Dominions 2. AI. Suggestions, that how to fix it.

[Particle]

It is nice to see you here Kris.
It is good to know that the devs are watching us from the skies. Errr...

[RadiantFleet]

Some other factors to consider when enhancing the AI. These are observations based on my preferred dom 1 playing style, one nation human, all of the rest impossible level AI, large map, and indep set to 9. On these settings, I usually win about 90% of the time.

-AI should take different decision trees depending on type of dominion/pretender that a nation has. Consider the following two configurations for Abysia:

Maxed dominion, growth, and order with a fortified city, a zero point pretender, max heat scale, max drain scale, no extra magic for pretender.

With this configuration, slow steady growth works really well. Build a temple in every territory, use defense heavily (minimum of 21, 41 for key territories, with later upgrades to 60). Armies are only for taking territory, not for defense. Your dominion is your main weapon, you take and hold territory, the spread of your dominion does the rest.

Maxed dominion, max heat scale, max drain scale, Castle or some other decent defense city, lich pretender (ie immortal) with a four in every sphere minimum.

With this pretender, magic summoning, immortality, and your dominion is the key. Build as many temples as possible. Buying units is unimportant, summon creatures that create new creatures (ie summer lions are nice, but fire elemental leaders are better, they make more fire elementals). Minimal defense in provinces (ie 21) to protect against sniping attacks, but armies are the serious defenders (because the only cost is leaders). Leaders that can create units should _never_ be sent into combat, only sit in castles and create more units.

These two configurations illustrate how different good strategies can be, depending on the characteristics of a nation and the pretender selected.

What I would really like is to have some way of specifying what pretender a computer player can use, or at least designing the ones in the "pool" of pretenders that are chosen from. Also, exposing the various decision formulas that the AI uses in files that can be modified would be great. It would help out illwinter by getting a bunch of people working on improved AI rules while letting illwinter focus on their great game engine, etc.

Originally posted by Chris Byler:

quote:

---

Originally posted by Kristoffer O:

quote:

---

Originally posted by Alneyan:


I would like a confirmation: is Dominions AI cheating? I would tend to believe it is not the case. If so, Dominions AI is really not that bad, and is one of the few AIs that does not rely on extra resources to provide a challenge. And I will be then in the side of "The AI isn't that bad, really".

The AI is not cheating on normal level. At higher levels it gets more design points and earns more money. Thats about it.

I got an idea regarding AI dependancy on LI when I was away. Vacations are 'foyson' for the mind.


Regarding fort construction. I'm not sure how the AI works, but where would you build a fortress? Consider fort type, nation, geography, income, resources etc.

What is the most important matter? How should they be weighted? How much shall current wars affect the spending of time and money. What army should build the fortress?

I wouldn't mind a numerical evaluation of this such as:
Castle cost / 2 < Income + res + gem income x 25 + neighbors x 5
Add a couple of other conditions.

Just to make you think. Eventually it might result in something good.

---

Some responses to these questions:

1. In your list of factors you don't list population type and sites in the candidate province. I often fort to enable building more of an indep troop (especially amazon sacred cavalries, which are all useful; but also e.g. heavy cavalry for Caelum or C'tis, longbows/crossbows for any nation with weak missile troops) and/or protecting a location with high gem income or useful mages (either pop based (amazons again) or site based).

A good first rule is that if you can't spend all your gold buying useful units in forts you already have, you need a new one; if your units are frequently starving because of insufficient supplies far from your forts, you need a new one in that area. (Which reminds me - supply production can be a factor in fort location too.)

Rarely build a fort next to another fort (it is occasionally worth it for something that can be only found in that province, for example I will always fort a province that contains Mount Chaining in Dom I - in Dom II it isn't as powerful but is still pretty good if you are a blood using nation).

2. You don't need an army to build a fort. I usually don't use one. There is some risk in building a fort where you don't have an army, but I find it is often better to have the army doing something else - either taking more indeps (if they are available) or distracting the enemy so he doesn't have time to Call of the Winds every place I might be building a fort. Get some province defense though - it's annoying to have your fort sieged by Call of the Winds (and of course even more annoying to have the construction interrupted).

3. Fort selection value: (Adjustment * (resources + supplies)) + (gems x 10) + value of rare units recruitable here (this value should depend on the rarity and power of the unit; longbows and crossbows count a little, amazon cavalries and mages count more, something like grey knights would count a lot; mages with paths different from your nation are more valuable than mages that duplicate your nation's paths) + misc site value (enter-to-summons, cost reducers).

I don't generally take income into account directly (although the first two terms value high population provinces more). "Adjustment" depends on the admin and gold cost of the castle; higher admin and cost should raise the value of the adjustment. Adjustment = 0 if there is an adjacent fort (this will reduce fort-next-to-fort except when gem income, rare unit value, or site value are high, which are the exceptions to the no-adjacent-fort rule).

4. Once you have built a fort, build a temple. Temples are vulnerable to enemy attack, but in a fort they are far less vulnerable. Then they spread your dominion, let you recruit priests and preach better. Mictlan might be an exception to this rule because their temples don't radiate - but blood sacrifices are more effective when performed in temples (IIRC), so they might want to build temples anyway so that they can sacrifice there. It's better to preach or sacrifice inside a fort because it protects the priest (high level priests especially are pretty expensive).

Build a lab if you are likely to want to recruit any mages (if you are fortifying because of local troops or sites you probably want a lab for that anyway; some nations like Marignon always want to recruit mages) or summon/construct troops there to supplement your recruited troops (C'tis usually wants to do this; but Desert Tombs C'tis can do without the lab, if you don't mind shipping in your unpriests).

---

[Mortifer]

I think it is not as important now. We have a nice list about the AI weaknesses, so that feature isn't needed.

[Aristoteles]

Okay, here is the list once more, that what should be upgraded in the strategic AI:


1. The AI is massing weak troops
2. The AI wont protect his pretender/commanders/priests/mages
3. The AI wont build forts
4. The AI's battlefield tactics could be better
5. The AI is making very odd things. IE. wandering around with his pretender alone etc.
6. The AI won't stay in a province, it is always moving his armies.
7. The AI won't make complex strategic maneuvers. IE. Making a focused attack against a province.
8. The AI won't equip his supercombatants..at least I've never seen that in the demo.


Feel free to add more, however I cannot think anything else right now.
Nerfix has posted that the AI is building too many temples, I don't know that this is good or bad.

[Serpico]

Hi, I am far from the experienced status, but I've seen these problematic points,-what you've mentioned on your list-, in my games with the demo.
I hope that this game developer -Illwinter- will improve these parts of the AI.
I love the singleplayer at least as much as the multiplayer.

[Gandalf Parker]

OK I think we still have time for this. If I modify the Aran .map file (eye is overused already) to provide the AI some of these items do you think it would clairify things abit?

1. The AI is massing weak troops

How about if each AI started the game with another army of hvy and elite units? Or even 2 more armys.

2. The AI wont protect his pretender/commanders/priests/mages

If I actually BUILT the pretenders for each AI then I could assign bodyguards to it. I can also bodyguard priests and mages but only if the .map file gives them.

3. The AI wont build forts

I could start the AI's with a couple of additional provinces and forts/temples/labs

4. The AI's battlefield tactics could be better

Hmmm I dont see where I could help this much. I could toss in a bunch of ally commands so that the AI's dont waste time and resources fighting each other.

5. The AI is making very odd things. IE. wandering around with his pretender alone etc.

Hmmm bodyguards is all I can think of.

6. The AI won't stay in a province, it is always moving his armies.

I think this is a good tactic for an AI. If you cant make it capable of good decisions based on whats coming then moving around alot helps.

7. The AI won't make complex strategic maneuvers. IE. Making a focused attack against a province.

Not sure what can be done here unless a new map image (tga file) is created with more bottlenecks.

8. The AI won't equip his supercombatants..at least I've never seen that in the demo.

That would tend to be very late game but if I create the pretenders I CAN also give them magic items.

The problem with so much pre-placement on such a map is that it wont have alot of replay value. The AIs would always start in the same provinces.

Also, should I only do this for the AI's that cannot be selected for human play? Or should I do it for all the races. If its all, then the human player will also have his pretender pre-selected and extra troops at start.

[ November 04, 2003, 15:51: Message edited by: Gandalf Parker ]

[licker]

While I appreciate your work Gandalf, I'm not sure that simply making map senarios is really what the folks are after here (though I expect you know that).

The fixes you propose (and you say this at the end of your post) are merely artificial boosts to try and make a specific map more difficult. I'm all for that, I love user created senarios that make life more difficult, but in the end, its not an improvement to the AI (unless there is the possibility for some *massive* scripting).

So I just ask that while your work is not done in vain, the focus of this discussion remains on how to improve the in game AI, not on how to make a senerio that is challenging.

Respecfully submitted

[Gandalf Parker]

Originally posted by licker:
While I appreciate your work Gandalf, I'm not sure that simply making map senarios is really what the folks are after here (though I expect you know that).

Respecfully submitted

Respectfully accepted. I have started a new thread for discussing a .map fix for these things.

Personally Im still not convinced that these are "problems" with the whole game. If they are based on settings then fixing some of these would cause problems when played with other settings (just a possibility). And if anything can be fixed with map mods then Id prefer that over a game rewrite since it would allow more variance in what can be accomplished with maps. Fixing these items might remove this level of AI play when some map makers might wish it.
Respectfully submitted Really

[licker]

Originally posted by Gandalf Parker:

quote:

---

Originally posted by licker:
While I appreciate your work Gandalf, I'm not sure that simply making map senarios is really what the folks are after here (though I expect you know that).

Respecfully submitted

Respectfully accepted. I have started a new thread for discussing a .map fix for these things.

Personally Im still not convinced that these are "problems" with the whole game. If they are based on settings then fixing some of these would cause problems when played with other settings (just a possibility). And if anything can be fixed with map mods then Id prefer that over a game rewrite since it would allow more variance in what can be accomplished with maps. Fixing these items might remove this level of AI play when some map makers might wish it.
Respectfully submitted Really

---

I agree that there are some (balance mostly) issues that may be negatively impacting the AI right now. However, I don't think that only modding is going to be the answer to 'fixing' the AI, unless the modders are able to somehow change the priorities of the AI in game. It is fine to make maps that should be more challenging, but it still doesn't strike at some of the underlieing (supposed) weaknesses. Starting the AI with more HI is fine, but if they don' continue to build more HI the advantage is soon lost (well depending on how many HI you start them with). Taking LI away from them completely (assuming you can do that by modding the units files) is another work around, but its also not really satisfactory in making the AI 'smarter'.

Alot of the discussion in this thread has tried to uncover the mechanisms by which the AI is making its army build selections, already it has been uncovered that perhaps the AI isn't building enough forts to allow for better troop recruitment, maybe the AI doesn't understand how to save money to build what it needs to, I don't know, but issues like that can only be 'patched' by scripting maps, they need to be fixed by some tweeking of the inputs to the AI, that's what I'm hoping is possible, as I don't want a complete rewrite, just the ability to tweek existing inputs.

[MStavros]

Gandalf, you are avery helpful person indeed.
However this won't help too much. It will offer more challenge, but the AI itself won't be changed, thus the problems won't be solved this way.
Still I love your map, especially with indy str 9.