could certainly be interesting, although I'm not sure if it would actually be a more successful strategy, as currently one of the better ways of beating an AI (assuming you don't already have an overwhelming advantage) is to let it kill its stacks on your well-defended city.

@Frogboy - this is great to hear, if you haven't already definitely checkout this thread "No Fog-of-War AI Evaluation" as there is some good feedback on AI starting actions:

https://forums.elementalgame.com/445389/page/1/#3368424

Cheers.

while you're at it: i'm pretty sure one reason why many people stomp even high level AI is how easy it is to get units the AI can't possibly deal with.

the  2 prime examples would be

- tamed bears with maul (mostly early game, but if used properly, there is no late game since you already won)

--> adjust maul to be less overpowered

- trained units with ridiculous initiative and fire damage buffs from essences (aura of grace & heart of fire)

 --> remove the 2 buffs, or tone them down to some sane level. as they are now, there's really no point investing time in tactical AI. with super buffed player units, there's no way the AI could possibly handle them, no matter how smart they move. and don't underestimate their effect. read some of the guides/AARs/watch some LP's and you'll realize that at least half (if not more) of the players who beat ridiculous/insane abuse those buffs to trivialize/bypass the whole tactical combat system.

One thing I notice with the tactical AI is that it tries, within reason, to kill one unit completely before killing others. This is a GOOD strategy against HEROES or other single-figure units (monsters, mostly), which do the same amount of damage even with smaller amounts of health. But if the enemy army is composed mostly of squads, then you are better off (in my experience) weakening a broader portion of them, since you can reduce the amount of damage they dole out with, in some cases, a single hit. This makes it so that your own units can survive longer, and can hold off completely killing an enemy unit until the end of the battle.

The AI doesn't seem to grasp this, and that might be its most significant tactical weakness.

The extreme faction differentiation makes conceptualizing how a great, straight-up AI solution would work difficult. I play so much to the strength of the specific faction I am playing, I would imagine it wouldn't be possible to really get competitive without designing for each faction (it will be hilarious if we ever see complaints like, "Kulan has allied with half the wilds, no fair"). I was amazed how well this worked in Gal Civ 2's expansions, but the challenge to do the same in this game seems far, far greater.

Now that we have access to some of the process in the game files, it will be really interesting to see what happens. This problem has always struck me as nearly impossible as a problem can be. I have always wondered how you approached it. If you ever describe your process in detail, or if you have and I have missed it, I would be fascinated (my desire is to do AI research). But I don't know if you consider it trade secrets or not, or if you consider it too technical for a wide audience. 

I agree. I do think Brad has mentioned that the major difficulty has been the game has been too fluid to get into that detail with the AI at present. Program the AI to take advantage of something, then the mechanic changed or is no-longer there. That I'm sure has been a difficulty throughout the process. But I see good things now for the AI as things are beginning to solidify and he can start to focus on strategies for the AI to use.

To your other topic. I believe we have access of doing specific strategies for ourselves, through this file

CoreAIDefs.xml

and you can define particular personalities for the AI to follow by setting appropriate weights for the game defining new AI personalities for the other sovereigns to use in CoreRaceConfigs.xml. Have fun if you would like to define unique AI from all these stats, it might be some fun to try and get the AI to do things the way you think they should.