AI in Civ 6
Images are sometimes not shown due to bandwidth/network limitations. Refreshing the page usually helps.
You are currently reading a thread in /v/ - Video Games
You are currently reading a thread in /v/ - Video Games
>Civilization V‘s oft-maligned A.I. is no more
>“The number one thing to know is that Civilization VI is not built on the previous engine,” says Beach in response to a question about critiques of Civilization V‘s ability to play the game on its own terms. “All of the gameplay systems have been entirely rewritten and re-architected. We’ve specifically set it up to be very modifiable. And we’ve rewritten the A.I. from the ground up, learning all the lessons that we had from Civilization V, so we know how best to solve some of these problems in military combat and so forth.”
>“It’s a much more goal-oriented A.I. than in the past. The A.I. in Civilization V was very much based on a flavored weighting system, and this one is much more able to follow all the chains of logic, like ‘I need these tech boosts to get to this part of the tree.’ So it’s a very different approach.”
>“The other thing is that we’ve changed the way leaders work in diplomacy. In previous versions, all of the leaders approached the game in basically the same way. Here you have very different key triggers, with A.I. that’s now driven by what we call leader agendas. A leader agenda is a specific historical way of approaching the world that we assign to each of the leaders based on what they were good at when they were ruling their empire.”
>“Each leader has a historical agenda that you’ll learn as you play the game repeatedly. But we also assign secret agendas to the leaders that you have to uncover through espionage. And so the diplomatic landscape is much richer. The more you mix all the different personalities into a big soup, the more you end up with a very interesting diplomatic landscape.”
Full article with more info on other features: http://time.com/4324490/civilization-6-interview/#prclt-FQ1DJIiB
Thoughts?
>“We’re aware that as we add depth, we have to make sure this game is just as easy to play as any of our other titles have been,” says Beach. “There’s a big push on user interface with this game. We’ve taken a whole new approach to user interface development, and there’s a lot that helps players understand what’s going on in the game.”
>“A decision isn’t difficult to make if you have context for that decision,” adds Feldges. “I think it’s going to make it much more accessible, but that’s the balancing act you always have with a Civilization. If you make something accessible, you get criticized for dumbing it down. If you make it too complex, you run the risk of alienating new players. And so we’ve been very conscious of that, and I believe on a very good track.”
At least they're self-aware
but it looks bad
>>338290689
>In previous versions, all of the leaders approached the game in basically the same way. Here you have very different key triggers,
This is objectively false, since in Civilization V some AI leaders took a bigger diploacy hit if you attacked a city state, or build a wonder they wanted, or sent religious missionaries towards them and so on.
Also there is a straight up stat that increases the likelihood of being backstabbed.
So this guy is lying about Civ5, hopefully out of ignorance and not malice.
>>338292141
Obviously he's not going to get very technical in an interview. The only civ I would consider to have very clear goals in Civ 5 is Greece.
>>338292475
Some leaders will almost always spread fast and wide.
Some leaders will almost always attack their neighbor fast.
There are stats like that, who builds more walls, who builds more infantry or more ships, etc.
They sure as fuck don't all approach the game the same way.
I am afraid that we will get this same exact system, with a couple of minor tweaks, and it will be called a brand new leader character so they are all different.
>>338290689
>And we’ve rewritten the A.I.
Cant they just usa a neural net to teach the ai/let the ai learn on its own to play the game like Google does?
Traditional game AI always relys on probalistic decisions - "with x% weight do action y", there's no real decisions and it always ends up doing retarded shit.
>>338290689
Yeah yeah, i'm sure it sounds great on paper.
Let's wait and see.
>>338293183
Civilization is a horribly complex game, many times more complex than chess. Do you know how long good chess AI takes to make a turn? Now imagine waiting for 7 such AIs, in a much more complex, and thus longer to process, game.
Having a neural network AI would be a tremendous amount of work, and not worthwhile.
>>338293459
If your computer has less power than this little thing you should probably upgrade.
>>338292865
The thing is, you ALWAYS know Alex will suck City-State dick. You ALWAYS know Alex will dominate. You ALWAYS know Alex will be your enemy. You ALWAYS let out a large sigh when you see his smug face first ride up to you in a game. You can't really say the same for the other AI unless they have high expansionism values. You don't really know why a certain civ denounces you out of nowhere for example. Seems like Fireaxis is going to make the AI very transparent with their main intentions and more fanatical about certain stuff based on that leader's history.
Also
> But we also assign secret agendas to the leaders that you have to uncover through espionage.
might b cool
>>338293674
If you think this is good chess AI you should probably kill yourself.
>>338293847
>> But we also assign secret agendas to the leaders that you have to uncover through espionage.
>might b cool
This also already happens.
>X is plotting against you
>X is constructing Y wonder
>X is preparing to launch a surprise attack on some civ
>>338293854
It doesnt have to beat all the grandmasters, if it beats 95% of the players it's awesome.
Equally the CIV game AI doesnt have to beat the best players who dedicate their life to the game, just most players if it's released from all artificial constraints.
>>338293945
But that just tells you what they're doing at the moment, not their overall personal goal or "agenda".
New engine that allows robust mods huh? Now where have I heard this before...
>>338294060
Deity Civ already beats the majority of players.
>>338294296
because it cheats, that isn't good ai, good ai should beat you on an even playing field, not by getting crazy resource and tech boosts
>>338290689
I'm just happy that the AI leaders will act like they're leaders of their country, not another player. I hated that in Civ IV I could make a lifelong friend, while in Civ V my buddy would turn on me because "you're going to win the game"
Shit, I'd play multiplayer if I wanted to play against people. I want to fuck around in history, facing off against history's greatest leaders, not some nameless and faceless player wearing their face.
>>338294296
By cheating.
>>338294114
They actually didn't speak about it during Civ5 or Beyond Earth, so I don't know where you've heard it.
>>338294464
All strategy game AI cheats, or is easy. These games are stupidly complex.
Choose, easy or cheating. I choose cheating every time.
>>338290689
>still 1upt
It can fuck off
>>338294656
>still mad
Top kek.
>>338293674
Chess is a super simple game with super simple rules when compared with something like Civ.
>>338295000
Don't bother, a month ago someone on /v/ was writing tl;dr walls of texts explaining how chess is more complex than DOTA2, a real time game where people cooperate to achieve vague goals and features some defeats being strategic victories, and things like pyrrhic victories.
>>338295000
You need a month at most until you can play CIV very well.
you need 5 years at least until you can play Chess very well.
>>338295154
You need 7-8 hours at most to understand why such statements are moronic.
The weekend is coming, read up on AI. It could save you a lot of humiliation when posting online.
>>338290689
>Civilization V‘s oft-maligned A.I. is no more
>marketing
>the old one we worked on for 6 years was trash! THIS is the real deal trust us ;)
>>338293183
AI like that still uses weights and probabilistic decisions, but they're just more nuanced. Still, training an AI player to be really good at Civ wouldn't necessarily create a fun experience. There's an old lecture out there by one of the developers for Civ III, as far as I can remember. To use the lecture's terminology, it was essentially about "good" (competent) versus "fun" (flavourful) AI. In summary, a game like Civ has to balance the two carefully.
If you were to train an AI for Civ, you'd somehow need to incomporate the "fun" aspect, which is kind of difficult. You want an opponent that looks human. You could train it to win, but it could become too good and feel cheaty. You could train it to work towards some agenda (e.g. to achieve world peace rather than winning), but it might end up learning strange, inhuman methods that you have no control over.
I don't think I put it in words very clearly, but hopefully you get the idea. Applying modern AI techniques becomes tricky when human behaviour is a goal. I do think it's possible, but if you manage to get good results out of it you'll have a very nice paper to publish.
>>338295154
Look you can simply run every possible outcome in chess then choose the best one. Civ has nearly endless possible combinations.
>>338295503
I get what you mean and i did mention "constraints" on the ai, it's pointless to let a new CIV player play against the CIV ai equivalent of Garry Kasparov, difficulty levels if you want to call it that.
What we have right now is the AI doing... something. I've never seen the AI decide to turtle up and pursue a science or cultural victory, it simply does a little bit of everything and nothing really well.
>>338294656
Eh, they're working on the main problem with 1upt by allowing you to not only merge similar units into a "corps", but also stack several different unit types into the same tile.
>>338295940
>you can simply run every possible outcome in chess then choose the best one
That's how the first Chess AI victory against a grandmaster was achieved but it's not how modern Chess AI works. They do make errors.
>>338296056
but muh doomstacks
muh doomstacks motherfucker
>>338295940
You've got that backwards, I'll just assume you're trolling and not seriously that dumb.
>>338296190
Yeah, instead of calculating the whole rest of the game for each possible move on every turn, they calculate only 30 turns ahead or something.
Same principle, and for the same reason impractical for a complex game like Civilization.
>>338296210
>i have no argument, so you are dumb
>In Civ 5 Gandhi was a genocidal madman who stopped at nothing to cause death and misery in every play thru
>In Civ 6 leaders will still have "a historical agenda"
I don't think the AI will be any different from previous games.
>>338296491
Gandhi was the friendliest AI opponent in Civ 5. The last one to ever start a genocide.
>>338296451
>impractical for a complex game like Civilization.
But it's not. A unit sitting on a certain tile can only move as much as it's movement value allows this turn.
>they calculate only 30 turns ahead or something.
That is literally how Humans play Chess.
So what's wrong with the CIV AI planning ahead the next 30 moves of its puny little Barbarian unit? If it manages to keep its unit from attacking my own defensively upgraded unit standing on a hill across a river it would be an awesome improvement.
>>338296491
>I ask what is, to my mind, the most important question of all. If a leader's traits are historical, will Gandhi be less fond of nuclear armageddon? "We have a special way of handling him. Initially he's going to seem very peaceful, but he'll have a dark side."
>>338296887
How much CPU power is that going to take though? Like the anon mentioned above, you have 7 AI players along with city-states and barbarians all taking turns. CiV already has problems with slow down in the late game on large maps, and I think this would just make it worse.
>>338297419
>CiV already has problems with slow down in the late game on large maps
Blame it on the developers for not optimizing their games.
Paradox games have a whole lot more going on on the map yet it's running far faster and in pseudo realtime.
>>338296887
>But it's not. A unit sitting on a certain tile can only move as much as it's movement value allows this turn.
And science, and culture, and how many units, and what types of units, and diplomacy, friends and foes, city states, money, long term goals, short term goals, etc etc etc.
Civilization is tremendously more complex than chess. It is so obvious that I am having a difficult time explaining it to you. My dog is looking at the screen and nodding in understanding, why cant you?
>>338293183
>can't this shitty stratgame company just make a world class neural network
Okay dude, keep giving us your insight on the video game industry.
>>338297861
>My dog is looking at the screen and nodding in understanding, why cant you?
Because your dog doesnt understand that the number of choices is always limited at every point in time in CIV. Why would you plan to put a certain unit on a certain tile at the other side of the map if you havent even revealed it yet?
Again, you think the Ai would calcuate every possible choice at every moment to act in the best possible way, that's not how humans play neither would it be how a learning AI would play.
>>338298129
Neural networks have existed since the 60s. Programming and the tools to program with have come a long way since then and programmers have become very cheap to employ.
>>338298424
>programmers have become very cheap to employ
>>338298547
>He doesnt know about H1B Indians
>>338298726
>he thinks indians will make him a neutral network that will beat him in a game of civilization
>>338298726
duder, h1b's can't program cat to fart
you do realize the requirements to be an h1b programmer are like zero compared to actually putting one US citizen through college, regardless of whether or not they have any sort of natural talent for it?
My 4X game I'm working has 5 advisor AIs (research, production, military, expansion, espionage) that push information up to a "Director" AI. The Director AI replaces itself after finishing certain goals, for example, the early game director AI replaces itself after the first settlement is thoroughly up and running with a more expansion oriented director. Each "civ" has its own chart of Directors so not every civ plays the same way from beginning to end.
I'm not sure why Civ doesn't use the same technique - it works really well. Having one AI with weighted properties trying to do everything is just silly and feels same-ish.
Ed Beach is a good game designer. He saved Civ5 with BNW
>>338295503
Id love an ai that really tries to win. Even if it employs cheap cheesy tactics. Nowadays they just cheat their ass off and overwhelm you with numbers.
Instead of creating a stupid ai and giving him double resource production, companies should create smart ais and handicap them for lower difficulties.
Machine learning/neural networks game ais when?
