AI doesn't try hard enough to win

[TheWilltoAct]

You seem to really like playing to win though, TheMeInTeam.

I'm not being sarcastic, I'm just suggesting that the other side of it is that some of us enjoy experiencing a little mystery, even if it might mean losing the game.

 

[TheMeInTeam]

It's true I like playing to win.

Also, mystery (perhaps unpredictability) is good. For example, gears of war II bots will target different locations to move randomly and it makes them slightly harder to deal with. Gears II bots do not randomly explode on death (without grenades), killing everyone in a 10 foot radius with such info never mentioned anywhere before it happens to you. That is the nature of what we see in civ.

As a role-play element was introduced into civ, so too were rules. Civ is one of the very few games I can think of where the RULES OF THE GAME THEMSELVES are hidden from the player. What disgusts me is this, not that the AI is unpredictable, or that it doesn't play optimally (even I don't play optimally), just that we are given a set of rules we are knowingly told, and that those rules are not only incomplete, but outright misleading.

1. Vassals average your diplo. Hidden rule. Misleading GUI.
2. The tips screen suggests that barbarians can spawn in the fog. Reality is that they can spawn only in areas with fog that are >2 tiles away from a unit. Again we're given WRONG info on the game rules. It took more than a year for most players to realize this.
3. Displayed diplo in the GUI is not accurate any time there are hidden modifiers, which is usually. More gameplay-significant rules, once again hidden.
4. Overflow hammers have been botched so badly that the last 3 versions of the game between official and unofficial each handle them differently! Once again we're not exactly given the information on what's actually happening.
5. There is a LOT of information that can be inferred with difficulty over time, that can be answered almost instantly by glancing at the code. This is why the code was provided possibly, but can a game be made any LESS beginner-friendly? It's like you have to have an elitist mindset just to have a chance of becoming good at this game. Fortunately for me, I do. I was willing to learn how to look into the XML for answers and memorize what AIs do what at each disposition...but note that the game's highest difficulties essentially FORCE you to learn some of that one way or another. I remind you of a thread started that complained that for the vast majority of players, difficulties above the MIDDLE in this game would be impossible. I don't share his weak opinion, but blocking the rules from players contributes to the hole we have at IMM+ for certain.

On top of all that, some of the hotkey controls don't work properly , but that's for another thread.

Anyway, it's one thing to lose the game due to unpredictable behavior. It is entirely another to lose because the GUI mislead you (I have used and seen others use the term lying, and that is not too strong actually), or you otherwise had rules of the game hidden from you.

People like to, for some reason, compare this game with chess on occasion. A good analogy to some of the crap fireaxis pulls in this game would be to say "a rook can jump over any other piece similar to a knight to make a capture, but only on every 3rd move" and "on every 5th move a knight has a 3 length 1 width L move instead of the normal 2-1". Oh, but, only your opponent has this information available. Special rules. That you're not told before the game, or even until you lose your queen to said rook or knight.

And then some ignorant quasi-elitist (aka a player much worse than you who plays against weak opposition) tells you that if you were good, you could overcome an equal-skill player taking your queen by surprise on turn 5. After all, he can, because he plays against 8 year olds who just learned the rules.

That is very much like what civ does to its rookies. HI! HIDDEN RULES! YOU LOST! Keep this in mind, each time someone loses to the vedic aryans (exception to the rules of barbs entering borders by turn x), a surprise UN vote (thanks to hidden modifiers, those civs actually liked your opponent more), a surprise DoW (so what if he had max diplo with you, that crappy vassal you helped him kill makes him secretly hate you now). Every time I see a rookie complain about these things on the forum, I'm reminded that though Civ IV is a great game, it has some glaring holes that are for some reason accepted as fine.

Hiding game rules isn't kosher. It represents a weakness in programming that is covered up so that only those willing to code dive of spam games have any chance of detecting them. If this happened in starcraft, heroes of might and magic, garbage like world of warcraft, or even pokemon, complaints would be somewhat profound.

 

[TheWilltoAct]

...but note that the game's highest difficulties essentially FORCE you to learn some of that one way or another. I remind you of a thread started that complained that for the vast majority of players, difficulties above the MIDDLE in this game would be impossible. I don't share his weak opinion, but blocking the rules from players contributes to the hole we have at IMM+ for certain.

This I can definitely believe, especially the part about needing to know every detail of the game in order to succeed on Immortal and Deity. I think the other mammoth task is actually developing an overall strategy, understanding where to place your cities and why is huge and can be rather overwhelming when you find yourself with a lot of empty territory that you must fill (hurdle I'm getting over atm ^.^).

Anyway I understand that this is something you feel passionately about. Perhaps as I improve at the game and start running into more and more of these roadblocks due to hidden rules I will come to better understand your point of view. Cheers!

 

[Silu]

1. Vassals average your diplo. Hidden rule. Misleading GUI.
2. The tips screen suggests that barbarians can spawn in the fog. Reality is that they can spawn only in areas with fog that are >2 tiles away from a unit. Again we're given WRONG info on the game rules. It took more than a year for most players to realize this.
3. Displayed diplo in the GUI is not accurate any time there are hidden modifiers, which is usually. More gameplay-significant rules, once again hidden.
4. Overflow hammers have been botched so badly that the last 3 versions of the game between official and unofficial each handle them differently! Once again we're not exactly given the information on what's actually happening.
5. There is a LOT of information that can be inferred with difficulty over time, that can be answered almost instantly by glancing at the code. This is why the code was provided possibly, but can a game be made any LESS beginner-friendly? It's like you have to have an elitist mindset just to have a chance of becoming good at this game. Fortunately for me, I do. I was willing to learn how to look into the XML for answers and memorize what AIs do what at each disposition...but note that the game's highest difficulties essentially FORCE you to learn some of that one way or another. I remind you of a thread started that complained that for the vast majority of players, difficulties above the MIDDLE in this game would be impossible. I don't share his weak opinion, but blocking the rules from players contributes to the hole we have at IMM+ for certain.

Much anger in you I sense

Am I the only one who doesn't think that list doesn't have anything important? I think diplo shouldn't be 100% deductible out of the box, because inherently, it isn't in any real context either (and I think the game manual doesn't say "+8 for them to vote for u" either). 2 and 4 can be simple overlooks/bugs without deliberate misleading. Pretty sure the dude writing the tips thought the spawnbustage thing was true.

The code certainly wasn't provided to "make up for bad documentation", since such a small percentage of players would use it for that. It was released for modding, and that's a huge success story of its own.

Anyway, it's one thing to lose the game due to unpredictable behavior. It is entirely another to lose because the GUI mislead you (I have used and seen others use the term lying, and that is not too strong actually), or you otherwise had rules of the game hidden from you.

Hiding game rules isn't kosher. It represents a weakness in programming that is covered up so that only those willing to code dive of spam games have any chance of detecting them. If this happened in starcraft, heroes of might and magic, garbage like world of warcraft, or even pokemon, complaints would be somewhat profound.

Stuff like this is present in every game. There're a million exploits or "hidden rules" in SC that took years to find and longer to become public knowledge and people have to either be forcibly enforced to not use those in any competitive context, which takes a big toll on organization or then new players just have to learn them on their own (similar to these Civ things complained about). Examples include muta stacking, return-minerals-antiblock, air units on top of turrets, etc etc etc...

For single player, in WC3 the AI could for years cast spells anywhere in their raw visibility radius distance - even up ledges covered in fog. WoW has been stricken with exploits at every turn, they just get patched fast because of the massive amounts of active manpower on that game. And these are Blizzard games, legendary for being as polished as they get.

My point is, a game as complex as Civ, and a team working on it, some things are bound to fall through the cracks. I'm pretty damn sure none of it is deliberate "misleading", just slight bugs, overlooks and possibly design decisions (diplo). At least the tools to know the real mechanics exist, unlike in 99% of games.

 

[PPQ_Purple]

I agree with TheMeInTeam.
The game should have provided all the rulles and data for us.

I don't say that it should tell us at what score the AI will or will not vote for us, or accept a trade but I do want to see all the things that effect that score. Hidden diplomatic modifiers are on the top of my hate list atm.

 

[TheLazyHase]

Stuff like this is present in every game. There're a million exploits or "hidden rules" in SC that took years to find and longer to become public knowledge and people have to either be forcibly enforced to not use those in any competitive context, which takes a big toll on organization or then new players just have to learn them on their own (similar to these Civ things complained about). Examples include muta stacking, return-minerals-antiblock, air units on top of turrets, etc etc etc...

For single player, in WC3 the AI could for years cast spells anywhere in their raw visibility radius distance - even up ledges covered in fog. WoW has been stricken with exploits at every turn, they just get patched fast because of the massive amounts of active manpower on that game. And these are Blizzard games, legendary for being as polished as they get.

My point is, a game as complex as Civ, and a team working on it, some things are bound to fall through the cracks. I'm pretty damn sure none of it is deliberate "misleading", just slight bugs, overlooks and possibly design decisions (diplo). At least the tools to know the real mechanics exist, unlike in 99% of games.

I totally agree. Civ IV disclose more thing and explain more of how its work than every other game, not the other way around like TMIT seem to believe.

 

[bestrfcplayer]

Here's what happend in a recent game. It was contentents, settler, quick, and duel. I forget who I was but the AI was Kubi. So, I hadn't even declared war on him AND he has a bigger empire than me. But just because I have way advanced tech lead on him doesn't change anything right? No! So I see Kubi becomming my vassel in like 40 AD! I can't belive how easy it was to win!

 

[davelisowski]

The space race definitely seems like the easiest way by far to win a typical game. I guess that's probably the reason why many people disable it.

Exactly. I can't remember the last time I built a spaceship part.

I never use it anymore. Now it's Conquest, Domination, or Cultural. Diplomatic is on for some games.

 

[Sincro]

Exactly. I can't remember the last time I built a spaceship part.

I never use it anymore. Now it's Conquest, Domination, or Cultural. Diplomatic is on for some games.

Um, doesn't turning off the space race remove the main chance the AI has of winning? While I'm not a great player, or even a good one (I cannot make the jump past Noble and it's pretty embarressing) I've played a lot of games, and never seen an AI victory aside from space or cultural. The AI seems to always actively shoot for the space race victory, and taking out that victory option seems awfully cheesy to me.

-Sinc

 

[TheMeInTeam]

Disabling space and culture prevent the only 2 real threats of AI victory...and note that culture is a faster VC than space.

 

[Iranon]

The designers did not try to make this a true competition; the AI plays intentionally poorly.

Just imagine what would happen if AIs deliberately avoided obsoleting their old units and made use of the insane upgrade discounts they get.
If you avoid Hunting and connecting iron, you can build both Macemen and Warriors. 15 hammers to build a warrior, 19 gold to upgrade for a Deity AI... and this does not take into account the direct bonuses.

I agree with deliberately using sub-par AIs and giving them handicaps... diplomatic victories would not happen otherwise, playing on a decent level would be impossible without finding and exploiting deficits of the AI and the game would be a lot less involving.
Some details could be handled better though, especially with regard to naval warfare.

 

[Mazzini]

The designers did not try to make this a true competition; the AI plays intentionally poorly.

Just imagine what would happen if AIs deliberately avoided obsoleting their old units and made use of the insane upgrade discounts they get.
If you avoid Hunting and connecting iron, you can build both Macemen and Warriors. 15 hammers to build a warrior, 19 gold to upgrade for a Deity AI... and this does not take into account the direct bonuses.

I agree with deliberately using sub-par AIs and giving them handicaps... diplomatic victories would not happen otherwise, playing on a decent level would be impossible without finding and exploiting deficits of the AI and the game would be a lot less involving.
Some details could be handled better though, especially with regard to naval warfare.

The reason the AI needs those cheap upgrade bonuses is BECAUSE it plays so poorly. If it was smart enough to take advantage of situations like that, would it really need those crazy cheats?

 

[PieceOfMind]

The reason the AI needs those cheap upgrade bonuses is BECAUSE it plays so poorly. If it was smart enough to take advantage of situations like that, would it really need those crazy cheats?

That's a good point actually. Some people have said in the past, including myself, that if the AI were truly competitive at high levels it would just rush the human player with its starting archers, but that isn't the point. The AI only gets those archers because it isn't able to take that strategy.

In other words, a lot of the handicaps the AI gets shouldn't be considered as rules or limitations the AI has to work with. Those handicaps could very well have to change if the AI is improved, and it means you shouldn't try to improve the AI specifically to take advantage of the handicaps. Thus, a Better AI should never be taught to abuse handicaps regarding cheap upgrades.

 

[Medi]

As I recall, didn't the Vanilla and Warlords AIs get absurd upgrade cost discounts, on the order of having to pay only 1/10th the cost? This is because the AI was really terrible at keeping an updated military, so it needed some help. Blake's BTS AI is better at that, so AIs have to upgrade at half price (boo hoo ).

In any case, I think people are being a bit pessimistic about human skills. If the AI got -16 "You're going to win a cultural victory, you jerk!", this wouldn't make cultural victory impossible: it would mean that a human trying to get a cultural victory would have to acquire buffer territory to protect her three important cities, tech at least to Assembly Line to assemble a capable defense, try to distract AIs by luring them into other wars, etcetera. And this is an extreme situation--a real "play-to-win" AI would much prefer to see his rivals stomping the human, while he pursues goals of his own.

 

[Badtz Maru]

I had a real interesting game the other night that highlights the difference between the AI and the player's willingness to win. I was playing as the Aztecs, and my next door neighbor, Gandhi, had been friendly thoughout the game (aside from an iron age war of aggression HE started). He had led me in score for a long time, but after some wars of conquest against some of my weaker neighbords I was pulling ahead of him. I decided to check the victory conditions screen to see how I was doing, and saw that Gandhi had 2 legendary cities and one that was 10 turns from being legendary.

Fortunately, I had a small invasion army (three transports with some highly promoted ex-macemen infantry with CRIII and a bunch of other promotions, some cannon, and a single destroyer) that I was planning on using on the Babylonians in the near future. It was 8 turns away from the closest coastal legendary city of Gandhi's. Gandhi was already at war with Napoleon, and I bribed Mehmed with some techs to also declare on Gandhi, and rushed around to the target city. I landed my cannons and they were wiped out that turn, but I was able to take the city (protected with riflemen and longbowmen) in two turns with amphibious attacks from my transports while Gandhi's frigates tried unsuccessfully to sink them. I averted loss and was able to win culturally myself 40 turns later.

Gandhi provided very weak protection for the city he needed to win the game, something I never do (when I get close my best defensive units start clustering in my legendary and near-legendary cities, and Gandhi was on roughly tech parity with me). After I razed one of his most ancient and largest cities, he gave me peace a few turns later when he could have seriously hurt me in a counterattack. I coasted to the cultural victory without anyone, even my enemies, trying to invade my land except for the Babylonians trying to take back cities that used to belong to them.

I'm glad I won, but I feel like I exploited the weakness of the AI to do so. On my way to raze his city, I was afraid he might have an adequate defense force (5 infantry probably would have stopped me, instead of the 2 rifles and 3 longbows he had), but I was pretty sure he would not as it was a core city far from his front.

 

[Antilogic]

If the AI played to win, it would be far less dynamic to counter it and the game far less practical. Let me give you an example of an almost-impossible-to-beat-on-high-levels AI I did in the past, using ONLY XML value changes:

1. It is creative and imperialistic.
2. It has a unitprob of 80, so it builds units twice as often as shaka
3. It is not willing to declare on a target unless it has over double its power
4. It will trade tech at any disposition
5. It will declare war at any disposition. In fact, the moment it has enough power it plans a war on that hapless target.
6. It is willing to trade resources at any disposition
7. Its favorite tile improvement is the cottage

A deity player tried a game with this AI at immortal and promptly got spanked (essentially, the way it was set up guaranteed the human was the target if on the same continent, and he ate a TON of units on that declaration...!). However, such an AI is a threat anywhere:

- It will tend to vassal every civ if on another continent, and actually has enough units in its naval stack to be a threat
- Since it trades MORE often than mansa musa, it isn't as far behind in tech as expected.
- If it's on your continent and the AIs are getting a lot of bonuses, you're probably toast. For one, it will never, ever leave you alone and the other AIs tend to like it.

There is a good reason fireaxis did not make an AI like that.

I'm wondering what an Agg/Cre version of your evil leader would do. Have you stuck with Cre/Imp solely for the expansion (more cities)? What about an Exp/Imp version like Joao--he always expands like crazy in my games.

I'm pretty sure that I could beat easily your AI*7 even on emperor ( maybe, with some luck on immortal ) by manipulation of the diplo, even if the civs were unbribable. In fact I would bet you could do the same I would only need to find a way of exploiting some initial border hates and play defense Greed is very exploitable, you know .....

Have you started a thread in the stories forum yet? It sounds like the gauntlet has been thrown down.

 

[Iranon]

@ Mazzine, PieceOfMind: It would be great if we could have an AI good enough to not need handicaps, but I don't see it. The opening alone would be daunting... do you really expect an AI to balance rush defense, rushing when there's an opportunity, expansion and tech without the handicaps against a semi-competent human?

Once we introduce handicaps, those are parts of the game. If I can make a chess game challenging by giving the weaker player rook odds, it's still a true if asymetrical contest.
This is far, far better than giving the weaker queen odds with the understanding that s/he'll try not to 'abuse' the advantage.

 

[Lone Wolf]

Actually, sometimes the nature of the game can make giving handicaps more or less harmful for the gameplay. It's no coincidence that the handicap system is far more used in Japanese Shogi then in Chess.

 

[PieceOfMind]

@ Mazzine, PieceOfMind: It would be great if we could have an AI good enough to not need handicaps, but I don't see it. The opening alone would be daunting... do you really expect an AI to balance rush defense, rushing when there's an opportunity, expansion and tech without the handicaps against a semi-competent human?

Once we introduce handicaps, those are parts of the game. If I can make a chess game challenging by giving the weaker player rook odds, it's still a true if asymetrical contest.
This is far, far better than giving the weaker queen odds with the understanding that s/he'll try not to 'abuse' the advantage.

The more you teach an AI to specifically take advantage of its handicaps, the more human players will get frustrated and accuse the AI of cheating. For example, the AI has a well known ability to spot units in the fog when it's considering units to attack. I'm sure everyone has noticed the AI's uncanny ability to seek you out from seemingly a large stretch across the map with its magic destroyer. I'm not suggesting this is somewhere where the developers deliberately made the AI take advantage of a handicap, but I don't think they fully considered the consequences when they established that rule for land units and generalised it to naval units. After all, for land units the difference in visual range and movement range is not large.

That said, there are some handicaps that are permanent enough that they should be taken advantage of. For example, AI starting with a worker at high levels. The thing is, you don't have to teach the AI the AI that at those levels it has to take advantage of its starting worker... Instead you just teach the AI that if it starts with a worker at any difficulty to use it more intelligently and perhaps not get another worker so soon. I would not call this coding to take advantage of a handicap.

In other words, I consider the free units and techs given to the AI to be more like rule changes than handicaps, and unlikely to need changing if AI is improved. Scalable handicaps like upgrade discounts, maintenance discounts etc. are all subject to revision after AI improvements are great enough. These are the main things I'm talking about in my previous post.

Here's a hypothetical... At most levels the AI, like the human, gets free wins vs. barbs (if I understand correctly... If the AI doesn't get these free wins let's suppose they do for the sake of the example). Should the AI be taught a tactic that it can build a nearly 10xp unit straight away by attacking a barb axeman with a warrior? It would be clever exploitation of the rules, and I've suggested attempting to exploit it at Prince and lower difficulties before for the human player (though it's difficult to pull off in practise).

 

[TheMeInTeam]

I'm wondering what an Agg/Cre version of your evil leader would do. Have you stuck with Cre/Imp solely for the expansion (more cities)? What about an Exp/Imp version like Joao--he always expands like crazy in my games.



Have you started a thread in the stories forum yet? It sounds like the gauntlet has been thrown down.

I think CRE is mandatory to put the AI at its potential. The way auto workers function now (the AI doesn't move them in place via planning ahead) gives CRE a HUGE lead on hooking up its best tiles ASAP...an AI w/o religion suffers its border pop for quite some time on occasion. Joao succeeds in spite of this simply by spamming cities.

Left between AGG, EXP, and IMP as the 2nd trait, I favored IMP because it does seem to speed AI expansion more than EXP by itself and it is the better trait for wars. I wanted the AI to grab as much land peacefully as it can ----> it needs a lot of units to declare and to do that it needs cities. IMP also makes a lot of sense for a spam AI, since the AI tends to settle great generals and thus it can get some good promoted troops soon anyway with all the war it does. While AGG is somewhat scary on massed melee or gunpowder, it doesn't help siege (which this AI is supposed to favor) or mounted, and IMP lowers the chance the AI builds units not helped by its traits.