Why is it programmed to win?

[No2AWing]

I have just left the game. I am so angry! I bought the game for fun - not to bang my head against a wall. I have been attacking a city for quite a few game years and I just cannot get the city. I get it down in strength to red then it gets it all back again! That, in my mind, is absolutely wrong! Especially as I am only playing in the second level! This is not the first time this has happened. We keep getting, "Please wait" It should read, "Wait while I get rid of all your units." When you can have a go with a unit it comes up with , "Unit out of turns" The game allows you one shot. So what is the point? It was not tested correctly? I have been playing Civ for 20 years and this is the worst one.

 

[PhilBowles]

I'm not sure what relation the mini-rant has to the title question, but the question itself is puzzling. Why is a chess program programmed to try and beat the player? Why is a Total War or Starcraft AI programmed to try and win the battles?

In terms of responding to the rant, there are specific things to bear in mind that, once you get a handle on them, prevent early aggression and make it easy to avoid losing units. Make sure to build or buy a Warrior and/or Archer within the first three or four things you build; you generally want to expand later than in earlier Civ games. Units need to be coordinated, and it's important to consider different unit types - such as ranged together with melee - rather than just stacking lots of the highest-tech units you have access to.

 

[thadian]

The job of the dungeon master is to design dangerous encounters that CAN kill the player, but it's not much fun if the Arquebus-wielding kobold scores 3 critical hits and kills the player. As a DM and having designed every form of encounter you could imagine and far far more -

the AI should not play to win, it should play for entertainment. Each AI should have "it's goals" and some of those goals should overlap one another at different moments. Conflicts should be dangerous - and winnable at the same time. Reminiscent of RFC UHV's.

Nobody at the table will go home having had a good night if everyones character dies because they all had a critical failure on a jump check to jump across a spiked pit! The pit isn't there to kill the player, it's there to provide danger - Danger, which through attrition, can indeed cause a player to die - and it does happen now and then. More common, the player is "injured" and has penalties to various actions associated with the injury and when the player comes out alive, they are always thrilled.

A game should make you thrilled - not mad. The AI should play to entertain and cater to the player. Based on the players actions, the AI should try to "intervene" to prevent success, in a way that the player can respond to. And the AI should be able to win through attrition if the player can't.

A higher level AI should not be smarter or have more buffs - it should have more "goals" in the game. The player, in an RPG never has to worry your AI ally will take all your parties gold and run in the middle of a resting session - why should i fear the same of my allies in civ?

Just saying that players need certain things "they can count on" and the AI needs to consider, based on what kind of challenge is appropriate, how to behave in a way that the player needs to "achieve a task".

Civilization has always been a sandbox, but it needs a "direction" of sort -Sandboxes are good, they are great but they need to ENGAGE the player differently than the current AI does.

 

[Andulias]

The job of the dungeon master is to design dangerous encounters that CAN kill the player, but it's not much fun if the Arquebus-wielding kobold scores 3 critical hits and kills the player. As a DM and having designed every form of encounter you could imagine and far far more -

the AI should not play to win, it should play for entertainment. Each AI should have "it's goals" and some of those goals should overlap one another at different moments. Conflicts should be dangerous - and winnable at the same time. Reminiscent of RFC UHV's.

Nobody at the table will go home having had a good night if everyones character dies because they all had a critical failure on a jump check to jump across a spiked pit! The pit isn't there to kill the player, it's there to provide danger - Danger, which through attrition, can indeed cause a player to die - and it does happen now and then. More common, the player is "injured" and has penalties to various actions associated with the injury and when the player comes out alive, they are always thrilled.

A game should make you thrilled - not mad. The AI should play to entertain and cater to the player. Based on the players actions, the AI should try to "intervene" to prevent success, in a way that the player can respond to. And the AI should be able to win through attrition if the player can't.

A higher level AI should not be smarter or have more buffs - it should have more "goals" in the game. The player, in an RPG never has to worry your AI ally will take all your parties gold and run in the middle of a resting session - why should i fear the same of my allies in civ?

Just saying that players need certain things "they can count on" and the AI needs to consider, based on what kind of challenge is appropriate, how to behave in a way that the player needs to "achieve a task".

Civilization has always been a sandbox, but it needs a "direction" of sort -Sandboxes are good, they are great but they need to ENGAGE the player differently than the current AI does.

Really nice post, now read the OP and tell me how what you wrote relates to it please. He's playing on Warlord and can't take a freaking city, while we are whining that it's in fact too easy to conquer the AI even on Immortal.

What exactly is the AI in this case supposed to do? Hand the city over for free because the player can't take it? I mean, seriously?! He's playing on Warlord and he can't take a city, so he's blaming the game for it? And you are defending him?!

 

[thadian]

Yes, he is a lower level player than some of us, and i am a lower level player than some of you (emperor).

That said, on the lowest difficulty settings, the AI needs to engage the player - First, the city. The city he is having a problem with probably has walls and he is probably using spearmen and warriors. When your first starting out, you don't know any better. The lowest levels of gameplay, with this in mind, should not build walls - and 3 warriors should be able to cap a city on those low difficulty modes.

Once the city is capped, 1-2 warriors should be built by the AI who immediately tries to retake the city - a battle the player should be able to win. This level is hardly about competitive gameplay, and more about learning the game.

So, this achieves that - he "cap's a city" - then has to "protect his gains", learning more about the game. Now, if he were on prince, the city would have walls up quickly and a spearman in the borders - The difficulty level should consider the needs of the player and use an AI that is responsive to those needs.

On the lowest levels of game play, the player needs more stability and hard lines, for example, the AI should NEVER pick the honor tree, and should moderate its expansion and tech progression to the player. The AI on a higher difficulty level should try to moderate itself to the game leader, player or not.

Edit: He might have not decided to attack the city until it was already "defensible" against his assault. However on the lowest levels, cities should not put up walls and guard with archers until at least ONE city has been captured by the player, "alerting" the AI that walls are needed. This would make the player feel an evolution - first the AI had free-fall cities, but this next city sure is tough.. oh, it has a wall! Oh no, an archer was just rushed, i better retreat and try again when i have an archer!

 

[The_J]

Moderator Action: Stop the bickering!
It's better to give concrete advices than to smash your heads against a wall.

 

[Putmalk]

Then he needs to get more skill, this is not an issue, and it's a rant.

If he wants to learn, he can play the tutorial or try Settler difficulty (come on, Chieftain is a joke, too...)

Everyone learns by having to bang their head against the wall....and the AI is a joke on those difficulties anyway. Human has bonuses, AI has penalties. What more handholding is necessary?

thadian, come on, man...

 

[thadian]

Im just saying the AI needs to think about the "Rhetorical Situation" of the game state, if you will.

Chieftan? check.

This player needs the AI to "sand bag" a little while he figures out different unit types. Of course the tutorial helps but i have NEVER read a tutorial - i learn more from forums anyway.

Skill is something people develop at a different pace. When i first played Civ 1 and Civ 2 as they were coming out, i couldn't even get passed having 2 cities on the lowest difficulties, i would just build a few cities and workers and rome would come in knocking... I didn;t know any better. I gave up until civ 3 came out, and a friend "assured me" that it was "balanced unlike the others" and for some reason, i was able to understand it and play on Sid mode with ease.

The games AI just needs to be a little more newb friendly is all.

As for the TS:

Start by building a scout - Then have it reveal the area RIGHT around your borders for nearby huts, build a second scout and have them both roam out. Then build a warrior, and by this time barbarians should be in the game. Move your units back to your border and take out the barbarians for EXP.

There is no "perfect moment" to expand, or perfect rate by which to expand. Just ask if you want less than 5 cities (tall empire, cities with population 20) or more than 5 cities (wide empire, about 12 cities with pop 3-5). Lots of good strategy articles written by players better than myself, so check them out!

 

[Kabloosh]

Most of the tools to get better are already in place. Read the tool tips of particular units.

You might want to bring a few siege weapons or at least ranged units to attack a city especially if it's defense rating is superior to any unit you can build.

Keep in mind the AI isn't going to spoon feed you the victory either. On the lower levels generally the reason why you would be behind or failing on assaults would be of your own undoing. You need to make sound tactical decisions since if you aren't careful you could end up losing a lot.

 

[Atlas627]

@Thadian

Your posts show a good understanding of proper game flow. That seems to have nothing to do with the OPs post (which made no sense to me), but gives me something to talk about.

As to being a DM: I actually just stopped writing up material for tonight's meeting as I post this. And you are correct, the point of the DM is NOT to defeat the players, but to give them a challenge (and the RISK of failure). Now, of course the risk is not real if it never happens. The difference here with your comparison to D&D is the game mechanics. In a tabletop RPG, the party can succeed, it can fail, and it can come out somewhere inbetween. And this is important. Another important difference is that each DM takes their time to develop something perfect for the players. They can adapt on the fly. Programming an AI is not like that, especially for a complicated game.

Which brings me to the point about the AI adapting based on difficulty level. This would be wonderful to have, of course. In fact, it should probably adapt regardless of difficulty level, as it thinks it doesn't need walls until you assault them and succeed! But programming an AI is difficult, especially because this is such a complicated game. And Firaxis is limited by time, money, and (consequently) manpower. You simply can't program that many different AIs. There is a reason why D&D is so successful. They have "unlimited" talent, manpower, time, and money, because YOU do all the work. Each DM makes a game, a mod if you will, and makes everything "perfect" for these specific players.

Now, how do you learn the game? In Civ3 I never got off of Chieftain. Why? Because there was SO little challenge, I never even interacted with the AI. I didn't realize there was more to the game because the AI didn't take initiative, so I never learned.

With the penalties to AI Chieftain difficulty provides, the AI DOES "sandbag" around, as you put it. They try to do all the same things as before, but it just can't. So in a sense the difficulty IS tailored to your learning curve, just not within the same difficulty (the AI is not programmed to go easy on you until you prove your competence).

@everyone else

Give a response, be civil. So far all I see are essentially 'Thadian is n00b' posts. And same for the OP. Be productive and helpful! Remember everyone reads the posts, not just people who respond to them or who start the thread. So even if you think the OP is just ranting, you can still provide useful advice. Maybe someone else will read it!

 

[PhilBowles]

The job of the dungeon master is to design dangerous encounters that CAN kill the player, but it's not much fun if the Arquebus-wielding kobold scores 3 critical hits and kills the player. As a DM and having designed every form of encounter you could imagine and far far more -

Very different style of game, and also one in which you aren't going to have players taking the role of the kobold; if they were you'd want to imagine that the kobold would do everything it can to survive the encounter and beat its aggressor. A strategy game is by its nature a game where the enjoyment is derived from having the superior strategy; it relies for enjoyment on the player feeling they have outperformed/outwitted their opponent. No, it's probably not a style of game designed to appeal to roleplayers - and indeed when I play computer RPGs like WoW I find I get bored and/or frustrated very quickly with the spoonfeeding these games' AI likes to do, where difficulties are set so that it's hard to get killed in the first place, but even if you are this is essentially without consequences. It ruins immersion and any sense that the challenges or events are real if you can't sense that the NPCs are looking out for their own interests and instead will fight to the death just because that helps the players gain experience. As for paper RPGs, I grew up on Warhammer Fantasy Roleplay (the original) - in which combat was brutal and a single critical hit can readily kill a player.

Consider that Civ has had both single- and multi-player options through most of its incarnations. If you're playing against humans, I doubt you'd come on here and throw a tantrum because you're frustrated that the other guy is trying to win. You expect him to. It makes no sense to expect the AI to play a game using the same game engine any differently; and if it did the single-player experience would probably pale in comparison to the multiplayer experience.

the AI should not play to win, it should play for entertainment. Each AI should have "it's goals" and some of those goals should overlap one another at different moments. Conflicts should be dangerous - and winnable at the same time. Reminiscent of RFC UHV's.

This seems to conflate two different issues. The AI does have goals; those goals include win conditions. And in my experience there is no more satisfying Civ experience than a late-game scramble to deny Denmark the diplomatic votes it needs, racing to complete the Utopia Project before the Russian spaceship takes off, or completing your own spaceship just as the Ottomans are beating at the gates of your capital desperate to prevent your victory (all examples from my Civ V games - and in one of them, the Utopia Project one, a loss to the AI). This is playing to win on the AI's part.

What you're talking about is difficulty beating the AI in combat. An AI that wipes you out, even if it's not going for domination, is not playing to win, it's playing to beat the human specifically. This is frustrating and bad for enjoyment as well as for the game. It can be good in some circumstances - in that game where the Ottomans were a turn away from taking my capital when the spaceship took off, neither they nor their Songhai allies could have won, and in geopolitical terms they would basically be handing victory to the Inca, who they didn't even like much.

EDIT: In terms of the city attack issue, I have experienced this recently too, and yes it can be remarkably frustrating! I would agree that this is something that could be looked at - it's good if it's hard to take a city per se, if the AI is in a position to defend it effectively and it's well-placed. But I had one game in which Attila was to all intents and purposes out of the game, having lost his entire army and, with Attila's Court under siege and every unit he brought out being killed, with no way of rebuilding to resume the attack. If the war has already been lost, it should not then take a millennium to capture the city itself (it was in a well-defended position and with one garrison unit).

But when it becomes commonplace and done just for the sake of it, and especially when you invest the time to half-complete a game to lose to a sneak attack rather than meaningfully losing the game because you were outcompeted, then it's boring and ultimately annoying. Not because the AI plays to win, but because it plays to cheese - in the general scheme of things anyone can accept losing games of Starcraft. But no one likes losing to Zergling rushes. It's irrelevant whether the player on the other end is a human or an AI.

Gods & Kings has helped to an extent - in my last full game the Germans hated me for much of the late game, but they wouldn't have gained anything by going to war with me since they were aiming for a science victory, were in the event able to achieve it before I was, and were on another continent at war with Greece anyway. So they never declared war, just rushed to win the game.

Though I've had similar experiences in vanilla, as well as random backstabs in G&K (quit one game after a random backstab from Isabella, and in my current game Bismarck decided he didn't like me even while we had a declaration of friendship), at the very least G&K diplomacy gives you more options to at least give the illusion of control over events. Mongolia has a giant fleet and wants to take Kuala Lumpur? Then if I boast that I'm going to protect it from them, I deserve what happens next. I used to be friends with the Siamese and they take a dislike to me? Well, I did try converting their holy city so they probably can't be blamed for getting annoyed. Even Bismarck in this game - I did at one point have a DoF with his Swedish enemies, and he then took a dislike to my Siamese friends - given that he's a bit paranoid generally, he might start to think there's a pattern... (He's not at war with me yet - at the end of my last session I signed defensive pacts with Ramkhanhaeg and Hiawatha, so it will be interesting to see what happens if he does declare war).

A game should make you thrilled - not mad. The AI should play to entertain and cater to the player. Based on the players actions, the AI should try to "intervene" to prevent success, in a way that the player can respond to.

Unfortunately Civ is badly-designed to allow this generally, a limitation of the core game design rather than Civ V specifically. It's not a particularly interactive game, and the major way to actively interfere with your opponent's strategy is by going to war with them; if you can't just get ahead by rushing to your win condition, that's usually your only option. So an AI aiming to prevent success is one that's necessarily more belligerent - ironically an AI that's not trying to win the game but just to present challenges for the player is more likely to result in these kinds of frustrating situation than one that is playing to win.

An AI that's playing to win won't have any need to destroy or attack other civs that don't represent a threat, any more than a player will - I don't usually go for domination victories, and rarely play to completely wipe out other civs otherwise, not least because doing so will make aggression from remaining civs more likely. An AI playing to win has no motive to behave any differently.

Really nice post, now read the OP and tell me how what you wrote relates to it please. He's playing on Warlord and can't take a freaking city, while we are whining that it's in fact too easy to conquer the AI even on Immortal.

What exactly is the AI in this case supposed to do? Hand the city over for free because the player can't take it? I mean, seriously?! He's playing on Warlord and he can't take a city, so he's blaming the game for it? And you are defending him?!

The way I put this in another thread is that the Civ V AI scales badly. In Civ IV and earlier Civ titles, you could get by at low levels with sandboxing, not specialising, not expanding, etc. etc., all strategies you'd get punished for at high levels. That made the game accessible to new players and difficult for experienced ones.

Civ V has much more of a blanket approach - you need to play as though you're experienced at pretty much any level of play. You don't specialise, you lose whether you're on Prince or Immortal. You don't build enough units - same applies. Which makes the game less accessible to begin with, but once you've mastered the basic strategies that work at lower levels, the same strategies work in much the same way at the higher levels - with the result that after you reach a certain threshold in the quality of your play, the game becomes "too easy". Emperor doesn't really force you to adjust the way you play to win once you've learned how to beat King.

Once the city is capped, 1-2 warriors should be built by the AI who immediately tries to retake the city - a battle the player should be able to win. This level is hardly about competitive gameplay, and more about learning the game.

So, this achieves that - he "cap's a city" - then has to "protect his gains", learning more about the game. Now, if he were on prince, the city would have walls up quickly and a spearman in the borders - The difficulty level should consider the needs of the player and use an AI that is responsive to those needs.

This is all very true, but I think that the AI does attempt this to some degree. I've noticed when playing low-level games with less experienced friends that the low-level AI will not fire with garrisoned forces, which it does by default at higher levels, and it does employ fewer units (simply as a consequence of getting fewer bonuses if nothing else). Indeed this is something I don't think was as true of older Civ AIs, when all that ever changed was the bonuses but the AI was essentially identical at all levels of play.

 

[veteranspy]

When you are asking for AI for pull its punches because the player is 'noob' you are essentially asking for a very advanced ai.

as for the game design philosophy of not really playing to win but playing to entertain, I disagree with that.

Strategy game is fun because you get to sit back and think about what you are doing. You are satisfied when you overcome adversity. The only way for you to get that is by knowing that if you play badly, you will lose.

The DM has very different goals from an AI in civ5. A setting with a group of players and a dm is different from a setting of civilization. in civ, the AI competes against you. A DM does not do that. A DM facilitates the game by being the person who operates the environment in which the players play. Given that a DM has infinite power, if the goal of DM is to win, he would just drop a rock and make people roll to see how painfully they die. An AI does not have infinite power, they are meant to match the player. Since they can't make an AI smart enough to match intellectually, they compensate by providing bonus to gameplay values.

To ask an AI to entertain and not to win is both impractical and a bad design.

 

[crasher_pt]

the AI should not play to win, it should play for entertainment.

"That's just, like, your opinion man..."

The games AI just needs to be a little more newb friendly is all.

How can the game benefit from a moronic AI? We had this example at vanilla release when the AI was dumb as rocks and threw civvies and archers at your melee units. Was this fun?

Put it another way: how does the game benefit from an AI who doesn't want to win? Is it to please those with low tolerance for frustration? The design has to draw the line somewhere.