|
Panzeh posted:I've not been that impressed with the results of machine learning, especially in the context of games as complex as Civ. Elias_Maluco posted:Have that been tried yet? I vaguely remember some project was going to try it on Civ 5, but I dont know if it ever got done It's a different game than Civ, but Google's DeepMind AI reached grand master level in Starcraft 2 which puts it in the top 0.2% of players worldwide. It competes slightly neutered as well, since its APM is capped to about 260 while pros can get up towards 500 APM in short bursts. They've also limited it to only see what a human player would see on-screen, so it's not just cheating by watching every part of the map simultaneously. https://www.theverge.com/2019/10/30/20939147/deepmind-google-alphastar-starcraft-2-research-grandmaster-level https://deepmind.com/blog/article/alphastar-mastering-real-time-strategy-game-starcraft-ii
|
#
?
Feb 26, 2020 13:31
|
|
|
|
| # ? Apr 25, 2024 23:42 |
|
|
Some one made an AI that learnt to play Free Civ. http://news.mit.edu/2011/language-from-games-0712
|
|
#
?
Feb 26, 2020 14:24
|
|
|
Are you limited to playing in a specific era when you choose a civ from that era? IE if you choose babylon does the game never proceed to the classical era, an if you choose Carthage does it start (and end) in the classical era?
|
|
#
?
Feb 26, 2020 14:40
|
|
|
No, you can choose to adapt to a new culture as you advance in eras, or keep your old culture for added bonuses. It's diversifying vs reinforcing your existing strengths and weaknesses.
|
|
#
?
Feb 26, 2020 14:47
|
|
|
Are they limited to cultures their original one spawned? Like how Carthage was a colony of the Phoenicians
|
|
#
?
Feb 26, 2020 14:51
|
|
|
No, you don't have to. It's a new set of options each era, though I think that they're first come, first served in terms of what's available.
|
|
#
?
Feb 26, 2020 15:06
|
|
|
Nitrousoxide posted:Are you limited to playing in a specific era when you choose a civ from that era? IE if you choose babylon does the game never proceed to the classical era, an if you choose Carthage does it start (and end) in the classical era? as i understand it there's no fixed civs in humankind, but rather as you progress through each age you pick one period-appropriate IRL civilization to absorb bonuses from. so you can be zhou-phonecian-roman-persian-french-american-moon colony over time or whatever
|
|
#
?
Feb 26, 2020 15:35
|
|
|
https://twitter.com/humankindgame/status/1229809102836183042 I think this got overlooked.
|
|
#
?
Feb 26, 2020 15:47
|
|
|
all these contextless civs do nothing for me
|
|
#
?
Feb 26, 2020 16:14
|
|
|
A great obelisk. Neat. Too bad I have no idea what it is, what it does or why I would get one. It could be an auto-built replacement for a palace that reduces peasant farting pollution by 17% for all we know. Why did they steal the stupid "will you play as x" from the civ 6 civ trailers/ads instead of the actually useful information?
|
|
#
?
Feb 26, 2020 16:25
|
|
|
Tree Bucket posted:Oh, absolutely. It felt really disappointing that Civ 5 had that really interesting intertwined ideology-happiness-culture system, which the AI totally ignored. I always wanted to see bloated AI empires disintegrating under their own weight- it would have been so much fun to engineer!- but it never really happened. It's a really tough balancing act to get an AI that interacts with these kinds of systems - they tend to complicate the standard decision tree-based AI's decision list enormously and can often lead to the AI either over or under prioritizing dealing with this stuff to disastrous results. An example that comes to mind is public order in Total War games. Often the AI is given such ludicrous PO bonuses that it basically ignores the effects of PO, but if you mod the game so the AI loses or reduces those bonuses they never really get anywhere and tend to fall apart due to constant rebellions and such.
|
|
#
?
Feb 26, 2020 16:32
|
|
|
Poil posted:A great obelisk. Neat. Too bad I have no idea what it is, what it does or why I would get one. It could be an auto-built replacement for a palace that reduces peasant farting pollution by 17% for all we know. Because it's a tweet and no one reads on twitter.
|
|
#
?
Feb 26, 2020 16:35
|
|
|
They probably don’t want to get locked into numbers via twitter marketing. Though they could probably suggest why you would build an obelisk in more generalities, but 
|
|
#
?
Feb 26, 2020 16:53
|
|
|
Yeah. Is it a cultural building? Is it a faith building? Is it a prestige building? Is it a government power building of some kind? 
|
|
#
?
Feb 26, 2020 19:50
|
|
|
Kanos posted:It's a really tough balancing act to get an AI that interacts with these kinds of systems - they tend to complicate the standard decision tree-based AI's decision list enormously and can often lead to the AI either over or under prioritizing dealing with this stuff to disastrous results. These days I wish strategy game designers would just give up on the idea of an AI that plays the game the same way as a human. They never get it right and either the AI is a walkover or you have to give it so many bonuses that it stops engaging with some of the game systems. Just jump straight to the outcome you want. Nobody's sad that the demons in Doom don't have to run around the level collecting weapons and health pickups before they can fight Doomguy, they just get what they get and they use it in a straightforward manner.
|
|
#
?
Feb 26, 2020 20:18
|
|
|
You know what happens when you Aksum
|
|
#
?
Feb 26, 2020 21:56
|
|
|
Davincie posted:all these contextless civs do nothing for me Me too. I'm kind of burnt out on this game already. I'd rather they not release any kind of teasers than this. I'm losing trust in it now.
|
|
#
?
Feb 26, 2020 22:00
|
|
|
AI seems to work best in games of assymetrical warefare no matter how complex it is. Perhaps because you don't get the same sense of "unfair advantages". I like Invisible Inc and Into the Breach for this reason, the AI isn't great but because they get wildly different units and even different win conditions, it's just fun to play against. But Civ etc. just aren't that sort of game. I do think, however, that in a 1UPT situation, the advantage that the AI should get is "stronger units" not "more units". More units just makes them worse at the game because of the nature of 1UPT. Civ 5 Vox Populi works on this (though doesn't totally solve it) by giving AI units shitloads of promotions.
|
|
#
?
Feb 26, 2020 22:01
|
|
|
Gort posted:These days I wish strategy game designers would just give up on the idea of an AI that plays the game the same way as a human. They never get it right and either the AI is a walkover or you have to give it so many bonuses that it stops engaging with some of the game systems. That works for many games, and many games should adopt that kind of system. However in a Symmetric Start game that idea is pretty frustrating to a player because like you said, you just give the AI so many bonuses that the game systems don't matter. For "history simulator" sandbox games I guess it's fine, but personally I'd rather a tight set of rules that the AI can actually play by rather then just 40 game systems that are orthogonal to each other.
|
|
#
?
Feb 26, 2020 22:01
|
|
|
The wiki article on AI machine training in games is actually an interesting read. Seems like AI can be trained fairly well for strategy games, but with limitations - e.i. a particular hero set in DOTA against another particular set, or protos vs protos on a particular map. Once you start introducing variation, the training time goes up into thousands of years - though, tbf, you can achieve 130 years of gameplay in a day with a server farm, so it’s all relative.
|
|
#
?
Feb 26, 2020 23:05
|
|
|
Mokotow posted:The wiki article on AI machine training in games is actually an interesting read. Seems like AI can be trained fairly well for strategy games, but with limitations - e.i. a particular hero set in DOTA against another particular set, or protos vs protos on a particular map. Once you start introducing variation, the training time goes up into thousands of years - though, tbf, you can achieve 130 years of gameplay in a day with a server farm, so it’s all relative. Yeah, the number of potential parameters at the beginning of a game, nation combos, map types, map RNG, etc. make it very very hard to do AIs in this regard. Starcraft 2 is a much much easier problem space by comparison. ate poo poo on live tv posted:That works for many games, and many games should adopt that kind of system. However in a Symmetric Start game that idea is pretty frustrating to a player because like you said, you just give the AI so many bonuses that the game systems don't matter. This is me, too, but I don't really think humankind is going for this. I like games that pick a lane and stick with it and, yeah, make something comprehensible by an AI and go- Through the Ages has a pretty decent AI for example, but that comes at a cost.
|
|
#
?
Feb 26, 2020 23:30
|
|
|
Poil posted:Yeah. Is it a cultural building? Is it a faith building? Is it a prestige building? Is it a government power building of some kind? And they've already said none of these are going to be unique buildings or units, so for all we know all these units and districts or buildings or whatever are going to be for everyone and they just get extra bonuses in the hands of their associated culture.
|
|
#
?
Feb 26, 2020 23:37
|
|
|
They are making some really boring choices and I hope to god they start saying what FIDSI everything gives, because the yield names Amplitude likes are the only difference from Civ I can see
|
|
#
?
Feb 27, 2020 02:25
|
|
|
This game looks so pretty I really want it to be good but the amalgamation of civilizations and dynamically changing between them over time makes it seem like they might just all end up being same-ish and indistinguishable from a game play perspective.
|
|
#
?
Feb 27, 2020 02:31
|
|
|
Popete posted:This game looks so pretty I really want it to be good but the amalgamation of civilizations and dynamically changing between them over time makes it seem like they might just all end up being same-ish and indistinguishable from a game play perspective. That is always a problem in game design. If you give the player too many choices, it makes each individual choice less impactful. I could easily see human kind falling into the trap of cultures just being a set of modifiers. Carthage give +5% to Navy's, whereas Phoenicians gives +5% to Army's. Meanwhile, England gives +20% to navy's and Germany give +20% to Industry! By the time you hit the modern era all your cultures are just a list of insignificant modifier's while you are battling with Giant Death Robots that benefit from nothing.
|
|
#
?
Feb 27, 2020 03:22
|
|
|
ate poo poo on live tv posted:That is always a problem in game design. If you give the player too many choices, it makes each individual choice less impactful. I could easily see human kind falling into the trap of cultures just being a set of modifiers. Carthage give +5% to Navy's, whereas Phoenicians gives +5% to Army's. Meanwhile, England gives +20% to navy's and Germany give +20% to Industry! That's the path I see this taking. It's either going to be a bunch of tiny changes that eventually make all civilizations the same by mid game, or they will each have unique abilities. In the case of the later, I'm not looking forward to memorizing or looking up which 9 unique abilities the Germans currently have for just this one game.
|
|
#
?
Feb 27, 2020 03:32
|
|
|
I certainly hope it's unique abilities instead of some modifiers. Or at least modifiers that change the way you play the game (like Rome's "production bonus for buildings already in the capital") Those were always far more interesting in Civ. Yeah i guess there's a risk of ending up with 6 unique abilities by the end game. But if you really have to put effort into memorising 6 things over several hours, things that you've been using throughout anyway, then i think the problem is you  . And who knows, maybe old unique abilities will obsolete as you progress through the game.Edit: sorry i misread the above post. Yeah, if they give each civ several unique abilities i will not be best pleased. Remembering all those in Endless Legend was a pain in the rear end.
|
|
#
?
Feb 27, 2020 08:18
|
|
|
Gort posted:These days I wish strategy game designers would just give up on the idea of an AI that plays the game the same way as a human. They never get it right and either the AI is a walkover or you have to give it so many bonuses that it stops engaging with some of the game systems. The AI getting so many bonuses that it stops engaging with some of the game systems is exactly them giving up on the idea of an AI that plays the game the same way as a human, though. Having the AI simply sidestep around systems that are too complicated for it to consistently deal with in an appropriate manner is like the definition of skipping to the outcome you want. You could spend a gorillion hours figuring out how to get the AI to manage its public order and war weariness effectively for a net result that the player will rarely notice, or you can just say "yeah the AI doesn't have to deal with public order and war weariness". Which is a better use of developer time and processing power? You can have games where the AI is playing a game completely divorced from anything the player is doing, like AI War, but that would probably turn players looking for a traditional 4X off. Kanos fucked around with this message at 09:25 on Feb 27, 2020 |
|
#
?
Feb 27, 2020 09:23
|
|
|
Thats kinda like the Endless Lengend AI, I guess? Ive been playing it, and at one hand the AI seems braindead, is mostly like playing alone. At the other hand, is hard to keep up with it in science/income/production etc, cause it has some ridiculous bonuses Well, either that or I suck at the game
|
|
#
?
Feb 27, 2020 12:07
|
|
|
Kanos posted:The AI getting so many bonuses that it stops engaging with some of the game systems is exactly them giving up on the idea of an AI that plays the game the same way as a human, though. Having the AI simply sidestep around systems that are too complicated for it to consistently deal with in an appropriate manner is like the definition of skipping to the outcome you want. The AI still engages with those systems, even if their bonuses make them immune to them. They'll still build public order buildings, or happiness buildings, or take options that reduce the war weariness their difficulty modifiers insulate them from. Similarly, the player will be able to take options that reduce their public order and happiness, even if those options are basically worthless. By dropping the pretense of "The AI plays the same game as the player", you'd get an AI that plays the game better, acts closer to their design, and a player who doesn't have worthless options to use against them. I usually use the example of a Genghis Khan-type AI opponent. The idea is that in the mid-game, he invades with a hundred thousand screaming horsemen, and the player gets a cool fight. So you give him a unique screaming horseman unit and an extra aggressive AI personality, but what actually happens? He gets bogged down in the tech tree before he can research his unit, and isn't a threat, or he just rushes with his starting units and gets beaten down and never recovers, or he gets caught up on some other system the AI isn't good at dealing with. If instead you just had him appear in the mid-game with a hundred thousand screaming horsemen, you'd actually get the result you were after in the first place, but by pretending he's playing the same game as the human he ceases to work.
|
|
#
?
Feb 27, 2020 12:50
|
|
|
Gort posted:The AI still engages with those systems, even if their bonuses make them immune to them. They'll still build public order buildings, or happiness buildings, or take options that reduce the war weariness their difficulty modifiers insulate them from. Similarly, the player will be able to take options that reduce their public order and happiness, even if those options are basically worthless. Okay yeah but what's the player doing until Genghis Khan shows up? What are the AIs the player is engaging with before that point doing? How are they operating? Calling back to Total War again, it's tradition to have some kind of big mid-lategame threat to tip the player's empire over - in Medieval 2 it was literally the Mongols and then the Timurids, who both invaded out of the east with gigantic super-stacks and generally hosed everything over until you stopped them. In Total Warhammer, in mid-lategame the hordes of Chaos show up with a ton of super stacks and generally raise hell until someone busts their chops. For the rest of the game you're still engaging with AI who are broadly playing by the same rules as the player. The only 4X-style strategy game I've ever played where the game dispensed entirely with the concept of a level playing field was AI War, which ended up feeling very different than something like Civ.
|
|
#
?
Feb 27, 2020 13:18
|
|
|
Personally I dont like that at all. Civ is not Doom, is a grand strategy game. Scripted stuff like that gets repetitive fast It kinda works in CK2 (the mongol and the aztec invasions) because they are few and works like endgame bosses. But most of the game is organic narratives emerging from the game world by themselves, which a lot more interesting and keep the game always fresh Paradox games are criticized a lot here lately, but both CK2 and EU4 are very good in having the AI play kinda the same game that the player does, and make decisions that for the most part make sense, and pose a decent challenge Elias_Maluco fucked around with this message at 13:42 on Feb 27, 2020 |
|
#
?
Feb 27, 2020 13:40
|
|
|
I'm beginning to think that the standard tech company solution is the way to go: don't worry about developing game AI; just have some genius dude in India play the enemy for a few dollars an hour.
|
|
#
?
Feb 27, 2020 13:49
|
|
|
Kanos posted:Okay yeah but what's the player doing until Genghis Khan shows up? What are the AIs the player is engaging with before that point doing? How are they operating? They'd be playing against other AIs who are also not playing the same game as the player, but in other different, specific-to-them ways that are much simpler than the systems the player has to deal with (EG: Maybe AI China starts with its entire empire already settled at game start). Again, the Doom comparison. Doomguy runs out of ammo, an imp does not. Doesn't really hurt gameplay in any way, just means the imp can have very simple AI without being anything other than the opponent it was designed to be.
|
|
#
?
Feb 27, 2020 14:08
|
|
|
Tree Bucket posted:I'm beginning to think that the standard tech company solution is the way to go: don't worry about developing game AI; just have some genius dude in India play the enemy for a few dollars an hour. This will be cloud gaming 2.0
|
|
#
?
Feb 27, 2020 14:27
|
|
|
Kanos posted:You could spend a gorillion hours figuring out how to get the AI to manage its public order and war weariness effectively for a net result that the player will rarely notice, or you can just say "yeah the AI doesn't have to deal with public order and war weariness". Which is a better use of developer time and processing power? OTOH, It means war weariness and public order become nothing more than a malus for the player to manage rather than an actual mechanic they can interact with, and is a big reason for the tremendous shallowness of most 4x games. Take a game like Stellaris as a great example, here - you'd think it would be wise strategically to play in such a way that your war knocks out the breadbasket food generating planets of the opposing civ. Starve them out and create chaos by claiming them as your own and occupying them early in the war. But it doesn't matter! None of it matters. The entire resource management system is decoupled from the actual gameplay so there's no meaningful strategic or creative decisions to make and it just becomes about whose fleet is larger. (or who got lucky with ally placement) Why even have war weariness and public order at all, at that point?
|
|
#
?
Feb 27, 2020 14:29
|
|
|
You could have the AI system respond in meaningful ways to the actions the player takes (more than the AI's choices) by matching results with expectations directly. Abstract away more of the mechanics. To use Civ amenities/happiness as an example, if you're roaming around actively pillaging luxuries and spreading dissent, have it treat this as an attack on a city's loyalty and roll for rebellions etc. But don't kneecap the AI's natural growth/expansion if it failed to settle luxuries early on per core game sim rules unless it seriously did a bad job or got seriously unlucky. This is in contrast to simply having them play "by the same rules!" followed by a huge inherent bonus that invalidates the mechanics and leaves the player with no way to interact with that civ using said mechanics. Another example, handwave food and growth using general rules (plains? desert?). Don't worry about crop carryover or intelligent farm adjacency bonuses or food trade route management for the AI. But if a storm hits or barbarians are burning fields or there's a siege, or you block the trade routes, then have a related result on population etc.
|
|
#
?
Feb 27, 2020 18:03
|
|
|
Elias_Maluco posted:Paradox games are criticized a lot here lately, but both CK2 and EU4 are very good in having the AI play kinda the same game that the player does, and make decisions that for the most part make sense, and pose a decent challenge But the AI in Paradox games don't have to deal with victory conditions (or even goals in the traditional sense), and that's a huge "advantage" to them. It masks a lot of the problems that other, more board game-like titles have to contend with
|
|
#
?
Feb 27, 2020 18:10
|
|
|
WarpDogs posted:But the AI in Paradox games don't have to deal with victory conditions (or even goals in the traditional sense), and that's a huge "advantage" to them. And it still barely and/or doesn't function in most circumstances.
|
|
#
?
Feb 27, 2020 18:11
|
|
|
|
| # ? Apr 25, 2024 23:42 |
|
|
Chomp8645 posted:And it still barely and/or doesn't function in most circumstances. I think it works well enough. In any case, its incomparably better than Civ AI, thats for sure Most of the time they will attack when it makes sense, they will form alliances and make deals that make sense, they are competitive enough etc WarpDogs posted:But the AI in Paradox games don't have to deal with victory conditions (or even goals in the traditional sense), and that's a huge "advantage" to them. It masks a lot of the problems that other, more board game-like titles have to contend with Yeah, thats true
|
|
#
?
Feb 27, 2020 18:16
|
|