AI Diplomacy, not so bad

[Trias]

Only to an extent. Why is my being their main rival with respect to win method A more of a valid motivator than my being close to victory by method B? In general it is rational to avoid trades with a player who is close to victory. However, the modifier comes into play long before anyone is close to victory.

I think this is a compromise in the rationality of the AI algorithms to avoid the situation where all AIs pounce on the game leader (often the human player). So, instead they limit this behaviour to AIs (that think that they are) going after the same victory condition.

Also avoiding trading with a player only when he is already near a victory typically is too late. I therefore makes sense to start to frustrate your nearest rival the moment you identify him.

The answer is yes, I have. However, it usually is a result of the AI being so close to me that I am able to warrior rush them before they can stop me. I wouldn't be concerned or confused with the modifiers [being discussed] if this was the situation for the AI. If there is a civ between myself and the civ that is coveting my land, then your explanation is moot. The civ should covet the land between us before coveting my land (and providing early troubles with diplomacy for both of us).

If I have to guess the following thing is happening: The AI identifies its preferred expansion locations, if any of these fall in another players territory it will get a "covet lands" attitude modifier. So far so good. The problem here is that these locations can be a bit erratic, as anybody that has view a civ4 game in debug mode has witnessed. (This probably somewhat necessary in order for the AI to be able to deal with no standard map types.) As a result the modifier triggers in situations that seem some what irrational to human player.

If my guess is correct, the proper way to fix this is to improve the algorithm the determines where the AI wants to expand, not removing the modifier against players that are "in the way".

There is absolutely no excuse for receiving a penalty for "trying to win the game the same" before any victory condition is being worked towards. I cannot comprehend how this is acceptable. Sure you can excuse it or make excuses for it, but in the end, it is simply ridiculous. If you want to have guaranteed bad relations with at least 1 civ from the start, then have a modifier "We just don't like you" and be honest about it. Since there is no such modifier, I must assume that the "winning the game the same" is broken.

Many players (and all the AIs) have a victory condition they are working towards from the start of the game. It is therefore reasonable for the AI to assume that other players have such a goal. Even if they don't it can still make an educated guess based on the progress of the game, what victory condition you are most likely to go for.

I don't quite get what you don't understand about it being reasonable for the AI not to be nice to the players that it views as its primary rivals. (most humans do this as well) See my response to gemist above as to why I think that the AI applies this only to players competing for the same goal.

Understanding what you say, and agreeing for the most part, I must disagree with your evaluation of my suggestions. Changing the "winning the game the same" mechanic and the "covet your lands (prior to having more than 2 border pops)" isn't "straightjacketing" the AI game mechanic, it's fixing a broken one.

This is exactly the problem I have with your appoach to fixing problems with the AI, the fact that you treat it as a game mechanic rather than AI decision making. (I think we both agree that somethings need fixing/improving, but we disagree on what should be fixed).

Yes. There is very little (long term) reason for players to cooperate. So there should be a reason. There needs to be a mechanic built in that encourages building long term relations with the intention of taking part in a victory. I would gladly accept a permanent alliance with a militarily strong civ and share the victory spoils if it means I can win my Culture victory without having to build an army for those final desperation DoW.

On this I completely agree.

It isn't natural for the human player to backstab in this game unless they plan on domination or are so far ahead militarily/victory condition-wise that they aren't worried about diplomacy. A backstab garners a pretty big hit diplomatically with the whole world (as it should). If I'm playing domination, you bet there will be backstabs since it is impossible to win a domination game without DoW (or at least going to war) on everyone including your friends. But not every game in domination. In fact, far less games are won by domination in Civ5 than in Civ4 (at least in my games)... and my strategies were generally those of a warmonger.

You are correct, good players will manipulate their target in DoWing instead of DoWing themselves. Conclusion remains the same, the typical human player will not hesitate to screw the AI over in anyway it can (get away with). It is therefore natural for the AI to be somewhat cautious of other players.

 

[apocalypse105]

well topic of thread is: AI, Diplomacy not so bad.

If you have to set your own rules - the game and its rules is obviously broken!

Same is for so many other stuff, worst is that DoW from AIs is totaly independent from its trading, therefore weired stuff like Dows the turn before RA ends or AI declares war the turn after it have given u all its money for lux itself can happen.

The main problem is the coding making trading and denounces, DoWs independent of each other when in fact it should be 1 system (which should also include demands).

You just said eveything what is wrong with diplomacy of civ 5 espeically the trade system if you should have research agreement or trade the Ai should'nt declare on you just because his numbers are biggers because it hurts them both BUt hey I guess diplomacy isn't so bad??....

 

[blitzkrieg1980]

Many players (and all the AIs) have a victory condition they are working towards from the start of the game. It is therefore reasonable for the AI to assume that other players have such a goal. Even if they don't it can still make an educated guess based on the progress of the game, what victory condition you are most likely to go for.

I don't quite get what you don't understand about it being reasonable for the AI not to be nice to the players that it views as its primary rivals. (most humans do this as well) See my response to gemist above as to why I think that the AI applies this only to players competing for the same goal.

I think it's because you haven't noticed that I'm talking about the modifier when it is often received before there is much progress in the game at all. Again, I've received this modifier before: settling a second city or having a high population and being relatively behind in tech (science), choosing more than 1 or 2 social policies since I tend to save them until I know what strategy I'm taking (culture), making friends with ANY CS (diplomatic victory), or building very much of an army or attacking anyone (domination).

So the algorithm used to have this modifier activate is either poorly laid out or broken. By activating this modifier so early, the AI is only damaging its own possible short term gains through trade, Research Agreements, possible ally in a war, etc with me. Again, if the idea is that we shouldn't be able to make good relations with everybody, then there should be a "We just don't like you / your culture" modifier.

What I don't understand is how this mechanic (which is what it is, it's a game mechanic whether I think of it this way or not, it's a programmed entity with set boundaries and constrictions) can be seen as acceptable when there is clearly no victory condition being worked towards yet. My position is NOT that this algorithm isn't working properly as it was designed, but that it was not designed well. There should absolutely be a constriction that requires some kind of actual advancement towards a victory condition before the diplo modifier activates.

For example, once a certain amount of culture points have been accumulated by the player (perhaps 1500) if the player has a higher rate than civ x and civ x is working towards a culture victory, then civ x activates its "you are trying to win the game the same" modifier. Or, if 50% of CS in the game have allies and the player has made friends with more CS than civ y and civ y is going for a diplomatic victory, then civ y activates its modifier. Or after reaching the industrial era, if the player has more techs than civ z and civ z is going for a science victory, then civ z activates its modifier. Domination is pretty obvious. If more than half the original capitals have been captured and the player has a larger military than civ w and civ w is going for a domination win, then they can activate their modifier. I think this seems pretty straightforward. It prevents anyone from receiving erroneous "win the game" diplo modifiers when the AI has no real evidence of the modifier's accusations. Once the evidence is there, the AI activates the modifier and begins to erode relations.

Without these constrictions (or something similar), the AI acts way too "gamey" and unrealistic. Really, though, it's going to hurt the AI more b/c it will gain more in trades and research agreements with me over the short term than it will by pissing me off early and making me consider chancing a "warmonger" label to remove them from my presence.

 

[apocalypse105]

Well, in fact it seems to me that u dont want ai DoW you even when your building Space Ship and having 0 military just cause it likes you - d winning not be too easy?

There are lot possibilities to not have war every 20-30 turns - but u have to fullfill several condition:
not be a threat but neither be a easy or the only target (and there is much more stuff)

His point is that the AI declares or is mad ad you at the beginning of the game when you dont even get close to a certain victory condition and start atacking on the player...

That doesn't make sence getting modifiers like they covet you're land if they are on a other island or not even close to you.. And you are winning the game modifier if you only have 1 citie...

I would'nt mind if the AI would get mad if I was building space ship parts or having 5 social policies fult in and he wants to declare me... That makes sence but in the beginning of the game no... Now it just feels like a real time strategy game where the AI just wants to kill you at the verry beginning

Question: Have you ever disliked an AI simply because of the location where it spawned or because it took an early lead? Even if the answer for you is no, for most human civ players is yes. This is exactly what these two messages express for the AI. They might trigger to often and/or in the wrong circumstances, but they certainly are perfectly legitimate reasons for the AI not to like you, and if they they do dislike you there is (and should be) very little you can do about it.

Note that the fact they you do not know what your grand strategy for the current game is, does not mean that the AI cannot think that you will be the greatest competition for the strategy they have chosen.

And you dont see the point every person is making in this topic that this kind of gameplay is broken and gives a really unfun game experience...

If i want the AI mock me at the beginning of the game I play a real time strategy game where I have to fight for the dead.. Not a game where there should be to be diplomacy but it isn't instead its a world war game...

In a game where trading is so important (research agreements) especially on high levels. Where the AI hates you at the verry beginning is a bad game design....

 

[blitzkrieg1980]

In a game where trading is so important (research agreements) especially on high levels. Where the AI hates you at the verry beginning is a bad game design....

This. And I understand that (for realism), not every AI should play nice every game. BUT, this should fit in with their personality (ie Montezuma, Alexander) and the modifier should be separated from the "trying to win the game the same" modifier. Fix that one, and then add in one "We just don't like your culture... OR you!".

 

[headcase]

You know, you could just not do that?

I have a rule of never doing a one off vs long time trade deal if I intend to declare war on that AI.

But the AI is playing to win. Trading and then deccing would be a valid strategy, one that's been done in real life and one that's technically open to the AI... but there's no counter to it.

They should bring up a box with any verifiable information upon any dec "Germany has Declared War on Spain, breaking a deal in which they gave 5 GPT for 200 gold up-front".

Then the AI could hate you for it.

This could be elaborated on with espionage, but simple info like that shouldn't require any.

Edit: The other thing is the AI does hate you for things that a human couldn't know, like "you promised you woulnd't go to war and then did". How did they know?! Again, that should be part of the box "Germany had an unsettling presence on Spain's border, and had promised not to go to war 17 turns ago."

 

[Trias]

What I don't understand is how this mechanic (which is what it is, it's a game mechanic whether I think of it this way or not, it's a programmed entity with set boundaries and constrictions) can be seen as acceptable when there is clearly no victory condition being worked towards yet. My position is NOT that this algorithm isn't working properly as it was designed, but that it was not designed well. There should absolutely be a constriction that requires some kind of actual advancement towards a victory condition before the diplo modifier activates.

(No its not game mechanic, as it is not part of the game design. It is part of the AI (and AI-human interface) design. The whole point behind the civ5 AI design, is to make it a player rather than a game mechanic.)

But practically there is always a victory condition being work towards. (Even if the player has not really decided yet.)

For example, once a certain amount of culture points have been accumulated by the player (perhaps 1500) if the player has a higher rate than civ x and civ x is working towards a culture victory, then civ x activates its "you are trying to win the game the same" modifier. Or, if 50% of CS in the game have allies and the player has made friends with more CS than civ y and civ y is going for a diplomatic victory, then civ y activates its modifier. Or after reaching the industrial era, if the player has more techs than civ z and civ z is going for a science victory, then civ z activates its modifier. Domination is pretty obvious. If more than half the original capitals have been captured and the player has a larger military than civ w and civ w is going for a domination win, then they can activate their modifier. I think this seems pretty straightforward. It prevents anyone from receiving erroneous "win the game" diplo modifiers when the AI has no real evidence of the modifier's accusations. Once the evidence is there, the AI activates the modifier and begins to erode relations.

The problem with such thresholds is two-fold:
1) They are not very robust against XML changes, and heighten the possibility of breaking the AI with an XML change.
2) There is a good chance that it will trigger to late.

That being said, it does seem that it currently may trigger to early. But again this should be improved by improving the alorithm through which the AI guesses its opponents strategy.

Without these constrictions (or something similar), the AI acts way too "gamey" and unrealistic.

Because in early human history no conflicts have been fought over the perceived aspirations of other nations?

 

[tommynt]

Quote:
Originally Posted by tommynt
well topic of thread is: AI, Diplomacy not so bad.

If you have to set your own rules - the game and its rules is obviously broken!

Same is for so many other stuff, worst is that DoW from AIs is totaly independent from its trading, therefore weired stuff like Dows the turn before RA ends or AI declares war the turn after it have given u all its money for lux itself can happen.

The main problem is the coding making trading and denounces, DoWs independent of each other when in fact it should be 1 system (which should also include demands).

You just said eveything what is wrong with diplomacy of civ 5 espeically the trade system if you should have research agreement or trade the Ai should'nt declare on you just because his numbers are biggers because it hurts them both BUt hey I guess diplomacy isn't so bad??....

well my point is that AI IS totaly brokken and TE is just halozinating, or just havent understood how easily u can both abuse its dumbness and get ed by its dumbness. It follows just way too simple patterns.

Same way too simple patterns are seen in other eras, like ai sending 8 undefended catapults right into your arms

maybe read this again ..

OK here is how I play sometimes (espacially deity games):
starting unit and 2nd units goes to find as many AIs as possible (prio over huts/barbs/CS).
As soon as AI 1 got around 200 gold (usually like turn 15) I trade all my gpt for its money and buy worker with that money and go work some lux and DOW (make sure scout isnt close to a ai warri)

I work 1 or 2 Lux and go find more AIs - at around turn 30 usually most ais are found (if 1 Landmass) and 1 or 2 Lux are worked.
Now go seell all your money and lux to AI1 for all its gold -->DOW
Same for AI2 --> DOW
...

Have 1000+ gold - buy whatever u want (but some military aswell)

As u seem now pretty strong and are a "threat" (even when miles away - AI is dumb ..) and most AIs will be happy to have Peace 10-20 turns later.
Just some warmmonger civs might stay in war (they build more units and arent afraid).
But being in war doesnt mean at all that ai will attack - in fact opposite might be true.

Even on top of that some civs will start like u now as u can resell your luxes to them and you are in war with the warmmonger civ (which may very likly be in war with some other ais).

well if this system isnt broken ...

 

[qartar]

(No its not game mechanic, as it is not part of the game design. It is part of the AI (and AI-human interface) design. The whole point behind the civ5 AI design, is to make it a player rather than a game mechanic.)

Which many people believe was a mistake. The AI functions as an obstacle; it is too unsophisticated to act as an effective adversary and the sacrifice of depth and immersion to the god of victory screens has eliminated the possibility of acting as an actual ally (but thank goodness we have city-states since shoveling cash at them is so rewarding). The better solution would have been to make the AI better at diplomacy, ie. less prone to gamesmanship, and more effective at achieving victory in its own right. Instead what we get is a sociopathic AI that is incapable of leveraging game mechanics to its advantage with stripped down diplomacy that can still be gamed.

But practically there is always a victory condition being work towards. (Even if the player has not really decided yet.)

Right, so why don't we dispense with the pretenses and have everybody declare war on the player from the get-go? After all they are trying to win one way or another. Oh right, because that would be unfulfilling and idiotic and this isn't Warcraft.

 

[blitzkrieg1980]

The problem with such thresholds is two-fold:
1) They are not very robust against XML changes, and heighten the possibility of breaking the AI with an XML change.
2) There is a good chance that it will trigger to late.

What makes them lack robustness with respect to XML? I don't see a problem with putting constrictions on a diplo modifier since they already have requirements and constrictions. There is most definitely a way to work this in. A lack of drive and motivation to do so is the only real roadblock and since I'm not part of the development team, it isn't my place to make such corrections.

That being said, it does seem that it currently may trigger to early. But again this should be improved by improving the alorithm through which the AI guesses its opponents strategy.

Which is exactly what I was saying. Really. All I wanted was a correction to this modifier, not to have it removed.

Because in early human history no conflicts have been fought over the perceived aspirations of other nations?

No, because in early human history, no one was eying up space ship victories and being a diplomatic master.

Again, you are simply saying that "The player should not always be able to start good relations with every leader." This is fine. However, I am saying, "There should be a separate modifier that has nothing to do with victory conditions."

In your comparison to the real world, you keep everything very neat and tidy. Early civilizations had conflict with other civs b/c of their aspirations. That's all well and good, but I'm pretty sure each civ actually HAD aspirations and were working TOWARDS those aspirations and each civ KNEW what the others aspirations were. They weren't sending thousands of men across hundreds of miles to potentially lose on a whim of what they might be doing.

To bring this back to the game, if there are no victory conditions CLEARLY being sought after yet (and YES there is a point at which it is far too early to make any good assumptions), then the "You are trying to win the game in the same manner..." is just plain wrong.

Once again, I have no problem with a new modifier that more aggressive or isolationist leaders are likely to have "We just don't like you." or "We just don't like your culture." But you have to be honest, this whole business of bad diplo relations starting in 2000BC because the AI thinks it knows my victory plans is just poor design. AGAIN, it is only going to hurt the AI since it will be missing out on trades, RA, and potential war allies because of a [likely] FAULTY assumption.

 

[CoolLizy]

Also, and maybe this is just me, but I don't mind the AI trying to win so much as I mind the fact that they've boiled "the AI winning" down to the AI acting like a bunch of bipolar, psychotic seven-year-olds who throw a fit anytime they're not in the lead. I'd prefer to sit down to a game where the AI, while all trying to win, make it feel like I'm sitting down to a game with friends who want to have a good time. Sure you'll always have that friend that talks big, won't shut up, backstabs everyone else around the table in an attempt to get ahead, and tries to hide extra cards under the table. AI characters like Monty should be playing like that.

But as it is, I'm actually more wary of "friendly" AI civs than "guarded" ones (and sometimes even "hostile" ones) because "friendly" tends to be nothing more than a five turn facade they throw up every time they're ready to declare war and think they're being witty about it. It's almost as bad as the CivRev AI players, where diplomacy boils down to them coming up and telling me they'll give me peace for x number of turns if I give them whatever junk they want at the moment.

 

[qemist]

This. And I understand that (for realism), not every AI should play nice every game.

In my current immortal game my nearest neighbor, Gandhi, has been cooperative for 120+ turns. So they don't "always DoW you by turn 50" as some say.

 

[qemist]

Also avoiding trading with a player only when he is already near a victory typically is too late. I therefore makes sense to start to frustrate your nearest rival the moment you identify him.

I don't know what you mean by "nearest rival". If I am the closest to victory my nearest rival is whoever is coming second. If I am in any other position my nearest rival is whoever is first. That is not the logic the AI currently employs. You can be third and they will be snubbing you because you are pursuing the same victory condition (which according to you means you are their nearest rival) while happily trading with #1.

 

[headcase]

In my current immortal game my nearest neighbor, Gandhi, has been cooperative for 120+ turns. So they don't "always DoW you by turn 50" as some say.

Yeah, he plays nice to avoid war. It might slow down his... research

 

[blitzkrieg1980]

In my current immortal game my nearest neighbor, Gandhi, has been cooperative for 120+ turns. So they don't "always DoW you by turn 50" as some say.

Oh, I know it. I have had plenty of DoW before turn 100 (epic), but it's usually only 1 leader (not all leaders). I'd say in 67% of my games, I've had a DoW before turn 100, but it's never been declared by more than 1 leader. Also, I've had a lot of games with Alexander and Monty as nearby neighbors, so you know how that goes.

I'm in agreement with you on this point. What I was trying to say was that I understand that there should often be tension between civilizations that spawn close to each other since it is realistic that early era civs that are geographically close would be likely to have conflict (ie not every leader should be able to be made friends with early on in every game). But this is well enough reflected in the "We covet your lands" modifier. It has nothing to do with the "winning the game the same way" modifier (which is what Trias and I were discussing).

 

[Trias]

Which is exactly what I was saying. Really. All I wanted was a correction to this modifier, not to have it removed.

I must have misunderstood your intentions then.

Again, you are simply saying that "The player should not always be able to start good relations with every leader." This is fine. However, I am saying, "There should be a separate modifier that has nothing to do with victory conditions."

The modifier is not (other than its current phrasing suggests) about victory conditions. Its a rudimentary diplo modifier against the opponent that the AI deems to be its strongest competitor. (The restriction to just competitors for the same goal, ignoring the other victory conditions, is (I think) artificial to prevent all the AIs ganging up on the same player.)

It is never too early to try to figure out who is going to be your biggest rival. (For example, this is an important factor for diplomatic positioning when blocks are starting to form.)

 

[Trias]

I don't know what you mean by "nearest rival". If I am the closest to victory my nearest rival is whoever is coming second. If I am in any other position my nearest rival is whoever is first. That is not the logic the AI currently employs. You can be third and they will be snubbing you because you are pursuing the same victory condition (which according to you means you are their nearest rival) while happily trading with #1.

For artificial reasons the AI only looks at the rivals to their chosen victory condition. So "nearest rival" in this case means player(s) that it perceives to furthest along towards pursuing that victory condition. Although we can't be entirely sure what metric it is using to determine this.

 

[blitzkrieg1980]

It is never too early to try to figure out who is going to be your biggest rival. (For example, this is an important factor for diplomatic positioning when blocks are starting to form.)

The player will almost always be the AI's biggest rival. Its [often] flawed assumptions that the human will be competing for the same victory conditions before there is any basis to form this [often] flawed assumption is handicapping the AI. If the human player has to make absurdly lopsided trades from the start b/c the AI thinks it is competing for the same victory in 2000BC, then the human will often opt to not trade with this AI and focus on the AI who haven't made absurd assumptions based on God-knows-what. Why would I choose to trade with an AI that requires not only 1 excess lux resource (and sometimes they ask for my ONLY source of a lux), but also a chunk of gold AND some strategic resources, when I can go to another AI for a straight lux for lux trade?

This is hurting the AI far more than the human (as I've been saying all along). Besides the loss of a trading and RA partner, the AI also makes itself a target early on since it has decided to start things off on a bad foot.

This "winning the game the same" modifier is busted and needs constrictions to make it work properly. I'm pretty sure you are the only one who sees it as working perfectly fine.

Put in a "We just don't like you" modifier or maybe a "We just don't trust you" modifier and be done with it. Leave any sort of "winning the game the same" modifiers for a portion of the game where a victory condition is actually being actively sought after.

 

[qemist]

Oh, I know it. I have had plenty of DoW before turn 100 (epic), but it's usually only 1 leader (not all leaders). I'd say in 67% of my games, I've had a DoW before turn 100, but it's never been declared by more than 1 leader. Also, I've had a lot of games with Alexander and Monty as nearby neighbors, so you know how that goes.

Gandhi was still playing nice at turn 230 when I gave the game up. You didn't say what level you play at but I've had double DoWs by turn 100 on immortal.

I'm in agreement with you on this point. What I was trying to say was that I understand that there should often be tension between civilizations that spawn close to each other since it is realistic that early era civs that are geographically close would be likely to have conflict (ie not every leader should be able to be made friends with early on in every game). But this is well enough reflected in the "We covet your lands" modifier. It has nothing to do with the "winning the game the same way" modifier (which is what Trias and I were discussing).

For sure. I don't think anyone except Trias defends the "winning the game the same way" modifier. It's so anti-immersion and illogical.

 

[Trias]

This is hurting the AI far more than the human (as I've been saying all along). Besides the loss of a trading and RA partner, the AI also makes itself a target early on since it has decided to start things off on a bad foot.

Why should the AI try to avoid being a target?

I'm also not sure if it really is hurting the AI, that much. First of, trading with the human player, most of the time is more beneficial to the human player. (because of the limits of the AI's cost benefit analysis) I don't think that should be a consideration of the AI (it should treat its human and AI opponents on the same level - as far as that is possible). Moreover, although the loss of an early trading partner maybe a temporary disadvantage, ending up in the "right" diplomatic block is somewhat of an offsetting advantage on the longer term.

As limited as the considerations of the AI are. At least the AI is following some sort of long term diplomatic game plan. (Which is a huge improvement over previous iterations of the game, where the AI simply had no long term plan whatsoever).

This "winning the game the same" modifier is busted and needs constrictions to make it work properly. I'm pretty sure you are the only one who sees it as working perfectly fine.

I've never said that its working fine. In fact, I've repeatedly said that it needs improving. (As do most AI algorithms) I've been combating your repeated assertion that having a modifier of this type at all is utterly ridiculous.

Put in a "We just don't like you" modifier or maybe a "We just don't trust you" modifier and be done with it. Leave any sort of "winning the game the same" modifiers for a portion of the game where a victory condition is actually being actively sought after.

Such a modifier would make no sense at all from the point of view of the AI. Introducing such an arbitrary modifier to the AI, would be a huge step backwards for the AI.