From Prince to Deity

[tozka]

I was wondering if deity AI is "smarter" than if you play on prince level (or any other lower level). Or the level just gives them better starting positions (like two settlers) , less happiness for me, etc. Because Deity is way too much for me. I get hammered too quickly there before I begin developing, but on prince, the AI keeps making stupid decisions. Like there was a moment that 7 civs declared war on me at the same turn. I didn't even have the strongest military (per my adviser, though I was fairly strong) but they kept sending troops one by one which I easily defeated. If they have coordinated a little, attack me at the same time I'd have been in trouble. It's just a pity. It kills the enjoyment of the game.

[OniaPL]

When you up the difficulty, the AI gets more bonuses and chooses research/production better. This doesnt affect the AI in the sense of combat strategy, however. I don't know if difficulty affects aggressivity.

[Easy Money]

From what I can gather from the files. The more units the AI has over you the more aggressive it is. It doesn't seem to matter if those units are superior or inferior to you, it is just the quantity. When they have a larger military they tend to be more aggressive to those who don't. That and their production bonuses and unit spam make them formitable, only in the fact that you must manage your units better. I would say that the AI is the same but their bonuses force YOUR game to be better if you want to survive.

[Pekish79]

I only played Emperor and Immortal (several time)

Emperor I can manage to win i won't say easily but just paying attention on Immortal i find they have a huge advantage and you have to play aggressive to keep it up
Aggressively in a smart way not military talking (since it's true they sux at moving army but they have too many more then you anyway) but aggressively as position on territory... you have to use your city to block their grow found city in area where they will block them position army on the map on key point in order to force them to go other direction to expand... ignoring every request of "open border" specially the one from ur neigh board....

you can just hope to mind ur own business and develop faster then them is impossible they have a bonus too high



i think the higher the lvl the more thing the AI ignore like happiness or army maintenance... they just don't have them

I really would rather see the AI getting better with lvl then the AI cheating more and more but i guess it's a lot more development making several lvl of AI rather then turn off some feat.

[Easy Money]

...I really would rather see the AI getting better with lvl then the AI cheating more and more but i guess it's a lot more development making several lvl of AI rather then turn off some feat.

So would I, but unfortunately a Civ V AI that can effectively compete with a human without bonuses, is beyond the reach of programming at this time. The only way to level the playing field for our human advantage is to give the AI bonuses. Either one requires the human to up their game, which is the goal of the programmers...

[OniaPL]

Funny example of immortal: When i had produced scout and a worker and was halfway to settler (no liberty), Caesar had 3 cities, scout, few warriors and 1 or 2 workers.

[OdinTGE]

The more units the AI has over you the more aggressive it is.

This entire aspect of the AI pisses me off to no end. How the hell do they know how many units I have or what they are? I can't see miles and miles into their borders?

How do they know I'm not hiding 43 tanks just outside their visibility...

[Easy Money]

This entire aspect of the AI pisses me off to no end. How the hell do they know how many units I have or what they are? I can't see miles and miles into their borders?

How do they know I'm not hiding 43 tanks just outside their visibility...

I feel your pain... The only comfort I can offer is to say that your "human" advantage is so unbalancing to the AI that if they were given no bonuses you would win most of the time and the times you didn't, it would be more your fault than the AI playing well.

[Pekish79]

So would I, but unfortunately a Civ V AI that can effectively compete with a human without bonuses, is beyond the reach of programming at this time. The only way to level the playing field for our human advantage is to give the AI bonuses. Either one requires the human to up their game, which is the goal of the programmers...

I know, I don't expect it actually, I was just saying...
I wonder which type of AI do they try to do

the one that calculate the chance and blindly follow the highest %

the one that follow some type of pre-imposted model/plan.

Most probably the chance one since it seems obvious that sometimes the army just move one way then you move and change the % of victory and he retreat even if losing a battle could lead to win the war.

Sometimes I think the pre-imposted model of reaction would work better a
trying to give a sort of a "philosophy" to the AI other then the % calculator of success that actually make the computer follow a sort of "plan". But perhaps they already doing something similar with the different type of character

Honestly i think they did a good job on the large scale they just need to tweak the small scale battle system AI need to stop fight randomly at a certain LVL of difficulty

[Sethos]

So would I, but unfortunately a Civ V AI that can effectively compete with a human without bonuses, is beyond the reach of programming at this time. The only way to level the playing field for our human advantage is to give the AI bonuses. Either one requires the human to up their game, which is the goal of the programmers...

Sorry but just because it cant act like a human doesnt mean the AI couldnt be far better. Not to mention that most of "AI" things could simply be made with scripts, create different AIs with different research order and build-strategy. Make them actually use the new combat system. Ad more values when it comes down to diplomacy. Decide their relation to you for example more with what they actually can see, for example your units moving near them, you attacking allies of them etc.

AI can pretty much outclass any humen when it comes down to build-order and science, the only thing that seems hard is fighting. Yet they fail at both ... (for example it shouldnt be very hard to programm an AI to do the perfect SC2 - XY Build-order) And thats what really annoys me, the AI is failing at things that should be really simple scripts, its not like there would be an actual need for the AI to decide itself what to research and where to build. Just create different AIs for different victory conditions with different "personalities" (and the higher the difficucly the more efficient they get). AIs failing to create normal positive income is just retarded ... (the hard part should be the interaction between player and AI not those simple things ...)

[Easy Money]

Sorry but just because it cant act like a human doesnt mean the AI couldnt be far better. Not to mention that most of "AI" things could simply be made with scripts, create different AIs with different research order and build-strategy.

Scripts would reduce the AI not enhance it. For example, a script is rigid and predictable. Exactly what this AI doesn't need.



Make them actually use the new combat system.

I've noticed they do on larger maps and higher difficulties. Not to the level of a human of course, but they do better on these settings.



AI can pretty much outclass any humen when it comes down to build-order and science, the only thing that seems hard is fighting. Yet they fail at both ... (for example it shouldnt be very hard to programm an AI to do the perfect SC2 - XY Build-order) And thats what really annoys me, the AI is failing at things that should be really simple scripts, its not like there would be an actual need for the AI to decide itself what to research and where to build. Just create different AIs for different victory conditions with different "personalities" (and the higher the difficucly the more efficient they get). AIs failing to create normal positive income is just retarded ... (the hard part should be the interaction between player and AI not those simple things ...)

Again, I think this lends itself to AI predictability and would be the wrong direction. I guess we'll have to agree to disagree ;)

[trainzebra]

Scripts would reduce the AI not enhance it. For example, a script is rigid and predictable. Exactly what this AI doesn't need.



I've noticed they do on larger maps and higher difficulties. Not to the level of a human of course, but they do better on these settings.



Again, I think this lends itself to AI predictability and would be the wrong direction. I guess we'll have to agree to disagree ;)

He's actually mostly correct. The macro game of Civ V is one big math problem that can be solved. Given your resources, what is the fastest path to the victory the AI player wants to achieve. If it's diplomatic the AI should mathematically optimize its gold, culture for culture, science for space race, and production for domination. The script should be flexible enough to account for changing conditions, but it's fundamentally a (very complex) math problem.

I think he's underestimating the difficulty a tad though. It's far, far more complicated than an SC2 build order. In SC2 your variables are resource gathering and build time. In Civ V your variables are city placement, expansion rate, available tiles, potential neighbor expansion, city growth, which tiles to work, and more. A human has to solve that equation before he can tell an AI how to =p

Combat is a whole different beast all together. There's never an easy solution to teaching a computer to intelligently play a game with such a huge available collection of actions. It can definitely be made better than it is now without too much trouble though =p

[inseeisyou]

Scripts would reduce the AI not enhance it. For example, a script is rigid and predictable. Exactly what this AI doesn't need.

I think it is great that you are trying to defend the game, but we need to be honest about some serious AI issues. There are some issues that are just blatantly a problem no matter how you think the AI should play. Ranged units simply passing their turn and not attacking when they have the opportunity (no they are not healing) is simply... broken. There is no way to spin that as OK AI behavior.

AI units moving toward a city and then retreating back without attacking and just taking indirect fire (no they are not somehow protecting another unit) is simply broken. The AI seemingly neglecting a coordinated attack on its capitol... again... you see the pattern.

They got lots of things right, but we don't want to sugar coat the problems.

[Morthis]

I think quite a few people are posting in here regarding AI but haven't actually done programming or worked on AI before to understand that sometimes what sounds like a very simple concept can be very difficult to program into an AI. Especially if you want to make one that doesn't just follow a series of scripts every single game.

I agree the AI needs work, and in the combat department, it fails pretty catastrophically sometimes, but let's at least be fair and not portray this as something that is incredibly easy to code. Hell, I don't believe the AI used a SoD in any of the Civ games until Civ IV warlords, and that's a pretty simple concept compared to the 1UPT we have now.

[Easy Money]

I think quite a few people are posting in here regarding AI but haven't actually done programming or worked on AI before to understand that sometimes what sounds like a very simple concept can be very difficult to program into an AI...

This has been the underlying point of all my posts and is the reason, like it or not, the AI needs bonuses to be competitive at higher levels. Until a way is discovered to effectively and efficently code an AI for human experience, I fear we'll be having these conversations for along time and getting nowhere.

My suggestion is play Deity on Pangea, Standard world or larger. If you can't beat this (and most people won't be able to) then your game needs work. The good thing about these higher levels is they do force YOU to play better. They are less forgiving of errors and to me, this is the goal of the series. To provide a challenge by the best means available, and currently (unfortunately to some) this is with AI bonuses.

With that being said; could it be done better in theory? Yes. I think we're all in agreement here; that an AI that needs no bonus and can consistently beat the best human players on the highest levels would be the pinacle of Civ AI achievement. But my cynical side wonders, if this were the case, how many of these same people who now complain about the current AI would also complain that this "new" AI is too difficult.

[inseeisyou]

This has been the underlying point of all my posts and is the reason, like it or not, the AI needs bonuses to be competitive at higher levels. Until a way is discovered to effectively and efficently code an AI for human experience, I fear we'll be having these conversations for along time and getting nowhere.

My suggestion is play Deity on Pangea, Standard world or larger. If you can't beat this (and most people won't be able to) then your game needs work. The good thing about these higher levels is they do force YOU to play better. They are less forgiving of errors and to me, this is the goal of the series. To provide a challenge by the best means available, and currently (unfortunately to some) this is with AI bonuses.

With that being said; could it be done better in theory? Yes. I think we're all in agreement here; that an AI that needs no bonus and can consistently beat the best human players on the highest levels would be the pinacle of Civ AI achievement. But my cynical side wonders, if this were the case, how many of these same people who now complain about the current AI would also complain that this "new" AI is too difficult.

Very thoughtful post, and I agree with the majority of it. In the absence of an AI that can consistently beat players with even forces, I agree that an AI that provides difficulty through superior numbers and technologies is better than nothing. It does force one to improve your strategies. I think it is just the blatantly stupid actions that really get under people's skin, however I do have faith that these issues will be addressed as long as we continue to bring them to the devs attention.

[Easy Money]

...I think it is just the blatantly stupid actions that really get under people's skin, however I do have faith that these issues will be addressed as long as we continue to bring them to the devs attention.

Agreed. But in my experience, these are the exceptions not the rule. We don't notice the decent moves the AI makes all the time, but the ignorant one sticks out like a sore thumb. Like in Civ IV when you lost a 90% chance to win. At that moment, you forgot that you've won countless others at those percentages, and that 90% means, on average, you will lose 1-in-10. Humans recognize the absurd by design more often than the norm, its an evolutionary mechanism.

As a comparision, it isn't like the AI is moving units randomly. That would be a problem. There is some strategy there, just look at the XML files to see all the things the AI tries to juggle. Humans make ignorant moves also; to expect the AI to perform better than a human and make fewer ignorant moves than a human is not realistic.

I've personally made several ridiculous moves in Civ V, more than any other series, if for no other reasons the ZOC and the occasional river :mad: and I've beat all the difficulty levels. Granted Diety was a tiny archipelago, but in my defense that was before I realized the AI built a minimal navy that in my experience, consists mainly of un-upgraded caravels (granted it may not have had enough oil to upgrade them)

In a nutshell. Until you can beat the best the game has to offer, asking for anything more difficult is odd.

[trainzebra]

Agreed. But in my experience, these are the exceptions not the rule. We don't notice the decent moves the AI makes all the time, but the ignorant one sticks out like a sore thumb. Like in Civ IV when you lost a 90% chance to win. At that moment, you forgot that you've won countless others at those percentages, and that 90% means, on average, you will lose 1-in-10. Humans recognize the absurd by design more often than the norm, its an evolutionary mechanism.

As a comparision, it isn't like the AI is moving units randomly. That would be a problem. There is some strategy there, just look at the XML files to see all the things the AI tries to juggle. Humans make ignorant moves also; to expect the AI to perform better than a human and make fewer ignorant moves than a human is not realistic.

I've personally made several ridiculous moves in Civ V, more than any other series, if for no other reasons the ZOC and the occasional river :mad: and I've beat all the difficulty levels. Granted Diety was a tiny archipelago, but in my defense that was before I realized the AI built a minimal navy that in my experience, consists mainly of un-upgraded caravels (granted it may not have had enough oil to upgrade them)

In a nutshell. Until you can beat the best the game has to offer, asking for anything more difficult is odd.

I'm going to have to disagree with this. Beating the game on Deity right now isn't about outplaying it so much as it is exploiting how bad it is at war. Read some of the stories about Deity wins over on Civ Fanatics. They generally find an area of the map to exploit how bad the AI is at war, wittle his forces down there, then conquer them.

It's like playing chess with your 5 year old cousin and giving her twice as many pieces as you to make the game interesting. Yeah it might be challenging, but it's not nearly as satisfying as playing against an intelligent opponent who understands the rules.

It's not a "theory" that the AI could be better, it's a fact. There are plenty of hex and grid based games out there that are more effective at combat than Civ V's AI. There's nothing wrong with asking for an intelligent, thinking opponent because you don't want to play a 5 year old with double your army.

[Easy Money]

...They generally find an area of the map to exploit how bad the AI is at war, wittle his forces down there, then conquer them...

I don't agree that this is a "legitimate" win. To beat it by an exploit seems cheap. But I understand your point, but disagree that programming an AI is as easy as most think it is (coming from a programmer, but not an AI programmer).

[Easy Money]

...Read some of the stories about Deity wins over on Civ Fanatics...

Could you post a link? I'd like to read these stories.

[trainzebra]

Could you post a link? I'd like to read these stories.

I couldn't point you to a specific thread unfortunately, I've just read different posts in random threads over the past week and a half. "Exploit" is probably too strong of a word, there is skill involved in it without a doubt. By exploit I mean things like setting up a strong defense and letting the AI decimate itself against it because it doesn't logically move its troops.

Programming an effective AI is difficult without a doubt, but programming an AI that's more effective than what Civ V is currently using is definitely within the realm of possibility. The fact that you have to point to Deity to demonstrate a challenging way to play the game is in itself a failing.

A week after Civ IV came out no one was even considering beating Deity, it took total mastery of the system to do it. Granted, Civ IV's AI didn't really come into its own until BTS and it is a much easier game to code combat AI for (make big stack, move big stack to city), but it does a good job of demonstrating how the difficulties should scale. A reasonably effective AI that's given more resources as difficulty increases. It's a strong challenge to the average player on Noble, it's insane on Deity. I don't have a problem with giving the computer "cheats" to increase difficulty, but the engine underneath has to be effective before I'm willing to accept it as a sound design.

[saintus]

So would I, but unfortunately a Civ V AI that can effectively compete with a human without bonuses, is beyond the reach of programming at this time. The only way to level the playing field for our human advantage is to give the AI bonuses. Either one requires the human to up their game, which is the goal of the programmers...

While I do agree in general, it is not as if we are expecting an IBM Blue machine to defeat the best chessplayer or something. We only want decent AI and that is possible to achieve. The problem is that not enough efforts have been put into that field. Back in the 1980:s there was a lot of turn-based strategy games and many of them had quite ok A.I. I refuse to accept the fact that AI programming is more difficult today, even though todays games like CIV V might appear more advanced it is still tile based tactical combat which should be quite easy to do well enough.

[saintus]

I think quite a few people are posting in here regarding AI but haven't actually done programming or worked on AI before to understand that sometimes what sounds like a very simple concept can be very difficult to program into an AI. Especially if you want to make one that doesn't just follow a series of scripts every single game.

I agree the AI needs work, and in the combat department, it fails pretty catastrophically sometimes, but let's at least be fair and not portray this as something that is incredibly easy to code. Hell, I don't believe the AI used a SoD in any of the Civ games until Civ IV warlords, and that's a pretty simple concept compared to the 1UPT we have now.

You are right. It is not easy. At least not for perhaps one or two persons responsible for that field. But why have perhaps 20 people working on graphics, art and sound and so little in AI ? If I´ve already waited 4 years for CIV V I can wait another 6 months for decent AI.

[Satoru]

I think people are comparing AI with chess or Starcraft are unfortunately wrong on several orders of magnitude. Chess actually has a finite set of moves. Even in this scenario it requires a computer the size of small apartment to compete with the best chess players in the world. The only comparable game currently would be Go. Here the best computer players can't even beat amateurs consistently. Civilization is orders of magnitude more complicated. You have to manage long term strategies of empire building, resource allocation, exploration. Short term strategies of combat and tile improvements. Then you have to do this 10-15 times over for each AI opponent since they each have their own competing goals. You then have to make this flexible enough because your map structure is random every single time.

Could the AI be better in Civ5? Sure I think we agree this is definitely needed. But the AI couldn't compete with a human player unless you either have an alien obelisk with some quantum computing power, or you want to wait 5 hours per turn.

[Satoru]

While I do agree in general, it is not as if we are expecting an IBM Blue machine to defeat the best chessplayer or something. We only want decent AI and that is possible to achieve. The problem is that not enough efforts have been put into that field. Back in the 1980:s there was a lot of turn-based strategy games and many of them had quite ok A.I. I refuse to accept the fact that AI programming is more difficult today, even though todays games like CIV V might appear more advanced it is still tile based tactical combat which should be quite easy to do well enough.

The IBM Blue was simply a brute force machine designed to scour through the entire game space of chess to come up with an optimal strategy. Even in chess which in theory has a fairly finite game space, they had to create algorithms to define how to best prune the game space tree in order to compute the best move available. Again this is in a game with strict rules, and a single opponent, with a finite number of rules. To make a Civ AI, it essentially needs to be several orders of magnitude more powerful to address all the complex things it needs to take into account.

AI programming is very very very hard. There are entire CS departments dedicated to trying to figure this out. They are making slow and incremental improvements. But you are seriously underestimating how difficult this problem is.

[saintus]

I do not expect an AI to beat 99% of the human players. I expect the best possible AI for this kind of game and I don´t think we have reached that goal yet with this current release. Besides, it´s not that Firaxis have bragged about the resources put into the AI or anything that have had us believe it was prioritized in anyway.

[tozka]

Well, I don't expect the AI to have human intelligence. It is just that there are some simple things - like getting their workers away from my armed troops.One worker sat 5 turns around my city, I even shot him with the city defense but did not capture it, so that I can see if the AI would think of pulling it back. Or wounded soldier goes deeper while surrounded by swordsmen. I know that it is not easy to program AI but there are some basic things the AI should be able to do.

[Azzer007]

Having just finished my 2nd game at Deity level, I have to say I'm left... disappointed. Pretty easy wins, no sense of achievement. I've played other games with far better AI's. As a modern example, Sup Com 2, with it's latest set of patches, has AI's that provide MUCH more challenge and strategy (eg they can alter their strategies if something they attempt fails, they learn where your most focused defense points are and attack weaker sides, they can see your armies moving and intercept them, they know how to use terrain advantages etc. - sure it's far from genius, but it's several orders of magnitude above Civ V's totally "blind" AI). But note that sup Com 2 had TERRIBLE AI on release, they had to spend numerous patches and months to get it to it's current levels... so if Civ V devs can put that same level of dedication in...

Sure sure, it's a different game, yadda yadda, but that doesn't mean they can't have AI's that can identify weak points, or recognise a failed assault and attempt a different strategy next time (in Sup Com 2 they can do this two ways, by probing different parts of the map for weaknesses, or if a small army fails, they hold back and amass a much larger army on the next assault to try to overwhelm your defences). There is no attempt at this whatsoever in Civ V. No altering of strategies. No learning from mistakes. No formations of troops. No using terrain advantages. These things are far from the realms of the impossible.


I don't agree that this is a "legitimate" win. To beat it by an exploit seems cheap. But I understand your point, but disagree that programming an AI is as easy as most think it is (coming from a programmer, but not an AI programmer).

Wait a minute - so, using Civ V's new one unit per tyle system, ranged and artillery system, to have a strategic advantage, is now an "unfair" win? THis is how I've beaten the AI's - using superior strategies, not dissimilar to what 2k Greg attempted (but failed miserably :P) at in that UStream video a short while ago. So the choices are "Utilise the game's new strategic systems and decimate even the hardest AI's" but this is completely unfair and an exploit, or "Blindly send wave after wave of attack with no thought of location range or tactics" so that the AI has a chance but defeating the point of many of Civ V's new features? Hmm. I see. So what you're saying is the AI's not THAT bad, just as long as we play one-handed and blindfold ourselves, rather than "exploit" by actually trying to play properly :P

[Easy Money]

...So what you're saying is the AI's not THAT bad, just as long as we play one-handed and blindfold ourselves, rather than "exploit" by actually trying to play properly :P

I wouldn't have put it that way, but yes, the AI won't get much better so, for now, its either use your human advantage or play one-handed and blindfolded ;). The AI for quite some time won't match the play of a human player, maybe it never will. Once the MP kinks are worked out though, I think the best players will migrate towards MP for the challenge they are looking for.

Also, don't take my "stating the facts about the Civ V AI" as in any way agreeing with the way it has been done. I'm saying that if it gets better, it won't be by much and will be nowhere near the level of playing against a human. We've had these issues since Civ I and although I realize that is not a defense, at the same time, I realize that it's a fact.

[Azzer007]

Also, don't take my "stating the facts about the Civ V AI" as in any way agreeing with the way it has been done. I'm saying that if it gets better, it won't be by much and will be nowhere near the level of playing against a human. We've had these issues since Civ I and although I realize that is not a defense, at the same time, I realize that it's a fact.

Hey don't be too pessimistic! As I said above (and sorry for those that don't like me "comparing games"), the AI in Sup Com 2 on release was AWFUL - just as bad as the current Civ V AI (as much as they can be compared from two different strategy games). After around 6 or 7 patches that all heavily altered the AI, it's now sublime. It's not the most amazing AI in the world - but it's really good, and enjoyable to play against (particularly with a mate - two of you versus a handful of hard AI's is a great laugh).

Even if the developers can't be bothered/are not skilled enough at AI coding, I'm hopeful that on the release of the C++ code moddy stuff (something I've heard mentioned a few times), there's a good chance that some serious "bedroom programmers" (who everyone know possess the REAL skill - the Sup Com 2 AI has been done by somebody who was 'just a modder' that modded AI for Sup Com 1 and got hired by the devs to work on Sup Com 2!) will step in and fix it for us.

We're both spot on though - until this happens... yes, we have to play blindfolded with one hand behind our back against the AI, which is just as unsatisfying as absolutely decimating them with superior tactics while they roll over :(