What is really new?

[NTJedi]

Twan said:
Hum as AI diplomacy has always been exploitable in any game having diplomacy IMHO it's useless to waste the time of someone "to prevent players finding exploits with the AI diplomacy".

The reason exploits can be found is because the AI is not able to learn from its mistakes the same as a human opponent. Unfortunately this type of AI computer opponent won't be created within our lifetime.
At least if one developer is focused on creating diplomacy he can limit the exploits and improve the diplomacy with patches. My suggestion at least prevents the usual exploit of sending small gifts to keep the AI opponents happy.

Twan said:
Now if AIs still works like St Patrik said ("AI only become aggressive when they notice that you are weak, or when you attack them", a good resume of dom2 AI) I think there is a flaw in the "diplomatic part of the strategic AI" that may be easy to correct with good effects on the SP game : make AIs attack not only a weaker player but the strongest one. I think it's very simple to implement : if a pretender is by large first in charts all neighbours should attack him instead of staying passive waiting for their turn. No need to make an alliance system for that, just make all the AIs aggressive against the potential winner once a critical power level is reached (usually by the player).

I've covered all this in another topic... perhaps we'll see the improvement within a patch or within Dominions_4.

http://www.shrapnelcommunity.com/thr...b=5&o=&fpart=1

Sindai said:
I'm 100% sure several games already do that. GalCiv certainly does. Developers aren't stupid. If you can come up with a solution in the time it takes to write a forum post you can be sure they thought of it ages ago.

Developers are definitely not stupid... which is why we have poor AI opponents in our games. There's a few exceptions and these games are only average at best. Developers are thinking of their future careers and thus focus much of their time on graphics because it's easier to promote your career by illustrating what you can create with pretty graphics compared to some extensive AI formula. Firaxis has actually taken a good step forward in hiring someone who purely focuses on the artificial intelligence... so hopefully we'll see some better AI opponents in whatever next game is being developed.

Sindai said:
The real problem with diplomatic AI is that it is usually incapable of betrayal. If it is then you have the nontrivial task of making it smart enough to betray the player only when it's logical. If it's not done cleverly and communicated to the user it will only cause players to complain about the AI being "random" and "attacking for no reason."

Yes betrayal would have to be worked into the personality formula of an AI opponent and most games don't even have AI personalities much less a betrayal factor. The betrayal factor is much more complex... from what I've heard and seen CIV_4 and GalCiv_2 do some betrayal.

[BigJMoney]

What about taking a different approach to AI? Instead of trying to make an AI "think" like a human or "seem" like a human, why not just start small?

First of all, to have the most basic aspects of diplomacy, the AI doesn't actually have to be interactable, it just needs to exhibit relations. Each AI nation chooses allies, neutrals and enemies. This can be completely random, it doesn't need to be based on relative strength and intel. If it were, the player could exploit this by focusing on large armies to keep enemies from declaring war. Neutral relations would mean they ignore you. Allies would send a message that says, "X wants to ally with you, do you accept?" They will then periodically send surplus resources to their ally, and stealthy units will never be discovered and attacked in their lands. Now, the player can't alter relations manually, but he can choose to get involved. If a nation is at war with another, he can send resources to the one he wants to see win.

A few necessities:

* war should only be declared on those adjacent
* war can be declared at any time
* if a nation is already at war, they should not choose to declare war with someone else
* neutrals should still scout and instill uprising
* spies caught instilling unrest in allied territory change relation to neutral
* assassins caught in assassinations in allied territory change the relation to enemy
* alliances can otherwise be broken at any time
* alliances and wars can be seen by all

It's not a lot, but it allows the player some ability to backstab and be backstabbed. It's also non-exploitable. The only thing the player can do is declare war. Maybe sending ally requests to neutrals have an N% chance of them saying yes. To add some risk, maybe also give it a smaller chance of triggering war.

=$=

[Gandalf Parker]

The Dom3 AI does seemt to declare war on known neighbors. The trouble is that "known" can come from scouts, spies, and spells. Also the AI must be able to decalre war and retaliate with spells when spells are used against it so "adjacent" isnt a good idea.

Also "not decare war on a second nation" can be abused alot by stealth nations. One of my favorite tactics is to hang around a war and cleanup weak points from both sides of it.

Same with "wars seen by all". I would abuse that alot. Forcing me to use scouts to see where the wars are seems ok to me.

Of course I would like more diplomacy with the AI. I would like for it to recognize gifts. But I also know that I want these things because I spend more time strategizing such things in games than other people do about strategizing combat.

[Kristoffer O]

> Twan:
> make AIs attack not only a weaker player but the strongest one.

Why? Is anyone attacking the US?
It is rare in history to attack the stong one. Ally with him and reap the benefit of a combined victory where your nation gets some of the spoils.

On the other hand, real life politics are not games and real life nations are not winners and loosers (OK, they might be loosers). So game balance wise (or from a there-can-only-be-one-god point of view) this might be a good solution:

> Twan:
> if a pretender is by large first in charts all neighbours should attack him instead of staying passive waiting for their turn. No need to make an alliance system for that, just make all the AIs aggressive against the potential winner once a critical power level is reached (usually by the player).


Perhaps the best diplo AI would include the ability to give other nations money, without any effect

Hmm, perhaps I'm too cynical

[DominionsFan]

Well 2 things about the diplo AI once more.

1. The AIs can ally with eachother -> Harder SP games.
2. Human players will exploit the diplo AI -> Easier SP games.

So the question is, is there a point to add a diplo AI to any game? Take a look at Civ 4. It takes a little effort to make an alliance with 1 or more nations in the game, that is hardcore exploiting. Just send them stuff and voila, you are all set. Galciv2 is the same...

[NTJedi]

Kristoffer O said:
> Twan:
> make AIs attack not only a weaker player but the strongest one.

Why? Is anyone attacking the US?
It is rare in history to attack the stong one. Ally with him and reap the benefit of a combined victory where your nation gets some of the spoils.

Well Dominions is currently setup where there can be only one winner which dominates all territory unlike our current world. Perhaps Dominions_3 has a "combined victory" option available or in the works... otherwise joining forces with a much stronger player will at best only have you achieve 2nd place(the_first_loser). By joining forces with other nations against the most powerful player in the game it allows a possible victory.

Kristoffer O said:
On the other hand, real life politics are not games and real life nations are not winners and loosers (OK, they might be loosers).

Well the best AI would be an AI opponent which learns from its mistakes, learns from strategies of human opponents, and uses formulas to behave clever and sneaky like human opponents... unfortunately I doubt this type of AI will become available within the next 30 years.
The next best AI we will see in TBS/RTS gaming will randomly select one of several multiple personalities this will make it difficult to determine what the AI opponents will be doing. This type of AI will also be moddable allowing gamers and programmers to improve existing AIs and even create new AI personalities. Any game will have much fewer complaints on AI opponents if modding is available allowing them to continously evolve.

[Gandalf Parker]

Beware of the Neural-Net AI

[Ballbarian]

The next best AI we will see in TBS/RTS gaming will randomly select one of several multiple personalities this will make it difficult to determine what the AI opponents will be doing. This type of AI will also be modable allowing gamers and programmers to improve existing AIs and even create new AI personalities. Any game will have much fewer complaints on AI opponents if modding is available allowing them to continously evolve.

I agree completely. I have been writing AI scripts for Age of Kings and Star Wars Galactic Battlegrounds for several years. For anyone into SWGB here is a link to my AI's home page. It has been a popular AI among SWGB random map players for some time.

I avoided using any real diplomacy with it since it is a RTS game and the engines capabilities are quite limited in that respect anyway, but it does respond and chat to allies and enemies with a fair bit of variety.

It uses a series of randomly selected strategies based on map type, resource levels, and civilization selected. It will also adjust its strategies according to changing circumstances. The scripting language and AI engine is basically just a state machine, but it can create some pretty cool results.

Anyhow, my point is that if the scripting language is kept pretty simple and straight forward, and it gives the scripter the ability to influence unit selection, access to what the AI can "see", and the ability to do through commands anything that the human player can, a lot of potential is unlocked.

[Neophyte]

A kind of "diplomacy" that makes sense to me would be to allow pantheons to form. In the current only one of each nation, perhaps restricting pantheon forming to pretenders whose paths are mutually exclusive would be in order.

Pantheons would some into their own if multiple instances of nations were allowed - then it is simple, all pretenders from a given nation could be a pantheon.

[Twan]

Kristoffer O said:
> Twan:
> make AIs attack not only a weaker player but the strongest one.

Why? Is anyone attacking the US?
It is rare in history to attack the stong one. Ally with him and reap the benefit of a combined victory where your nation gets some of the spoils.

Hum there can be only one true god. So if a pretender is near ascension everyone try to stop him, no ?

It's not exactly the real world situation (except in the delirium of some fanatics, and in most cases they like the attacking the US concept ).