AI Improvement Suggestions: Consolidated Thread for Brad's Attention

By on September 23, 2013 4:15:01 AM from Elemental Forums Elemental Forums

onomastikon

Join Date 02/2006
+18


While the AI in 1.3 has seen improvement, I believe the following helpful advice gleaned from various threads over the last weeks might help to greatly improve its performance. In the hopes that His Right Honourable Msgr. Bradley Wardell might grace this meagre thread with a perfunctory perusal, I herewith scribe:

1. AI Unitbuilding and Unit Design:

- Suboptimality: AI designs units with traits it cannot use, for example Wraiths (Resoln) with Plate Mail Proficiency. Less severe: early-game defenders with Plate Mail Proficiency.

- Suggestion 1: AI should design separate unit Templates for each faction, noting some of the most common strengths and weaknesses within each faction.

- Suggestion 2: The AI should rely MUCH less on "upgrading" existing units, and simply upon designing new ones as technology becomes available. Increase the number of templates and individual designs to reflect various stages in technology, particularly the early game, in which armor is not available

- Suggestion 3: AI unit designs should become available to the AI to build from after certain triggers have been activated, most notably, the discovery of a prerequisite technology.

2. AI Citybuilding:

- Suboptimality: The AI does not plan or use its cities effectively. Example of usage: Conclaves producing units while Fortresses produce research; available enchantment slots in Conclaves include Aura of Might when producing Wealth. Examples of planning: Tiles without essence are chosen for Conclaves.

- Suggestion 1: The AI should favor settling tiles with Essence more than those without. The AI should plan to make Conclaves out of villages with at least 2 essence, otherwise it should make towns or forts.

- Suggestion 2: AI should use fortresses primarily for generating units, not producing research; AI should make fewer conclaves, build more research- or mana-generating structures in conclaves, use conclaves primarily for generating research or mana, train fewer units in conclaves; AI should build more food- or wealth-generating structures in towns, use towns primarily for generating food or wealth, train fewer units in towns

- Suggestion 3: The AI should not settle tiles with 6 or fewer combined resources (unless there are 3 or more harvestable resources within range and no other tiles are appropriate). AI should greatly favor settling on tiles with 9 combined resources, even if there are no harvestable tiles nearby; AI should favor settling on tiles with 8 combined resources, even if there are no harvestable tiles nearby; AI should favor settling on tiles with 7 combined resources if they contain 1 essence, even if there are no harvestable resources nearby.

3. AI Championbuilding:

- Suboptimality: The AI will frequently equip a "natural" mage, such as Procipine, as an axe-wielding warrior, resulting in greatly ineffective champion.

- Suggestion 1: AI Sovereigns should have more detailed templates according to which the AI chooses equipment, traits, and behavior. Some sovereigns, such as Procipine, should focus much more on using magic spells.

. AI Tactical Behavior:

- Suboptimality: The AI will often choose suboptimal behavior for its champions in tactical combat. The AI *will* spend mana on strategic unit blesses, it appears, but seems much less likely to spend mana tactically. The AI will choose (also for human player champions with a "beserk"-causing spell or item) to keep a melee unit back with archers instead of forward.

- Suggestion 1: The AI should be much more willing to spend mana to cast buffs and debuffs. The AI should be more willing to cast spells.

- Suggestion 2: The AI should be more willing to place its melee champions in a position to attack in future rounds, also helping to "defend" its back ranks.

5. AI Strategic Behavior:

- Suboptimality: The AI will often not move its units (at all), but keep single units just outside of cities, or near enemies / ruffians, getting them picked off easily. The AI will much too frequently undervalue peace. The AI will leave unattended lairs, treasure chests, or enemy outposts uncaptured. The AI will much too frequently rely on mediocre combinations of chaff units produced in highly unproductive towns or conclaves or villages to defend its vital interests. The AI relies on units to attack more than it does Champions backed up by units. The AI will build "stacks" of scouts to "support" its champions, instead of ranged units (to support melee champions) or melee units (to support ranged champions). The AI does not compliment its champions (1 melee, 1 support) or its troops well (support champion needs fitting troops, and needs to cast spells to support them).

6. AI Pathfinding:

- Suboptimality: The AI is greatly hindered by the current rule mechanic disallowing most non-hostile entities to occupy the same tile simultaneously, causing them to go for dozens of turns out of their way in the faulty hope of thusly being able to bypass a bottleneck. (This also hinders pathfinding of human units.)

- Suggestion 1: change ruleset to allow non-hostile entities to occupy the same tile simultaneously; non-hostile units can occupy same tile as other non-hostile units and structures.

- Suggestion 2: Should a destination be "blocked" by (hostile) units, the unit pathfinder should not find a route-around, but a different desirable destination.

 

Locked Post 79 Replies
Search this post
Subscription Options


Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 5:21:55 AM from Elemental Forums Elemental Forums

Regarding strategic behavior, there's one thing that's been bugging me. In general, a player works towards providing value to his faction. I see value in resources, number and level of cities, unit count and quality, technological advancement, etc.

But war. Being in a state of war is of no value. War is just a tool that enables you to attack someone else. The AI seems to think otherwise. It just declares war and then sits back "Ah, HA! Showed you there. Now we're at war. Can't top that smart move, can you?"

My last expert game started off really promisingly. I think the AI is doing an excellent job at expanding and was really boxing me in. I had to settle some suboptimal tiles, because I wasn't able to clear the monsters next to the best sites in time. Great work there by the AI. But then a neighbout declared war on me. We had good relations, but their power rating was higher, so in theory this was an ok move. The only problem was, that all the AI forces were on the other side of their longish empire. I just walked through their territory, taking cities as I went, with the single stack I had. In the end, I was able to capture all cities without ever fighting their powerful stack with the sovereign.

Here's hoping that Brad can find the time to implement a pre-war state or something to the game.

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 6:39:37 AM from Elemental Forums Elemental Forums


A.I diplomacy is wacked.Peace and good allies one minute with a so called peacefull empire and then they declare war out of the blue and it seems to loop with all factions.The diplomacy makes me shelf it.Not a fun or immersive experience.

 

I also think the A.I cheats when it comes to getting third parties to declare war on you.The cost of far too expensive for the human player to do at a wim.

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 11:22:31 AM from Elemental Forums Elemental Forums


thanks for making this thread.   if these things get addressed, we'd have a much better game.

 

I'm smeagolheart and I endorse this thread.

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 1:03:07 PM from Elemental Forums Elemental Forums

Yes, the game can always be more precious.

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 1:45:00 PM from Stardock Forums Stardock Forums

Indeed.

 

If it's possible then it should be done.

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 1:47:49 PM from Elemental Forums Elemental Forums

Quoting StevenAus,

Yes, the game can always be more precious.
 

Look we already have bunny slippers.  What's next a bathrobe?  Enchanted Earrings of Enchantment?

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 1:52:09 PM from Elemental Forums Elemental Forums

Quoting smeagolheart,



Quoting StevenAus,
reply 4

Yes, the game can always be more precious.  


Look we already have bunny slippers.  What's next a bathrobe?  Enchanted Earrings of Enchantment?

at the very least....if SD was to introduce eastereggs or spoofs, I would have thought they would have had "the one ring".....definately not bunny slippers....

 

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 5:45:36 PM from Elemental Forums Elemental Forums

AI seems to waste all its mana on useless spells mostly. On my caravans, more specifically. They love crippling my caravans, and keep casting strategic spells on it, and it has a good time braging about how evil it is to me and all. Makes me sad, really... Poor caravans Why not directly attack them instead, or just stop wasting mana on it?

Reason for Karma (Optional)
Successfully updated karma reason!
September 23, 2013 5:47:18 PM from Elemental Forums Elemental Forums

Quoting UnUtilisateur,

AI seems to waste all its mana on useless spells mostly. On my caravans, more specifically. They love crippling my caravans, and keep casting strategic spells on it, and it has a good time braging about how evil it is to me and all. Makes me sad, really... Poor caravans Why not directly attack them instead, or just stop wasting mana on it?

Haven't you seen the elemental winning flowchart?

 

Elemental #WINNING flowchart

1. Wither Caravans and force your opponent to their knees

2. Profit

Reason for Karma (Optional)
Successfully updated karma reason!
September 24, 2013 4:44:45 AM from Elemental Forums Elemental Forums

I suggest that all units enchantment city spells should be GLOBAL. Easy and effective!

May be it will benefit a human player also - player may build units in more cities. I don't see any high cheating bonuses for a human player. But AI will benefit from this 100%. There will be no much need in a highly essensed starting locations (in opinion of human players), so no need to press Ctrl+N dozens of times As well as AI will live healthly with continuing to build anything anywhere.

I don't know, is it possible to mod?

Quoting Vallu751,
I was able to capture all cities without ever fighting their powerful stack with the sovereign.

Captured cities with high unrest should need an option to revolt so that a player should need a big garrison to hold it. Razing a city should take 2x(level of a settlement) turns. So villages are razed in 2 turns, 5 level cities need 10 turns to raze.

Reason for Karma (Optional)
Successfully updated karma reason!
September 24, 2013 9:23:29 AM from Elemental Forums Elemental Forums

Quoting smeagolheart,


Haven't you seen the elemental winning flowchart?

 

Elemental #WINNING flowchart

1. Wither Caravans and force your opponent to their knees

2. Profit

Apparently Derek has had this fixed in 1.4

You are going to have to find another way to win.

 

Reason for Karma (Optional)
Successfully updated karma reason!
October 8, 2013 4:15:03 AM from Elemental Forums Elemental Forums

Ouch, just saw Pariden, against whom I am playing, had built 2 of 3 towers needed for the Forge of the Overlord, while I declared war on her. Approaching her cities, I can see that the third tower is in the building queue in a Conclave -- BUT in front of this, she had queued the training of an archer, taking 6 turns. The city directly adjacent to this conclave is a Fortress, which was set for producing Research. Excuse me, I hate to say it, but that's really awful. The AI might have actually had a chance at winning for once? If the fortress had produced the unit? Or instead of producing research, produced the tower prerequisite to casting the spell of making? Oh, this really needs a tiny bit of work sir!

Reason for Karma (Optional)
Successfully updated karma reason!
October 8, 2013 6:08:02 AM from Elemental Forums Elemental Forums

Quoting onomastikon,
Oh, this really needs a tiny bit of work sir!

By modding make units cheap, allow training yards in all types of cities - and that conclave would produce that archer quickly and that archer will be decent warrior. Delete features "Produce research" ... etc and here you are! The game is raw but you can adjust it to your needs. Mod it!

Reason for Karma (Optional)
Successfully updated karma reason!
October 8, 2013 9:06:43 AM from Elemental Forums Elemental Forums

Quoting onomastikon,
Ouch, just saw Pariden, against whom I am playing, had built 2 of 3 towers needed for the Forge of the Overlord, while I declared war on her. Approaching her cities, I can see that the third tower is in the building queue in a Conclave -- BUT in front of this, she had queued the training of an archer, taking 6 turns. The city directly adjacent to this conclave is a Fortress, which was set for producing Research. Excuse me, I hate to say it, but that's really awful. The AI might have actually had a chance at winning for once? If the fortress had produced the unit? Or instead of producing research, produced the tower prerequisite to casting the spell of making? Oh, this really needs a tiny bit of work sir!

Ouch. Yeah. That should never happen.

To SD: Can the AI be finetuned a little in preferance towards which city types they first attempt to build specific things in for the 1.4 update??

Reason for Karma (Optional)
Successfully updated karma reason!
October 8, 2013 10:31:45 AM from Elemental Forums Elemental Forums

Excellent suggestions!

Well, except for one:

Quoting ,



3. AI Championbuilding:

- Suboptimality: The AI will frequently equip a "natural" mage, such as Procipine, as an axe-wielding warrior, resulting in greatly ineffective champion.

- Suggestion 1: AI Sovereigns should have more detailed templates according to which the AI chooses equipment, traits, and behavior. Some sovereigns, such as Procipine, should focus much more on using magic spells.

Mana is expensive, so my mages cast spells sparingly. If you design AI sovereigns as spells-only characters, the AI will quickly run out of mana, and the Sovs out of actions in battle.

That being said, Procipinee should never equip an axe, only swords, daggers or ranged staves.

Reason for Karma (Optional)
Successfully updated karma reason!
October 9, 2013 8:34:20 AM from Elemental Forums Elemental Forums

Quoting onomastikon,
The city directly adjacent to this conclave is a Fortress, which was set for producing Research

I was under the impression that the AI had got better at using its Fortresses to produce units. One of the exceptions that I raised to this was that you might wish to train units for defence. Usually if I can see an AI city that I'm at war with then the city has < 10 turns to live, so it should focus on defence. However if it was that focused on defence I would have expected the Fortress to be producing a unit as well as the threatened Conclave. Certainly, producing Research in  Fortress when you're at war and producing units elsewhere seems pretty pointless.

The AI can go for the Spell of Making, I've lost to it once for exactly that reason in my "Insane challenge", and could easily have lost to it via the Spell of Making on my last (successful) attempt. That's with massive research, mana, production and resource bonuses though, I'm not sure how likely the AI is to get to the Spell of Making without massive advantages.

Reason for Karma (Optional)
Successfully updated karma reason!
October 16, 2013 2:59:19 PM from Elemental Forums Elemental Forums

In every game I have played the AI only ever does two things, it....

Declares war then sues for peace, then declares war then sues for peace, then declares war then sues for peace, then declares war then sues for peace, then declares war then sues for peace, then declares war then sues for peace.  You get the picture?

 

The thickest AI I have ever come across.

Reason for Karma (Optional)
Successfully updated karma reason!
October 16, 2013 6:32:13 PM from Elemental Forums Elemental Forums

Just wanted to post a thanks!

Also, if any of you have some games on HUGE maps that are pretty late that you'd like me to look at, post a link to a save here.

Reason for Karma (Optional)
Successfully updated karma reason!
October 17, 2013 7:16:39 AM from Elemental Forums Elemental Forums

Thank you for posting, and for listening. I do think that the AI still needs serious work.

Sadly, I cannot play on HUGE maps, because of the difficulties surround unit pathing or non-hostile units not being able to share tiles with other non-hostile entities. Most maps larger than medium require my units to move too much, and this is simply a nightmare when a friendly outpost, shard, resource, town, or unit happens to be in a bottleneck. Gosh if you could fix that (allow non-hostile units to share locations with other non-hostile units), I would never complain about AI again and I would immediately buy GalCiv3. ;.) (You don't remember, Brad, but you wrote 6 years ago that you really liked my friendly but constructively critical post explaining why I was giving up on GalCiv2... so not convinced yet!)

Reason for Karma (Optional)
Successfully updated karma reason!
October 18, 2013 4:34:15 AM from Elemental Forums Elemental Forums

Quoting onomastikon,

Thank you for posting, and for listening. I do think that the AI still needs serious work.

Sadly, I cannot play on HUGE maps, because of the difficulties surround unit pathing or non-hostile units not being able to share tiles with other non-hostile entities. Most maps larger than medium require my units to move too much, and this is simply a nightmare when a friendly outpost, shard, resource, town, or unit happens to be in a bottleneck. Gosh if you could fix that (allow non-hostile units to share locations with other non-hostile units), I would never complain about AI again and I would immediately buy GalCiv3. ;.) (You don't remember, Brad, but you wrote 6 years ago that you really liked my friendly but constructively critical post explaining why I was giving up on GalCiv2... so not convinced yet!)

I support this message.

If only friendly armies could share a tile...

and there was a way to choose if units on the same tile should merge or not, LH would be almost perfect.

Reason for Karma (Optional)
Successfully updated karma reason!
October 18, 2013 7:07:09 PM from Elemental Forums Elemental Forums

Just posting to say that I agree the pathing around neutral territory is incredibly irritating and causes me multiple mistakes every game. Each mistake wastes two turns too because the units will move the wrong way trying to find an alternative path, then you need to spend a second turn moving them back to where they were, before you can manually move them through the neutral territory.

Chokepoints are also unbelievably annoying, particularly since in 1.3 the AI seems to stop moving and never resume moving if you ever run into one of their units in a choke point, even after you back out to give them room. This happened 3 times to me in the same 1.3 game. I since stopped playing that game (and LH) and while I can't really pin the blame on the choke point bug (it was more my general lack of spare time) it was certainly a factor - the third and last time it happened, just before I quit, my main army was faced with a 15-20 turn walk back around the lake, or declaring war on my supposed ally.

Reason for Karma (Optional)
Successfully updated karma reason!
November 12, 2013 3:03:49 AM from Elemental Forums Elemental Forums

I know you are very busy with GalCivIII, but you aren't leaving FE:LH yet, right? Any news? (If only a solution to the non-hostile tile-sharing can be implemented, I'd be keen on purchasing DLC...)

Reason for Karma (Optional)
Successfully updated karma reason!
November 12, 2013 9:28:29 AM from Elemental Forums Elemental Forums

AI is always low on mana, like critically low.

I don't think I have ever seen an AI have more than about 100 mana in a game.

They enchant up their units really high giving them a huge reduction in mana generated per turn, maybe even putting them into negative generation.

They can't really save up for the devastating spells for that reason, I think.

I would like a hard coded rule that the AI will never suck up more than half of its mana generation per turn with unit enchantments.

Reason for Karma (Optional)
Successfully updated karma reason!
November 12, 2013 10:12:09 AM from Elemental Forums Elemental Forums

What Raid said. A hundred times over.

Reason for Karma (Optional)
Successfully updated karma reason!
November 12, 2013 11:09:58 AM from Stardock Forums Stardock Forums

Quoting XWerewolfX,
What Raid said. A hundred times over.

I agree. but the downside, is that the AI could win by magic victory far too easily.

Reason for Karma (Optional)
Successfully updated karma reason!
Stardock Forums v1.0.0.0    #108432  walnut2   Server Load Time: 00:00:00.0000359   Page Render Time: