Does AI always know the location of Iron?

[TheMeInTeam]

In case you wonder why I'm addressing this: I believe that such false statements about the AI are one of the reasons why Civ5 has such an abysmal AI. You can listen to Sid Meier talking about game AI in his GDC 2010 keynote address, he says that an AI must not be too clever because the player will just say that it cheats. (Source: here). Civ4 has a pretty good AI that removed many cheats that the Civ3 AI still had - but people are still just saying that it cheats when it does something well. The logical conclusion for game developers is that it doesn't pay to develop a very good AI because lots of players will simply refuse to acknowledge it. Therefore it's much more cost efficient to develop a game with a weak AI -> see Civ5.

Bogus. The AI isn't good in any of the 3 games, regularly getting trounced by mediocre play without very heavy bonuses. The "good" civ IV AI has no concept of any victory conditions aside from culture (bts only though) and space, and it's not even very good at space! It has 0 tactical prowess and chooses techs horribly. The majority of the "cheating" complaints you see in force on the forums are either actual cheating (seeing into the fog w/o moving there, displaying false information on the UI, knowing about trades they can't possibly know about yet, etc) or strictly unlucky RNG outcomes (the most common complaint) which have absolutely nothing to do with AI play quality and only a little to do with its bonuses (IE more units + more fights = more chances at a high-odds loss). Sids comment about people complaining that a "good" AI cheats is baseless and frankly comes off as an excuse for bad AI, which every iteration of civ has.

Civ then becomes a game of "gaming" AI stupidity instead of maximizing output based on what the map/situation allow (IE RA + gold farming AI in civ V overwhelm general good play), and good players are literally forced into MP to have a game which has dynamic strategy/valid options...of course whether it is balanced even there remains debatable, given how ban-happy some communities are compared to others (banning BLOCKADES is in danger of becoming standard on realms beyond! Blockades!).

Anyway the claim that people think the AI cheats when it does something well is baseless. I see that only very rarely. Even in the threads about the AI knowing about the resources, people are suspecting the AI because it is playing POORLY (IE settling utter junk land) and getting LUCKY (site happens to have iron). Settling junk cities so early like that is undeniably AWFUL play; nobody good does this when pressed like in MP. However, low and mid-level AIs hamstring themselves this way while high level AIs pay so little in maintenance they can get away with it. Derp bonuses =/= good AI. Bad strategy (settling junk land that MIGHT have a resource but we don't know it yet) does not constitute a good AI. AIs that DONT TRY TO WIN BY THE GAME RULES? How is that supposed to be "relatively good"? Strictly speaking, none of the AIs in the flagship series know how to play their own game.

The AI does get to see everything in the squares it can move to, right? Which means for late game naval units, they have a sight range of about 5+?

Yes, this is accurate.

 

[Psyringe]

The AI does get to see everything in the squares it can move to, right? Which means for late game naval units, they have a sight range of about 5+?

This is correct. Soren explained in his Googletech talk why he felt the need to do this: Human players can see in which direction the enemy unit is moving into the fog, so they can make a good guess how to follow it. The AI cannot see the moving unit (they just don't see it anymore, meaning the AI actually forgets that there ever was a unit), and (within the Civ4 code framework) cannot be programmed to do so in any efficient manner. Hence, Soren felt the need to balance this AI disadvantage by giving its units a larger visibility range.

Personally, I think that this was one of the poorer design decisions, because now obviously the player is at a disadvantage. I think it would have been better to let the AI sometimes fail to track a unit inside this extended visibility range, that would be more akin to the player's method of educated guessing.

The interesting thing about this specific issue is that it's often (and correctly) brought up as an argument that "the AI cheats", but usually the posters who do so overlook the fact that without it, the AI would actually be at a disadvantage.

========================================

Bogus. The AI isn't good in any of the 3 games,

That depends on what you compare it to. If you compare the Civ4 AI to an idealized "perfect" AI that plays a game of Civ4's complexity competently and without cheating, then of course it isn't very good, because it does require cheats to stay competent, and even then it has severe weaknesses (lack of goal orientation in the unmodded game, issues with tactical movement). However, such an ideal AI is absolutely impossible to program with the AI concepts and hardware restrictions we currently have, so it's not a very practical approach. Personally, I prefer to compare it to other actually existing AIs. And in this comparison, the Civ4 AI is easily in the top layer. Compared to the Civ1 AI ("Oh, it's 1900 and the player is in the lead, let all AIs attack him!", Civ2 AI ("Oh, this AI is lagging behind, let's gift it a free world wonder!"), and Civ3 AI ("Oh, let's settle this worthless spot in the middle of the desert, because I just KNOW that 5000 years from now I'll find oil there!"), the Civ4 AI is easily the best of the series. Compared to other game series, it also holds its ground very well, considering the complexity of the game, the adaptivity of the AI (enabling it to understand a lot of mod-added content without requiring additional code), and the time constraints of a TBS game (player has usually a lot of time for his turn, the AI needs to handle several rivals very quickly to keep the lag bearable).

Anyway the claim that people think the AI cheats when it does something well is baseless. I see that only very rarely.

It has thinned out. It was a very common occurrence when the game was released (which was probably the time when feedback mattered most and/or was most likely to be spotted by Firaxians). Examples I remember off the top of my head:

- People claiming that the AI cheated because it attacked from far away, i.e. the AI couldn't have seen that the player defended his cities with only one spearman each, in good old Civ3 tradition. Actually the AI had just reacted to the power graph, which is visible to the player as well.

- People claiming that the AI cheated because they didn't win a battle they expected to win. Actually the AI had just chosen appropriate promotions for the defending units.

- People claiming that the AI cheated because they ended up in a war against multiple AIs. Actually they either hadn't paid much attention to the state of their relations (underestimating the effect of religion was rather common in the early days), or several AIs went into dogpile war mode, and the player interpreted that as an anti-player-bias, not understanding that it happened to weak AI civs as well

I agree with you that Sid is going into the wrong direction by directing the development focus away from the AI, but I do not agree that this was simply a baseless excuse on his part. I think there's (unfortunately) been enough feedback even here on the more hardcore forums to warrant the very unfortunate conclusion he arrived at.

In such a situation, the hardcore fans _could_ become a factor in the equation if they pushed for a good AI. Unfortunately, hardcore fans who gripe about the existing weaknesses in an already quite good AI without offering workable solutions how to match these very high expectations, don't help that cause. Imagine you're a game publisher. You notice that AI is a part of development that's expensive and fickle, and even if it marks a massive improvement over previous iterations in the series, lots of players won't even recognize that (because they simply assume the AI to cheat), while the hardcore fanbase doesn't appreciate it either (because it has unrealistic expectations and focuses on the weaknesses that are still there). What would you do, where would you put your money? Food for thought.

 

[TheMeInTeam]

the Civ4 AI is easily the best of the series. Compared to other game series, it also holds its ground very well, considering the complexity of the game

Please tell me you are kidding. Civ IV AI only rarely tries to win and can't even execute a general concept of putting national wonders in a city where their multipliers are most relevant. An AI that doesn't try to win is somehow GOOD? No. I won't buy it. I don't expect an AI to play 100% perfectly (as you said, that's impossible), but this is ridiculous. Most AI know how to play their game and try to win. Late 90's AIs played their game better than civ IV AI in a LOT of respects. The civ V AI is even arguably better than the civ IV one, because it has a higher incidence of trying to win (IE a higher incidence of playing the game). I'm not sure how much credit civ IV AI should get for being "best in the series" considering how awful it has been on a consistent basis in the series, but even THAT point is arguable!

- People claiming that the AI cheated because it attacked from far away, i.e. the AI couldn't have seen that the player defended his cities with only one spearman each, in good old Civ3 tradition. Actually the AI had just reacted to the power graph, which is visible to the player as well.

I'm well aware of how the power graph checks work. Are you prepared to defend a decision to pick shoddy target cities, declare across the world with a worst enemy likely to attack adjacent to you, and attack people who have 150% power? If not, then you have no room to claim people are complaining on this point due to "good" AI play, as the DoW mechanics have been explored very thoroughly. Even HERE, we have an example of poor AI play getting lucky, because as far as the AI is concerned, 1 spear per city or 150% of its power mean exactly the same thing.

In summary, AI considers empire with 0 defenders = empire with 50+ defenders. Is that a good AI to you?

- People claiming that the AI cheated because they didn't win a battle they expected to win. Actually the AI had just chosen appropriate promotions for the defending units.

I can punch holes through this argument top to bottom to top again. Frankly, this argument is terrible. The AI does not promote well, and almost certainly the actual complaint was that the human lost a battle on RNG outcomes, which has nothing to do with AI play quality whatsoever. You're not sneaking this one by me. I see these woodsman melee and combat archery defending cities and I cringe...and now you're telling me that an AI that chooses terrain specific promotions outside of that terrain is good? Maybe it got LUCKY and randomly picked decent promotions that time, and someone complained. That still isn't truly a complaint over "good" AI cheating.

- People claiming that the AI cheated because they ended up in a war against multiple AIs. Actually they either hadn't paid much attention to the state of their relations (underestimating the effect of religion was rather common in the early days), or several AIs went into dogpile war mode, and the player interpreted that as an anti-player-bias, not understanding that it happened to weak AI civs as well

AI dogpiles being fruitful or not is ALSO lucky. The AI will agree to these wars even if it is across the world and can't reach its target or doesn't have a means to capture a single city. At least the dogpile function is a LITTLE bit closer to decent play, unlike your other examples. I'd like to see a better basic algorithm for dogpile though, beyond just declaring then doing nothing most of the time.

I agree with you that Sid is going into the wrong direction by directing the development focus away from the AI, but I do not agree that this was simply a baseless excuse on his part.

It's an excuse, but my long-time most hated direction taken by failaxis is that they simply don't care/refuse to fix basic gameplay 101 issues, to the point of releasing new content or patching obscure things over making it so that the controls work.

"unselect unit"=select all units
"ranged attack"=move next to enemy melee unit so it can kill you on its turn
"select all units of a given type" = select all units
"open trade screen" = declare war (without prompt)

Just to name a few mechanical issues with the most recent 2 games in the series...which also process turns enormously slowly (doing ludicrous things like unit move animations in the fog) and have questionable multiplayer issues (net code is atrocious, how do they get by with that travesty in a release title? What are they, treyarch?!)

I'm not asking for a perfect AI. I'm asking for an AI that does the following:

1. Knows the rules of the game
2. Actually plays the game (IE attempts to reach a victory condition)
3. Uses its resources in a way that, on average, attains #2 (no iron works in cities with low hammers, no oxford in the 10th best research city, etc).
4. Can handle each normal situation possible in the base game (including intercontinental attacks. Dropping 5 catapults and 1 melee unit isn't something it does by land. Why does it do it by sea?)

In such a situation, the hardcore fans _could_ become a factor in the equation if they pushed for a good AI. Unfortunately, hardcore fans who gripe about the existing weaknesses in an already quite good AI without offering workable solutions how to match these very high expectations, don't help that cause.

I'm afraid I'm not at liberty to discuss what goes on with "the cause", but I know a thing or two about it. I was one of those "hardcore players" pulled to help out, you see.

What would you do, where would you put your money?

1. Gameplay rules: Before graphics, sound, AI, or anything else. Get the skeleton of a game where the options are strategically viable and the game has depth.
2. User interface/controls: Make the game seamlessly playable, to both rookies and veterans. None of this crap where units suicide before the player can give input or having units do an action different from what the player ordered. Failaxis has been a disgrace in this regard for 2 games straight, despite that it is gameplay #1 and a top priority in ANY game.
3. Balance: Is one unit class being used exclusively for example? Do games always end in era x? Teams of players should be working out these nuances, and then...
4. AI: At this point you have a chance to give the AI a core as well as have it execute some scripts seen by competitive players within the testing system. Once the game has a acceptable balance it should be possible to make a reasonable AI, IE one that plays to win within the rules...and if the balance is there you can make different AI personalities tend to attempt to win in differing ways.
5. Running the game: Streamline this. Stop making people spend 2 hours every game (civ V) waiting for every 250 turns played. It's VERY telling that playing civ V in strategic view decreases AI turn times 5-6x. Ain't runnin' no complex calcs in that extra time, that's for sure.
6. Graphics + Sound: The bells and whistles come last. They are important, sure, but they aren't what makes a game "stand the test of time" and become truly great. How much money went into the opening movie of civ V alone, while multiplayer allowed double moving and mass drops when 4+ people were in a game over a year after release?

To pull this off, you need a tremendously good project manager and enough checks set into the process to stand a chance. You also need somebody better than idiot 2k CEO "xcom should be a shooter".

It's hard to do, but SOME companies still put out good titles.

 

[Steve the Noble]

1. Gameplay rules

...

6. Graphics + Sound: The bells and whistles come last.

For developing a long-lasting strategy game, that order sounds ideal... but between game rules and graphics + sound, you know which one is going to get more oohs-and-aahs at game conference demos...

 

[Psyringe]

An AI that doesn't try to win is somehow GOOD? No. I won't buy it. I don't expect an AI to play 100% perfectly (as you said, that's impossible), but this is ridiculous. Most AI know how to play their game and try to win. Late 90's AIs played their game better than civ IV AI in a LOT of respects.

Name three (of a complexity comparable to Civ4), please.

To my knowledge, the most popular non-Civ TBS games of the 90s are:

Alpha Centauri: The AI doesn't even understand half of the game and simply ignores lots of things. It never builds boreholes, has no grasp of the importance of satellites, ignores lots of very obvious unit builds, has no idea how to use crawlers ... etc. etc. SMAC is a great game, but its AI is definitely one of its weakest points. It doesn't pursue a specific victory goal and it has huge difficulties to adapt to mods.

Master of Orion II: The AI doesn't understand several technologies, especially those that offer non-standard ways of combat. Research decisions are horrible. The AI is incapable if ferrying population around in a manner that would make its empire efficient. Combat strategies are extremely simple. Again, it's a great game, but the AI is weak. It doesn't pursue a specific victory goal and it has huge difficulties to adapt to mods.

Master of Orion 1: The AI is rudimentary at best, it basically gets enormous bonuses, makes purely flavor-based decisions, and gives the illusion of an actual AI. Combat algorithms are very simple and easily exploited. Great game, weak AI, doesn't pursue a specific victory goal, incapable of adapting to mods. We see a pattern here.

Master of Magic: The AI is essentially broken. It gets absolutely insane bonuses to make up for its incompetence, but still fails. It has enormous difficulties to leave its home island, doesn't coordinate its research efforts at all, fails to understand lots of important spells, uses very simple combat algorithms, neglects lots of very obvious spell combos, it can hardly play the game at all. Same pattern as above.

Heroes of Might and Magic III: The AI follows some rather simple rules, which - since the game is rather simple as well - are quite efficient. Of all the games mentioned so far, HoMM3's AI definitely has the best grasp of its own game. However, that's not because the AI is so great, but because the game _is_ much easier to grasp. Whether or not it's "trying to win" is arguable - it basically attacks enemy cities and armies one after the other and since that is the only victory condition, it will do so sooner or later if not stopped. It doesn't follow any actual plan though, and like pretty much all AIs of popular 90s TBS games, it makes its moves purely situational.

The Civ4 AI is a massive step forward over the AIs mentioned above. It understands its own game better, plays more competently, adapts better to mods, and cheats way less. It still (mostly) doesn't pursue a specific victory condition, but that's really something that the AIs mentioned above couldn't even dream of. Actually, the 90s are exactly the decade in which the complexity of the games made huge leaps forward while the AIs just couldn't follow. This led then, in the next decade, to some games with more focus on AI development (Civ4, GalCiv). This trend, however, has ended some years ago.

I'm well aware of how the power graph checks work. Are you prepared to defend a decision to pick shoddy target cities, declare across the world with a worst enemy likely to attack adjacent to you, and attack people who have 150% power?

You're using a special case to try to refute a point that I wasn't even making. My point was that in the cases where the AI did something well, and did not cheat, there were still people complaining about a cheating AI. Whether or not the AI always makes good or stupid DoW decisions is of no consequence to that point.

The AI does not promote well, and almost certainly the actual complaint was that the human lost a battle on RNG outcomes, which has nothing to do with AI play quality whatsoever.

Well, re-read the threads. As I remember, the vast majority of complaints came from people who didn't understand the promotion system and who (for example) had attacked cities with fortified defenders who had promotions that helped them further. Choosing helpful promotions for defenders is a sign of a decent AI. Whether or not there are _even better_ promotions that it could have chosen is not relevant here - because, again, you're trying to refute a point I didn't even make. See above.

AI dogpiles being fruitful or not is ALSO lucky. The AI will agree to these wars even if it is across the world and can't reach its target or doesn't have a means to capture a single city. At least the dogpile function is a LITTLE bit closer to decent play, unlike your other examples. I'd like to see a better basic algorithm for dogpile though, beyond just declaring then doing nothing most of the time.

And again, you're doing the same thing as above. Yes, obviously the dogpile algorithm could be better. And you focus on that facet of the issue and let it fade out everything else. But my point here was that players complained about being cheated when they simply weren't. That is independent of how improvable the Civ4 algorithm is.

I'm not asking for a perfect AI. I'm asking for an AI that does the following:

1. Knows the rules of the game
2. Actually plays the game (IE attempts to reach a victory condition)
3. Uses its resources in a way that, on average, attains #2 (no iron works in cities with low hammers, no oxford in the 10th best research city, etc).
4. Can handle each normal situation possible in the base game (including intercontinental attacks. Dropping 5 catapults and 1 melee unit isn't something it does by land. Why does it do it by sea?)

Well, that's what I meant. You're asking for something that is currently unattainable in the context of complex TBS games (well, unless you build the whole game around the AI, which GalCiv tried with quite some success, but that comes at a cost as the game will then feel bland to many players). If you want a game of the complexity of Civ4, then it's simply not possible to adhere to all 4 of your expectations.

However, that doesn't mean that there's no room for improvement in good contemporary AIs. On the contrary. There are lots of steps that could be taken to make AIs better (giving each AI its own thread to do calculations parallel to the player's turn would be a good start). However, the path to achieve this is, imho, not to diss the existing good AIs by focusing on their flaws and ridiculing the name of the developing studio, but to make a balanced assessment of what worked well and what's improvable, and to come up with workable suggestions on how to improve the latter. And that's something that I don't see in your post(s), which makes me a bit sad.

 

[Archon_Wing]

Hmm, after dumping a lot of time into trying to edit Starcraft I's AI; and having a lot of people play it, I think I have an idea on what people find enjoyable/challenging.

-- Most people wouldn't mind higher levels giving the AI bonuses, but this kind of bonus whoring must not look that obvious to the player. Example: The city expands to 10 cities by 1500 BC or rushes with bronze units @ 2500 BC; in these cases it feels like the AI is playing an entirely different game. A AI with bonuses should ideally always feel a little ahead of the player and not deliver killshots at impossible timings.

-- The AI must practice the basics. Basic defense, expansion, economy, and army. Obviously it won't be the most efficiently done, but I better not see trash like one AI expanding to 4 cities and never teching.

-- Flavor is fine, but being too extreme creates joke AIs that ruin the game by softplaying. Reluctant to tech trade civs can remain that way, but in Civ IV it's comical how the AIs cripple themselves by trading so little. It's stupid.

The best way to promote flavor is to use a strong base build, and use a dice roll to pick several branches that have a chance of being viable and will see occasional use. But the foundation must be there to begin with. Civ IV AIs simply does not have this.

-- More variety. The AI is disturbingly uniform when it comes to tech choices with a few variations. This means that certain tech paths just completely blow because everyone runs towards it? Why?

-- It must try to win. I understand some want to be more peaceful, but there must be higher chance to build military and go to war for some of the peaceniks. As it stands, the peaceniks are utterly worthless and often become nonfactors. This doesn't mean it should refuse to vote for you in the UN regardless of stance, but come on! Someone that hates you and sees you're about to launch a spaceship should indeed have a reason to attack you.

Having AIs that implode, or just provide extremely easy picking, or even worse peacevassal whores is really not challenging or fun. It's just annoying.

 

[Utenlok]

Archon was pretty much money on his post.

There are many AI that you know have zero chance to win or even be a factor. Toku comes to mind. Everyone knows Izzy will effect who likes who with her religious work, but will be a total pushover to capture when the player decides to do so. I have been right next to Gandhi's capital on a rush and he doesnt even whip an archer from his 5 pop city and just dies without a struggle. We all know Monty will build a ton of horse archers (why dont aggressive's favor ground units that get the free promotion) and show up to fight your rifles with them.

 

[Knightly_]

Hey,
first of all i absolutely agree that there are loser AIs in civ4 (as Archon mentioned) and situations like stop expanding after 4 cities even if enough land are ridiculous. But when it comes to victory conditions and how to finally "win" a game i think Firaxis has done a decent job with the AIs. Yes they play to lose (i think thats even an official statement from Soren) and if they go for victory they perform bad. I see your point in "AI gets less bonus but tries to pursue a VC more actively" but the problem behind this are the VCs itself. Multiple VCs add variety for the player and i agree that Civ4 VCs are a matter of debate on its own but given these VCs i dont think AIs using efficient strategies would be much fun. Saving up great artist for an 8-bomb kind of instant win? Being able to actively bribe other AIs into different religions for Diplo win? Instant DoW against the next AI after a capitulation when going for Domination? I doubt that this would be fun to the player. An even greater problem is when an AI cares to win, it has to be aware of the other AIs/player victory state. While this would surely avoid the utterly broken culture-AI vassal to strong-military-AI it would end in nothing but dows in the end of the game. So Firaxis made the AIs only go for culture/space which are the two victory conditions a player can foresee and gives it into the hand of the player to prevent them from winning. Imho a good decision.

So while i generally agree that a good AI should play to win the game the civ4 VCs are not suited for this because a good AI would not be a fun one.

Knightly

 

[strijder20]

Hey,
first of all i absolutely agree that there are loser AIs in civ4 (as Archon mentioned) and situations like stop expanding after 4 cities even if enough land are ridiculous. But when it comes to victory conditions and how to finally "win" a game i think Firaxis has done a decent job with the AIs. Yes they play to lose (i think thats even an official statement from Soren) and if they go for victory they perform bad. I see your point in "AI gets less bonus but tries to pursue a VC more actively" but the problem behind this are the VCs itself. Multiple VCs add variety for the player and i agree that Civ4 VCs are a matter of debate on its own but given these VCs i dont think AIs using efficient strategies would be much fun. Saving up great artist for an 8-bomb kind of instant win? Being able to actively bribe other AIs into different religions for Diplo win? Instant DoW against the next AI after a capitulation when going for Domination? I doubt that this would be fun to the player. An even greater problem is when an AI cares to win, it has to be aware of the other AIs/player victory state. While this would surely avoid the utterly broken culture-AI vassal to strong-military-AI it would end in nothing but dows in the end of the game. So Firaxis made the AIs only go for culture/space which are the two victory conditions a player can foresee and gives it into the hand of the player to prevent them from winning. Imho a good decision.

So while i generally agree that a good AI should play to win the game the civ4 VCs are not suited for this because a good AI would not be a fun one.

Knightly

So you basically say MP is not fun. As good MP players would use all those strategies.

 

[Knightly_]

I have to admit beside from team games against AI teams i have never played a MP only game vs other players. But from what i have read the war, or say domination route is the only viable one. Correct me if im wrong.

Knightly

 

[strijder20]

I have to admit beside from team games against AI teams i have never played a MP only game vs other players. But from what i have read the war, or say domination route is the only viable one. Correct me if im wrong.

Knightly

I have not tried MP a lot either - but coincidentally, I just played one. 3 vs 3 teamer. I was playing as the Dutch, teamed up with Portugal and Zululand. The opponents were the Dutch, Persia and Greece.
I was alone on an island with Persia and Greece. You can guess what happened. I got swamped by troops, but managed to held out to feudalism - and then got very lucky and killled most of their stack with two longbows - on defence, of course. The other team gave up after this.
But I have also played an island game once, where you can be in relative peace until 1200 AD. Hannibal - a player, not an AI- spammed privateers at me, but I got to frigates and later ironclads, so it was not that much of a problem.
The amount of war largely depends on the mapscript, as you see. Islands provides peace until the 1000 ADs (although beware if the Portuguese are in game), while Pangaea means pretty much war all the way to victory, or defeat.
There are some games where you can go for another victory condition, but those are rare. Very rare. It's better to divert all your resources in expansion/ the annihilation of your opponent. Of course, always peace games are an exception
You can also try teaming against the Ais, as you did.

 

[Knightly_]

So in fact there is only one way to victory in MP. Possibly thats a reason why so many people dont play multiplayer. I suspect its kind of an entirely different game anyway...

 

[TheMeInTeam]

Heroes of Might and Magic III: The AI follows some rather simple rules, which - since the game is rather simple as well - are quite efficient.

You think so? Every movement point, every creature move decision, every gold piece allocated matters in that game. If you don't think that game is deep, you don't know how to play it. I've LP'd it for hours upon hours upon hours, and I'm still terrible at it compared to good MP players...but against anyone who hasn't played it competitively I'd spank them so hard they couldn't see straight. The game has way more depth than you're giving it credit, and YES, comparable complexity to the civilization franchise. Although rarely is there more than 1 or at most 2 VC, you're still looking at 10+ meaningful decisions to be made per turn, often double that. Can you claim civ IV has significantly more than that?

Warlords III: Warlords IV kind of fell apart, but this was the last great game in the series, and man was it spectacular. Lots of unit types, empire setup possibilities, hero and spell choices, order of conquest, quest decisions, etc. Probably most here won't have knowledge of it, but while the strategic studies group was still around, they knew how to make games. This one was amazing and while the AI (like in HOMM III) was lacking compared to top notch players, it knew the rules and took the basic winning approach well. Once again, a player could spend 50 hours on this game or more and still be pretty bad at it, so it definitely had its depth.

Unfortunately, those are the only 2 I can say for certain beat civ IV AI in a lot of respects (which is what I initially claimed). Other TBS probably have a few more:

http://en.wikipedia.org/wiki/Category:Turn-based_strategy_video_games

Most notably, I played the first deadlock and while rough around the edges it was a fairly deep game with a tech tree and similar econ vs military balance (even shared the = won't work concept ). Once again, not like the AI was ever great there but it tried to win. If later iterations of that series were any good it would be competitive with civ also.

You're using a special case to try to refute a point that I wasn't even making. My point was that in the cases where the AI did something well, and did not cheat, there were still people complaining about a cheating AI. Whether or not the AI always makes good or stupid DoW decisions is of no consequence to that point.

Your point was destroyed, you just don't want to acknowledge it. My claim is that the AI "did something well" not because it's actually good AI, but because the RNG gods gave it a LUCKY roll that happened to make it do something well this time. You then use people's complaints about it as a basis for skimping on "good" AI, when the reality is strictly that "good" AI in this aspect never existed, only a random + stupid AI + a bad human player.

Well, re-read the threads. As I remember, the vast majority of complaints came from people who didn't understand the promotion system and who (for example) had attacked cities with fortified defenders who had promotions that helped them further. Choosing helpful promotions for defenders is a sign of a decent AI. Whether or not there are _even better_ promotions that it could have chosen is not relevant here - because, again, you're trying to refute a point I didn't even make. See above.

:wallbang:. Seriously? You're telling me the devs listen to complaints by terrible players and use it as an excuse for bad AI. You're doing it right in this paragraph. Choosing promotions at RANDOM does NOT constitute a good AI! The entire basis of this point is an awful player. The only thing that proves is that the AI at that time was better than the player at that time. Considering that the player is throwing units into fortified cities and not taking it, we have a clear picture here; awful play < bad AI... not good AI > sensible player. In other words, this complaints core origin is not the AI but the crummy player and isn't a valid example at all.

And again, you're doing the same thing as above. Yes, obviously the dogpile algorithm could be better. And you focus on that facet of the issue and let it fade out everything else. But my point here was that players complained about being cheated when they simply weren't. That is independent of how improvable the Civ4 algorithm is.

This misses the point. The point is that it is also independent of the quality of the AI. Nothing is better evidence of that point than the fact that complaints over RNG battle outcomes STILL exist.

Can you claim that "functional and fair RNG" = "good AI"? Not sensibly. However that is essentially the same as making any of your above claims as examples of people complaining about "GOOD AI" cheating. The AI wasn't good there at all, with fairly objective evidence provided. The source of complaints were people not understanding the game, or griping about bad RNG outcomes doing preventable damage and therefore losing to a poor AI.

Well, that's what I meant. You're asking for something that is currently unattainable in the context of complex TBS games (well, unless you build the whole game around the AI, which GalCiv tried with quite some success, but that comes at a cost as the game will then feel bland to many players). If you want a game of the complexity of Civ4, then it's simply not possible to adhere to all 4 of your expectations.

I refuse to believe this, and have reason to believe it isn't true also. Just look at the work done by a small, unpaid team over on better BTS AI mod. BBAI has algorithms for attempting every victory condition (aside from time I think), is better at pillaging/choking, picks better national wonder sites, and will raze cities it knows it can't hold (and it has a basis to evaluate this!). Jdog even put a lot of effort into tech selection. In other words, one guy completely obliterated all of the work done by the entirety of the civ IV staff that was done from vanilla ----> warlords ----> bts. Jdog did have some help, but I'll go out on a limb and say his contributions alone did more for his mod AI than the entirety of firaxian patches/expansions.

No, the current state of the AI in both games was a choice. It's a shame that the best TBS publishers from days gone by aren't around any more. Hopefully HOMM VI can capture III's magic and destroy the balance of TBS.

I have to admit beside from team games against AI teams i have never played a MP only game vs other players. But from what i have read the war, or say domination route is the only viable one. Correct me if im wrong.

Okay, read some of the realms beyond PBEM outcomes. I'm correcting you . Pitboss #1 there was also decided by space.

However, the complaints over MP aren't justified. That's not an AI failing, it's a DESIGN failing. If you want a BALANCED game, design it so that all of the VC are valid, even in MP!

Failaxis has never done this in any of its civ games, and so military continues to be the dominant force in ever game in the franchise. That's why you balance it with people as strategists first. Blizzard does it that way and there is a reason that game is VERY close to balanced, despite extremely different races. Civ couldn't do that with 18+ civs, but it could do it between VC and between stock units. It doesn't though.

 

[Kid R]

Seems to me from reading these boards over the years that the game mechanics of Civ 4 were sufficiently complex that winning strategies only emerged slowly, by player study and experimentation during the months and even years after release.

The devs can hardly have anticipated and coded all the good strategies, or foreseen and countered all the kind of weird player strategies which would get adopted as "standard play" by many. It's only to be expected various flaws and loopholes would be exposed by players after many times the study time that the original programmers had. And to the AI's credit it at least continues to play a decent "spoiler" game against whatever we throw at it.

Modders and gameplay strategy innovators all do a great job, but we should not be immodest and forget we all stand on the shoulders of others

 

[Knightly_]

However, the complaints over MP aren't justified. That's not an AI failing, it's a DESIGN failing. If you want a BALANCED game, design it so that all of the VC are valid, even in MP!

I suppose balancing such orthogonal VCs is not a trivial task. So in the end its just a matter of RoI and because civ4 is designed for single player they approached it the other way, that is create multiple VCs for the joy of the player but not necessarily balance them. From that perspective they made the AI decent in my opinion. I have no argument why your approach should fail. Although its questionable if you can create a powerful enough AI play the game after you have balanced it with humans i think it would work.

Another thing: I am very disappointed about the trend in the gaming community to action-up lots of tactical/strategy stuff. Its not only the xcom you mentioned, but also for example jagged alliance. Another game i only recently played for the first time was mass effect. Yes the story is nice (as you would expect from bioware) and the style of the game encourages one to feel more like a hero but all battles felt so incredible dumbed down and hollow to me... i was stunned. Reading your post im actually motivated in taking a look into warlords as i have never played it.

Knightly_

 

[Archon_Wing]

Failaxis has never done this in any of its civ games, and so military continues to be the dominant force in ever game in the franchise. That's why you balance it with people as strategists first. Blizzard does it that way and there is a reason that game is VERY close to balanced, despite extremely different races. Civ couldn't do that with 18+ civs, but it could do it between VC and between stock units. It doesn't though.

Failaxis lol. But yea, their ability to balance stuff is pretty sad; I mean a quick look at the UU/UB/trait combos makes it pretty laughable. Starcraft and Starcraft 2 have their own set of issues but the fact it could be balanced with 3 factions with very different units and mechanics, while we see stuff like the batista elephants, space elevators, lolchas, and the Holy Failure Empire it's a bit frustrating.

 

[karadoc]

The AI does get to see everything in the squares it can move to, right? Which means for late game naval units, they have a sight range of about 5+?

They don't get to see everything that they could move to. But there are a handful of cases where they get to see enemy units which are just outside their vision. The can see enemy units that are 2 steps away, regardless of anything blocking the vision. Naval units cannot see 5+ tiles, but they can see any enemy units that are in attack range.

The thing about the AI is that it is programmed in such a way that if the programmer makes a mistake, then the AI can cheat... so there are probably a bunch of cases where the AI cheats when it isn't meant to. For example, those blue circle recommendation for city locations - they can't see resources without the appropriate tech, but they can see resources that are still covered by the fog of war. The original code basically just said "if the resource is in range, count its value", when what it should have said is "if the resource is in range, and we can see it get its value". This one of things that I recently corrected in my mod (so that now, the city recommendations for the AI and for the human players do not take into account any unrevealed tiles.)

There are also some cases where the AI knows where your cities are even if it hasn't seen them... but generally these kinds of things are programming mistakes rather than deliberate attempts to give the AI an unfair advantage.

 

[Mylene]

I couldn't resist to comment on the Homm3 AI statement
Easily one of the most horrible AIs ever programmed.

They will sit still with their heroes when you are near, and have their army split up.
They will waste all spell points on fights like 3 Angels vs. 50 gnolls.
It doesn't understand the concept of making 1 hero strong, so it will always fail vs. an experienced human.
And the "Impossible" setting is a walk in the park, i had to ally up 7 AIs and still beat them easily.

 

[reddishrecue]

Knowing how the AI reacts can be something helpful for this question. An issue that the AI really does is build in undesirable places such as tundras. This is one way the AI plays. More noticeable AI habits in resource gathering is offering to trade a good (such as spice) for iron when the AI lacks for iron. No, unless the AI got iron working.

 

[MantaRevan]

edit: Oracling IW is a pretty bad idea

Yeah. On normal speed Iron Working isn't that advanced. If you really want to, it's pretty easy to beeline.



---
I am here: http://maps.google.com/maps?ll=38.253384,-85.481592