How Early Will the AI Declare War?

[bardolph]

I find it odd that the AI won't try to steal workers or attack my undefended capital in the early BC years.

After all, if they have archers vs my warriors, they are virtually guaranteed victory!

Does the AI have some kind of rule that it can't attack early in the game?

 

[Galileo44]

I think it is coded to stop ultra-early war. If it weren't, then I think that Deity would be impossible, all the AIs would just archer-rush you with their starting units.

 

[bardolph]

Hrm... I'd rather see them with fewer starting units, but being more aggressive with them. Besides, they already have the research/manufacturing bonuses. Giving them free units, then coding them so that they can't use those free units, is pretty artificial and cheesy.

 

[oyzar]

i got archer rushed by ragnar in one game. I had just hooked up copper but my one axeman couldnt realy beat his 5 archers...

 

[axident]

I find it odd that the AI won't try to steal workers or attack my undefended capital in the early BC years.

After all, if they have archers vs my warriors, they are virtually guaranteed victory!

Does the AI have some kind of rule that it can't attack early in the game?

I think Firaxis coded it to prevent immediately archer-rushing because, well, that wouldn't make the game very fun. Less fun = fewer sales of the game.

In a recent game, I got DoW'd the earliest I've ever seen. There was Ragnar on one flank with USA behind Ragnar farther away, and Kublai Khan kinda far away on my other flank. I worker-stole from USA and declared peace ASAP, as they were no better than Cautious with everyone and had to get through Ragnar's borders to get to me (no way that was going to happen). I was planning on using Ragnar as my game-long ally to seal off that flank, and in the meantime I built my first axemen and was going to attack a barb city nearby soon when Ragnar DoW'd on me with 2 archers and an axeman. WTH? I shifted directions immediately, whipped a few more axes, and Ragnar's capital was mine shortly after. Now I am a little cautious about early-game DoW's from my neighbors, practically right after I get my first axemen.

 

[jeremiahrounds]

AI should act like their character imo. And they try to do that to a certain extent as much as you can create a personality out of a broad set of "rules". THough a custom option should be "cut-throat" AI which plays as aggressively as a set of "Rules" could manifest.

I still dont think the object of the singleplayer AI is to "win". It should be to prosue goals consistant with the story of the AI.

Which leads to the question: why would they kill you off if they dont not like you and arnt aggressive.

I entertain an aggressive civ might invade you as soon as they sense your weak and you dip into "cautious" which is the starting value. Otherwise why would we call them aggressive? But everyone else I doubt they should invade you simply because your weak. Thats just not consistant with presenting a personality.

And its important I think to display a personality in the rule set that governs a particular AI.

BTW the ultimate level of AI programming achievement is not perfect play , but mistakes on purpose consistant with a character you are trying to portray with your limited mechanics at your disposal. This I learned from chess AI programs. Where the state of the art for programming a chess AI has advanced to the point where they are no longer just trying to make a perfect opponent because that is of limited interest for the average chess player, but rather various companies have begun programming personalities that make intended mistakes. Examples would be undervaluing certain pieces, consistantly making the same sort of mistaking, not being able to see how to properly position some types of pieces. The reason is to create a personality beyond "perfect play"

I did some quake2 bot programming. And I thumbed through the better AI code. I graduate from college this may. And I have a fantasy of going through it and seeing if I cant strengthen the personality of AIs in the ruleset rather than the play of the AI. Its just a fantasy at this stage.

I would like to see aggressive civs be more opportunistic then nonaggressive civs in looking for weak opponents while simultaneously not fighting lost causes. A truely aggressive civ would prolly invade someone for just being weak and not pleasing. That sort of stuff. Strengthenign the character of the AI and not neccesarily the play. Prolly could handle a bit of it by taking a good look at the weightings.

If nine different civs are on the board they should all play nine different ways. If there is only one best way (a local or global maximum) then atleast eight of them have to be playing less than perfectly. And nine different ways shouldnt be a sequence of 9 different random events. Thats no fun for the player. It should be nine different predictable paths of advancement based on the particular AIs.

You can of course see a little of this in the current AI scheme. Ghandi doesnt play like napolean.

 

[Mizar]

Hrm... I'd rather see them with fewer starting units, but being more aggressive with them. Besides, they already have the research/manufacturing bonuses. Giving them free units, then coding them so that they can't use those free units, is pretty artificial and cheesy.

The fact that the AI doesn't attack very early is indeed coded to make the game playable on high levels.
The AI gets free units on the higher difficult levels not to rush the human player but to prevent beeing rushed by him. And of course, the additional units speed up their expansion too.
An AI that doesn't need such starting advantages would of course be better
but the fact that early warmongering is an effective strategy on immortal/deity shows clearly that without it the current AI would not be much
of a challenge for good players.

 

[bardolph]

i got archer rushed by ragnar in one game. I had just hooked up copper but my one axeman couldnt realy beat his 5 archers...

What year did they declare war?

AI should act like their character imo. And they try to do that to a certain extent as much as you can create a personality out of a broad set of "rules". THough a custom option should be "cut-throat" AI which plays as aggressively as a set of "Rules" could manifest.

I still dont think the object of the singleplayer AI is to "win". It should be to prosue goals consistant with the story of the AI.

Which leads to the question: why would they kill you off if they dont not like you and arnt aggressive.

I entertain an aggressive civ might invade you as soon as they sense your weak and you dip into "cautious" which is the starting value. Otherwise why would we call them aggressive? But everyone else I doubt they should invade you simply because your weak. Thats just not consistant with presenting a personality.

Aggressive AI's should be more likely to attack because their combat bonus makes them more likely to win.

Every leaderhead in CIV, with the possible exception of Gandhi, would be willing to go to war if doing so would benefit their country. That's one of the reasons why they were considered "great leaders" in the first place.

One would think that the variance in traits, favorite Civics, UUs, UBs, and starting positions would give AI leaderheads enough variation in gameplay to make them behave differently from each other, even if every one of them pursued a single "optimal" strategy for survival and victory.

Given the challenges that the AI already faces when confronted with a human player, I think it's bad to program them to make "artificial mistakes" at all, because even when playing to the best of its ability, the AI still makes mistakes which the human will exploit!

Also, if a game is well-designed, there should be considerable debate about what exactly an "optimal" strategy would be anyway, so the various leaderheads could pursue different successful strategies from each other. The key word here is "successful" strategies.

I don't think there's any point in programming an AI to adopt strategies that are known to be unsuccessful strategies. For example, it's a known fact that all civs can benefit from aggressive tech trading (however, it might be debatable as to which techs to trade, when, and with whom). So, programming a leaderhead to avoid tech trading (*cough* Tokugawa *cough*) is only a death sentence for that leaderhead.

There are many ways to differentiate leaderheads, without crippling their ability to play successfully. Likewise, there are many ways for human players to play successfully against other humans!

And if there really is only one "perfect" strategy that wins every time, then it's time to redesign the game.

I think Firaxis coded it to prevent immediately archer-rushing because, well, that wouldn't make the game very fun. Less fun = fewer sales of the game.

That's what the lower difficulty levels are for. I'm all for coding in certain handicaps to the AI when playing at Warlord or lower, but once you start playing Noble and higher, take those handicaps away!!

The current handicap system has a really strange effect on Ancient Era game play for single player:

It makes Archery a useless tech in most games.

Why? Because the human player has no reason to build Archers:

1. The human player cannot build Archers for offense, because defenders are guaranteed to already have Archers defending their cities.
2. The human player should not build Archers for defense, because the AI is guaranteed not to attack with their archers until well after the Human player has the capability to build Chariots and/or Axemen.

That's just strange.

It's also strange that AI personalities are all given Archery as a free tech. It just railroads the human player into a very specific and limited decision tree in the Ancient era.

Now, it might just be an artifact of the way Civ 4 is designed, that the only way to give the AI a fighting chance is to overload them with defensive units at the start of the game. The problem with this "solution," however, is that if they actually use these free units, then it's the player who has no chance! So the first solution requires a second solution (limiting the AI's ability to attack in the Ancient era), which creates a third problem, and that's the "useless Tech" situation that I described above.

 

[Galileo44]

I would challenge anyone to win a game on Deity other than separate continent starts if the AI would be programmed to archer rush. They would take ~10 turns to get to you, and that would be death, as even if you have a warrior in that time, the archer still might win, or just choke the player.

 

[bardolph]

I would challenge anyone to win a game on Deity other than separate continent starts if the AI would be programmed to archer rush. They would take ~10 turns to get to you, and that would be death, as even if you have a warrior in that time, the archer still might win, or just choke the player.

Then give the AIs something other than free archers.

It's just ludicrous to me that, at DEITY, the AI is being handicapped. Shouldn't it be the other way around?

How about giving the AI free scouts? That will give them the opportunity for some free gold and technologies, but the benefit will be randomized somewhat, and they can't use the scouts to ambush the player's capital.

 

[Galileo44]

Then give the AIs something other than free archers.

It's just ludicrous to me that, at DEITY, the AI is being handicapped. Shouldn't it be the other way around?

How about giving the AI free scouts? That will give them the opportunity for some free gold and technologies, but the benefit will be randomized somewhat, and they can't use the scouts to ambush the player's capital.

The AI does have free scouts as well. I think that the basic problem is AI bonuses in general are not an effective way of increasing difficulty. It is good to see the Better AI project reducing the amount of bonuses that the AI needs. However, with the current system, every AI would archer-rush you at the same time if they were told to war ultra-early. I also wouldn't say that no ultra-early war is a handicap. The human player can't do it either. The only true war like that that I have seen is HOF games with rushing with a single Quecha on Deity and reloading to get the 10% win for a very early (IMO cheating) way. (3850BC win)

 

[sylvanllewelyn]

The AI has to discover you, and KNOW that you're weak, before declaring war on you. We are assuming that there are no such things as power graphs (which I consider to be cheating). The AI doesn't know that you're weaker than the other AI's, so it'll take time before scouting you out (without open borders, good luck) and realizing that you are, in fact, weak. This may be beyond the ancient era already, and you really shouldn't be building only warriors by that point.

If it's THAT realistic, I see no problems with AI's being aggressive. It'll correct itself. Plus, if the AI's were oppertunistic, once you attack an AI, the rest should dogpile it, making the game way too easy.

 

[Vynd]

Playing on Monach level once, Montezuma attacked me very early and wiped me out. I can't remember exactly when, but I believe it was in the mid-B.C. years. He had some Jaguars so it certainly wasn't right at the beginning of the game. But it was definitely well before the time I was used to seeing the AI launch wars, when I thought I was still "safe" even from homicidal lunatics like him. I was working on founding religions for a cultural victory and was exceptionally weak militarily, and I think the Ai recognized it and took advantage of it.

Archery is not a useless tech for the the human player because of barbarians. They're quite numerous and aggressive at high difficulty levels. Also, you seem to be assuming that the human player will always be able to get Copper and/or Horses hooked up early in the game, which isn't always the case. With that said, it's definitely true that Archery is not a very desirable tech. So what? Depending on one's stratgy there are always techs that will be better than others.

 

[bardolph]

The AI does have free scouts as well. I think that the basic problem is AI bonuses in general are not an effective way of increasing difficulty. It is good to see the Better AI project reducing the amount of bonuses that the AI needs. However, with the current system, every AI would archer-rush you at the same time if they were told to war ultra-early. I also wouldn't say that no ultra-early war is a handicap. The human player can't do it either. The only true war like that that I have seen is HOF games with rushing with a single Quecha on Deity and reloading to get the 10% win for a very early (IMO cheating) way. (3850BC win)

I think only civs with hunting get free scouts. Giving bonuses to the AI is necessary, because there is no other feasible way to make the AI challenging. The question is, which bonuses are the most appropriate, and provide the most challenging and fun experience for the player?

The only reason that a human player can't war ultra-early is....

You guessed it. The free archers given to the AI.

Do you see my point? They give free archers to the AI to prevent the human player from declaring war too early, then said, "Uh-oh, these archers are dangerous to the human player. Better nerf the AI and program it so it can't use the archers we give it." It's a sloppy solution. It means that they gave the wrong bonuses to the AI in the first place. There must be a better way.

(By the way, the human player can declare war ultra-early, to steal workers and pillage improvements. Not all wars need to capture cities.)

I think replacing every free archer with one or two free scouts is a possible solution. That would mean that some AI civs will gain a big early tech advantage, while some AI civs will be stuck with just the "normal" research/production bonuses. Consequently, the human player might actually benefit from an archer rush, if the target AI happened to get unlucky with its goodie huts.

This solution would give an extra strategy choice to the human player, instead of railroading you into "Bronze Working first; if you don't have Copper, then Animal Husbandry; if you don't have horses either, only then do you ever bother with Archery" formula.

The AI has to discover you, and KNOW that you're weak, before declaring war on you. We are assuming that there are no such things as power graphs (which I consider to be cheating). The AI doesn't know that you're weaker than the other AI's, so it'll take time before scouting you out (without open borders, good luck) and realizing that you are, in fact, weak. This may be beyond the ancient era already, and you really shouldn't be building only warriors by that point.

If it's THAT realistic, I see no problems with AI's being aggressive. It'll correct itself. Plus, if the AI's were oppertunistic, once you attack an AI, the rest should dogpile it, making the game way too easy.

I don't consider the Power graph to be cheating, since it's information that's available to all players, including the human one. Think of it as an intelligence network.

The human player can DoW an AI even before scouting the capital city; often, to steal an undefended worker. You can also DoW just to scout the area, and to pillage improvements. In multiplayer, human players are certainly capable of these strategies.

If these tools are available to the human player, why not make them available to the AI as well?

Of course, once you've loaded them up with free archers, you can't allow the AI to do this, because you've just given the AI an overwhelming military advantage in free units, which, if the AI actually uses, would spell certain doom for the human.

But even given this fact, what is the AI trigger for being able to declare war in the early game? In my experience, it seems dependent on acquiring an offensive unit other than archers: chariots, axemen, swordsmen, or any UU replacements for these units.

Which means you'll never see a Skirmisher or Quechua rush from the AI. (EDIT: on second thought, I've never seen an Axe rush from the AI, either. Have you?)

Oh well.