How weak is the AI?

[Miravlix]

When Rome declares war on me my first instinct isn't to decompile the code to review it. My first instinct is to swear at the screen and vow to destroy him.

I'm not sure what the point is to tell us your instinct is self destructive. When dealing with computers you should always figure it out, because computers is dumb and easy to predict, especially since Firaxis refuses to put any effort into AI and just pile on bonuses.

 

[Ornen]

I agree that AI bonuses are fine and to be expected. That said, I'm still concerned the game is too reliant on them. Even at Emperor, a couple steps below the highest setting, the AI practically starts with a bonus city and starts building a carpet of doom very early.

We'll see when it comes out. I'm not naive, and not demanding no AI bonuses. But I'd rather play against an ai that leans on its bonuses without relying on them completely.

 

[rastak]

I'm not sure what the point is to tell us your instinct is self destructive. When dealing with computers you should always figure it out, because computers is dumb and easy to predict, especially since Firaxis refuses to put any effort into AI and just pile on bonuses.

You are correct that the rational course of action would be decompile the code and understand exactly every move it will make or at least know the exact probabilities. That sounds like fun! Then again the even more rational thing to do would be not waste valuable time playing games.

 

[Magil]

I agree that AI bonuses are fine and to be expected. That said, I'm still concerned the game is too reliant on them. Even at Emperor, a couple steps below the highest setting, the AI practically starts with a bonus city and starts building a carpet of doom very early.

We'll see when it comes out. I'm not naive, and not demanding no AI bonuses. But I'd rather play against an ai that leans on its bonuses without relying on them completely.

When 1UPT was introduced, the AI could not use it at all. Keep that much in mind. Now consider the new complex systems Civ VI has introduced. While your expectations are not unreasonable... I'd temper them a bit. The AI is most likely going to be carried hard by its bonuses this time around, because I have no faith in the AI's ability to manage its empire with the many intricate and interconnected systems at play.

You are correct that the rational course of action would be decompile the code and understand exactly every move it will make or at least know the exact probabilities. That sounds like fun! Then again the even more rational thing to do would be not waste valuable time playing games.

Let us remember that Civ is a strategy game. For strategy to be possible, one thing should logically follow from another. This, to some degree, includes the behavior of your opponents (especially when they are AI--human players are more unpredictable, but can still ruin the game if they play in a way that is not strategically sound).

"Decompiling the code" is a bit much, but acting without understanding is about as far from strategy as you can get.

 

[stealth_nsk]

What's the problem with decompilation? In Civ5 all AI source code is available to be viewed already, it even has comments in it. While you're waiting for Civ6 you could look at the code and try to understand what AI is doing.

 

[Magil]

What's the problem with decompilation? In Civ5 all AI source code is available to be viewed already, it even has comments in it. While you're waiting for Civ6 you could look at the code and try to understand what AI is doing.

I wouldn't expect the average player to decompile the code in order to understand why the AI acts the way it does. That strikes me as expecting a bit much. I would expect the average player to understand why the AI makes a particular decision--most of the time. During gameplay, using the information provided to them during the game.

Not saying there's anything wrong with decompiling the code, but it should be obvious enough such that it's not necessary for the majority of cases.

 

[stealth_nsk]

I wouldn't expect the average player to decompile the code in order to understand why the AI acts the way it does. That strikes me as expecting a bit much. I would expect the average player to understand why the AI makes a particular decision--most of the time. During gameplay, using the information provided to them during the game.

Not saying there's anything wrong with decompiling the code, but it should be obvious enough such that it's not necessary for the majority of cases.

It was a bit of joke, actually. You can't understand the code as a whole, only parts of it. Similarly, for big AI actions you can't precisely tell why it was done - good AI weights many things.

Also, there's a problem with studying how AI works internally - you could discover exploits you'd be unable to see normally, breaking immersion.

 

[Magil]

It was a bit of joke, actually. You can't understand the code as a whole, only parts of it. Similarly, for big AI actions you can't precisely tell why it was done - good AI weights many things.

Depends on what you mean by "precisely". Sometimes you can't see all of the reasons, but often you should be able to see enough of them. That is the point of things like leader agendas, warmonger penalties, and other visible diplomatic modifiers in Civ VI after all--these are all presented to the player for the purpose of giving him or her a way to understand the AI's priorities and its decision-making process. They exist in the game for a variety of reasons, like creating an engaging experience and gameplay balance, but they are clearly visible to the player so that the player can make informed decisions. Recall that vanilla Civ V's diplomacy was extremely poorly received at least partly because it went out of its way to hide these sorts of things from the player.

Also, there's a problem with studying how AI works internally - you could discover exploits you'd be unable to see normally, breaking immersion.

Irrational behavior tends to be more immersion-breaking to me as opposed to behavior that can be logically explained, even if only retroactively. Furthermore, I'd like to think the people who program and/or mod the AI can still be immersed in the game to some degree.

 

[rambow13]

Anyone who says they are not disappointed in finding out AI doesn't scale past Chiefdom is lying to themselves. The scaling bonuses and boosts have always been one of the most annoying aspects of CIV. Yes it makes the game more difficult, so it achieves the end goal, but it seems cheap and grows old.

I was really hoping with CIV 6 and all the complex systems and changes that they made, that they would finally really scale AI. Hopefully this is a big system upgrade in the first expansion, because it would be really fun to see how well a deity level AI would use all the systems at its disposal.

 

[stealth_nsk]

Depends on what you mean by "precisely". Sometimes you can't see all of the reasons, but often you should be able to see enough of them. That is the point of things like leader agendas, warmonger penalties, and other visible diplomatic modifiers in Civ VI after all--these are all presented to the player for the purpose of giving him or her a way to understand the AI's priorities and its decision-making process. They exist in the game for a variety of reasons, like creating an engaging experience and gameplay balance, but they are clearly visible to the player so that the player can make informed decisions. Recall that vanilla Civ V's diplomacy was extremely poorly received at least partly because it went out of its way to hide these sorts of things from the player.

The problem is - computer has no problem calculating thousands of parameters. For example, it may take several parameters to calculate how good each spot is for a city. It may take this data as one of the parameters to determine which side to settle. Those 2 parameters could be used as a weight to giving priorities for producing a new settler. In addition, where are other parameters like current policy cards which affects these weights. And additional algorithms which decides where to build globally needed things. So you look at a city building a settler and even knowing all ins and outs you don't know why AI decided to build it. It just works differently than human brain.

Irrational behavior tends to be more immersion-breaking to me as opposed to behavior that can be logically explained, even if only retroactively. Furthermore, I'd like to think the people who program and/or mod the AI can still be immersed in the game to some degree.

My favorite example is Civ3 AI exploit. If you don't know about it, you see AI attacking a poorly defended city and it's good enough for you to have challenge. You may not know why AI is attacking this particular city, but it's not a big problem. If you know about the exploit... Even if you make a "home rule" to avoid using it, your immersion is broken.

 

[Trias]

Anyone who says they are not disappointed in finding out AI doesn't scale past Chiefdom is lying to themselves. The scaling bonuses and boosts have always been one of the most annoying aspects of CIV. Yes it makes the game more difficult, so it achieves the end goal, but it seems cheap and grows old.

I was really hoping with CIV 6 and all the complex systems and changes that they made, that they would finally really scale AI. Hopefully this is a big system upgrade in the first expansion, because it would be really fun to see how well a deity level AI would use all the systems at its disposal.

Building an AI that truly scales is incredibly hard (as in much, much harder than building a single AI that is decent at the game). Since thus far civ games have shipped with an AI that meh at best, it seems like an incredibly waste of resources for Firaxis to work on a scaling system. I'd much rather they invest their effort in improving the base AI to play better. If I remember correctly, the "even" level in civ4 was prince, so the baseline for the AI being at chieftain for civ6 is a bit of a disappointment.

The frustrating aspect of AI bonuses in civ has usually been that the bonuses do a poor job at simulating better play by the AI. Often not adequately covering the weaknesses of the AI, thereby requiring much larger bonuses to provide a challenge.

 

[Magil]

The problem is - computer has no problem calculating thousands of parameters. For example, it may take several parameters to calculate how good each spot is for a city. It may take this data as one of the parameters to determine which side to settle. Those 2 parameters could be used as a weight to giving priorities for producing a new settler. In addition, where are other parameters like current policy cards which affects these weights. And additional algorithms which decides where to build globally needed things. So you look at a city building a settler and even knowing all ins and outs you don't know why AI decided to build it. It just works differently than human brain.

You're making this a lot more complicated than it needs to be. Yes, the AI has a lot of different algorithms to run and factors to consider when it's making a given decision. But quite often it can be boiled down to a few easily visible factors. Again, I will point to things like diplomatic modifiers and agendas. You know why Victoria sent her settler to another continent to settle--that's her agenda. You know why she denounced you, because of a warmonger penalty.

Sometimes, yes, not readily available information will cause the AI to behave differently than expected. But that isn't--or rather, shouldn't be--the norm, or anywhere close to it. My point is--and I will firmly stand by it--that it takes the strategy out of a strategy game if the AI isn't largely understandable--if not totally predictable--for a player who wants to understand what's going on.

 

[isau]

I don't know yet about the AI. I have played the Civ V Community Balance Patch at higher difficulties and had to drop down a few difficulty levels to survive. I'm sure they have access to that code and can see what changes were made to make at least possible for AI to slow the player down. A lot of the changes in Civ VI seem to have been inspired by the CBP.

I still remember the time Harald of Denmark became a global terror and ended up swallowing my entire continent. In the Civ IV ruleset he would have won a dom victory too. Too bad Civ V and VI use the "conquer the capitals" rule instead of tile grabs, because the AI was really good at winning under those conditions.

 

[Trias]

When 1UPT was introduced, the AI could not use it at all. Keep that much in mind. Now consider the new complex systems Civ VI has introduced. While your expectations are not unreasonable... I'd temper them a bit. The AI is most likely going to be carried hard by its bonuses this time around, because I have no faith in the AI's ability to manage its empire with the many intricate and interconnected systems at play.

Actually many of the new systems make me slightly optimistic about the prospects for the AI. Having many intricate and interconnected systems in the game may ultimately help the AI. For most of these systems it is not that hard for the AI to consider all options. The problem faced by the AI is that especially in civ V on the short term most options seemed very same-ish, and since long term stratagy is not the AI's strong suit, it struggled. Things like adjacency bonuses may help the AI differentiate more between options, and guide it long-term.

One thing the some of the previewers have been reporting is that they find themselves focussing more on short term goals. That is good news for the AI. Short term is approachable for the AI. Just maybe the designers have actually been thinking about what the AI can and cannot do well when designing the new mechanics. (Then again maybe not.)

 

[Cyberian]

Anyone who says they are not disappointed in finding out AI doesn't scale past Chiefdom is lying to themselves. The scaling bonuses and boosts have always been one of the most annoying aspects of CIV. Yes it makes the game more difficult, so it achieves the end goal, but it seems cheap and grows old.

I am not disappointed and I don't think I am lying to myself. I would argue that the peoples who want a scaling AI difficulty that can compete with players in strategy games know very little about the history and current state of AI in computer games and have unrealistic expectations because of that.

There is not a single strategy game on the market able to achieve that. But every release of another strategy game those peoples come out of the closets and complain about stupid AI that can only compete on higher difficulty because of handicaps to the player and/or bonus to the AI.

 

[Magil]

Actually many of the new systems make me slightly optimistic about the prospects for the AI. Having many intricate and interconnected systems in the game may ultimately help the AI. For most of these systems it is not that hard for the AI to consider all options. The problem faced by the AI is that especially in civ V on the short term most options seemed very same-ish, and since long term stratagy is not the AI's strong suit, it struggled. Things like adjacency bonuses may help the AI differentiate more between options, and guide it long-term.

One thing the some of the previewers have been reporting is that they find themselves focussing more on short term goals. That is good news for the AI. Short term is approachable for the AI. Just maybe the designers have actually been thinking about what the AI can and cannot do well when designing the new mechanics. (Then again maybe not.)

I don't know about that, Civ V seems to reward advanced planning in many respects. I might determine that a city is good for an industrial complex long before I actually build the complex, because it has a cluster of hills/mining resources, informing my city placement and specialization (and planning to have space for the district in regards to population when it becomes availible). Or I might invest 3 turns into Masonry, and then stop, and move to another tech, because I know that in another 3 turns my builder will finish, and put down a quarry. I might determine that my next settlement might irritate Victoria because it's going to be on another continent, so I send her a delegation and arrange beneficial trade deals in advance of that to make sure they get through. Past experiences with Civ AI shows it's not the greatest at handling this kind of stuff. But who knows? Anything's possible. I just don't expect it on any level.

 

[rastak]

Look, I'll be honest. I am certainly concerned with the AI based on the you tube guys. That said, I play these games to relax and have a good time. I get more than enough heavy intellectual exercise at work, I don't look to Civ to provide that. It obviously requires thought and planning but so does designing your back yard landscaping.

I'm hoping the game allows some immersion, is competitive enough where I can't just roll over everyone and I can tell a story for a few hours. I'm hoping it's stable enough to play a few games with my son-in-law and a buddy that stops by every couple of weeks. How they make that happen behind the scenes doesn't matter a whole lot to me.

 

[Haggbart]

I am not disappointed and I don't think I am lying to myself. I would argue that the peoples who want a scaling AI difficulty that can compete with players in strategy games know very little about the history and current state of AI in computer games and have unrealistic expectations because of that.

There is not a single strategy game on the market able to achieve that. But every release of another strategy game those peoples come out of the closets and complain about stupid AI that can only compete on higher difficulty because of handicaps to the player and/or bonus to the AI.

100% agree with this.

Also the AI primary function is not to be an equal opponent, it's to be a character in your play, or a roadblock if you prefer the mechanical description.

That's why people on the forums were furious in vanilla CiV when the AI would DOW without "roleplaying" reasons. Because AI primarily playing to win the game is not what people generally want.

 

[Trias]

I don't know about that, Civ V seems to reward advanced planning in many respects. I might determine that a city is good for an industrial complex long before I actually build the complex, because it has a cluster of hills/mining resources, informing my city placement and specialization (and planning to have space for the district in regards to population when it becomes availible). Or I might invest 3 turns into Masonry, and then stop, and move to another tech, because I know that in another 3 turns my builder will finish, and put down a quarry. I might determine that my next settlement might irritate Victoria because it's going to be on another continent, so I send her a delegation and arrange beneficial trade deals in advance of that to make sure they get through. Past experiences with Civ AI shows it's not the greatest at handling this kind of stuff. But who knows? Anything's possible. I just don't expect it on any level.

That was exactly my point. The AI is bad at planning in the long term. In particular, it will typically be bad at finding moves that make little difference now, but make a huge difference far along the road. The civ6 systems seem much more focussed at providing immediate short term bonuses differentiating moves. If this has been setup properly in such a way that following the short term queues will result in a decent (but not necessarily brilliant) long term strategy, this could help the AI tremendously.

 

[isau]

100% agree with this.

Also the AI primary function is not to be an equal opponent, it's to be a character in your play, or a roadblock if you prefer the mechanical description.

That's why people on the forums were furious in vanilla CiV when the AI would DOW without "roleplaying" reasons. Because AI primarily playing to win the game is not what people generally want.

Agreed. The AI is always playing with blinders, because it's playing a role. That is why it will always need bonuses of some kind.

That said I do wish they'd go back to the Civ IV version of the domination victory. The AI is terrible at dom victories because it can't make the plans to conquer the capitals. But in Civ IV it definitely could win a Dom victory if you let it go unchecked. I think that change alone is what significantly affected the perception that Civ V is too easy. It's too easy to just ignore the AI entirely because they simply can't win a Dom victory without taking your capital, and they'll always be terrible at that.