1. We have added a Gift Upgrades feature that allows you to gift an account upgrade to another member, just in time for the holiday season. You can see the gift option when going to the Account Upgrades screen, or on any user profile screen.
    Dismiss Notice

So apparently, the AI has not been improved

Discussion in 'Civ6 - General Discussions' started by Kruos, Feb 12, 2019.

  1. Gorbles

    Gorbles Load Balanced

    Joined:
    Nov 24, 2014
    Messages:
    5,583
    Location:
    UK
    Something being noticed in QA, or being recognised as undesirable behaviour, doesn't automatically translate into the work hours required to address it. A bug being in the game doesn't mean QA missed it - this is one of the most misleading claims made about QA in general.
     
  2. Ferocitus

    Ferocitus Deity

    Joined:
    Aug 7, 2016
    Messages:
    4,007
    Gender:
    Male
    Location:
    Adelaide, South Australia
    It's those kinds of anomalous behaviours that will be valuable guides for modders.
    However, that doesn't mean that a simple tweak, or weighting adjustment, will solve the problem, or do so without a raft of unintended consequences.

    Some parts of previous mods like VP will be useful, but only the most AI-naive person could think that's all that's required.
    It's the equivalent of thinking that you can push an AIturbo button and the game will magically behave as desired.

    Look at the history of Civ5 or Civ6 mods and you will see how many iterations they went through before all the wrinkles were ironed out. What seemed correct when the mod was released and, in many cases, what looked like perfectly reasonable improvements at first glance, were very often exposed as buggy, or inappropriate when combined with other game features.

    I know the AI is deficient in many aspects, but I can bypass those annoyances by playing on large (modded) maps against 20+ civs so that the AI's mistakes are not completely ruinous to my enjoyment of the game.
    BTW, just so you don't think I'm a shill - I wouldn't touch the standard releases of Civ6 or R&F with a bargepole if it weren't for the Terra Maps, and even those splendid attempts needed a few iterations to eliminate some chronic faults.
     
  3. ashendashin

    ashendashin King

    Joined:
    Sep 19, 2016
    Messages:
    985
    I'm not gonna argue against the fact that you can't just copy VP's code into Civ6 and expect it to work(technically it's the community patch dll which VP is built on top of), but I will say that VP is far more complex than vanilla Civ5. Civ6's design may be AI unfriendly, but so was 5's, despite the relative simplicity.

    Also wish this was just merged with the previous AI thread rather than this talk about a subforum. These threads come and go, so just leave one at a time. Since they pretty much always have AI in the title I don't see why people view the mere presence of a consistent topic with such vitriol.
     
    TheBigFundamental likes this.
  4. Casworon

    Casworon Prince

    Joined:
    Aug 9, 2014
    Messages:
    527
    I do think the more interesting discussion which is connected to the AI discussion is whether we think they will release the DLL or not.

    Has it been mentioned by Firaxis at all recently? I think i remember Ed Beach saying they will look into it but no promises. But this was years ago
     
    TheBigFundamental and oedali like this.
  5. Gorbles

    Gorbles Load Balanced

    Joined:
    Nov 24, 2014
    Messages:
    5,583
    Location:
    UK
    Because it's sometimes more important to correct mistaken assumptions than let them continue on @ashendashin. Regardless of how heated things get (and let's not phrase it like there's only vitriol on a particular side - we're human, everybody gets frustrated sometimes).

    For example, it's not that Civ VI's design is AI-unfriendly. It's that the rules you take for granted in CiV may either be modified in Civ. VI or completely different and replaced with new rulesets or behaviour. It's not a matter of the game being friendly or not (this is entirely down to what the game allows you to access, and how you can modify that within the scope available. Gedemon did a good breakdown recently when I asked them for limitations in modifying current behaviour), nor how complex VP is (full credit to anyone involved in it). It's more that the same principles aren't guaranteed to work to the same extent in Civ. VI, because things will operate in different ways and interact in different ways on top of that. It would be building a new system; a VP 2.0.

    I think @Ferocitus's point is that this would take far more time (and effort) than people seem to think. Particularly when a lot of the complaints about the vanilla AI is "well Vox Populi did better, so why can't they". I could be reading them wrong of course. AI has never been something I've looked closely at, in modding or general development.
     
  6. ashendashin

    ashendashin King

    Joined:
    Sep 19, 2016
    Messages:
    985
    I looked at his point on complexity from a game design perspective. If he was strictly referring to the programming aspect on all points, I already conceded that copying code isn't so easy. 'cause yeah I'd never go back to vanilla civ even with the Community Patch which contains all the AI improvements without what's pretty much a conversion. AI improvements alone wouldn't save Civ6 for me, but just having FXS consider the AI when implementing mechanics would help the series immensely IMO.
     
    TheBigFundamental likes this.
  7. Elhoim

    Elhoim Iron Tower Studio Dev

    Joined:
    Nov 3, 2004
    Messages:
    2,524
    Gender:
    Male
    Location:
    San Isidro, Argentina.
    Yes, I didn't wish to imply that the VP code would be transferable, it wouldn't be at all. But the main point was that it's possible to have a better tactical AI with a dedicated person/team focusing on it, without the need of a supercomputer or anything. Sure, the code would have to be done from scratch, but it's not an issue of "we don't have the technology yet for a good tactical Civ AI" (emphasis on good, not perfect nor unbeatable).
     
    TheBigFundamental likes this.
  8. Ferocitus

    Ferocitus Deity

    Joined:
    Aug 7, 2016
    Messages:
    4,007
    Gender:
    Male
    Location:
    Adelaide, South Australia
    Thanks. Just to be clear: I didn't mean development time, but the execution time on platforms that are commonly available.
    We could have a much better AI if more selfish people were prepared to install a few dozen good graphics cards. But then they'd just complain about their power bills and the effect on the "meta environment". :)

    Of course it's possible to improve the tactical side. There's no shortage of ideas there.
    How much performance are you prepared to sacrifice to improve that single aspect?
    With limited available RAM, you might also have to sacrifice features that other players value more than the tactical improvements you crave. Good luck knowing what other players value more and accommodating their priorities.
    It has to be done via mods IMO.
     
    Last edited by a moderator: Feb 13, 2019
  9. Elhoim

    Elhoim Iron Tower Studio Dev

    Joined:
    Nov 3, 2004
    Messages:
    2,524
    Gender:
    Male
    Location:
    San Isidro, Argentina.
    Definitely. Personally, for Civ VII I would go for a more AI friendly set-up than a huge board with thousands of tiles and dozens of units to manage. My preferred idea would be the introductions of armies, which contain several units, but move as a single one with a bigger ZOC and engagement radius. When the armies clash, you play the combat in an area delimited by 10 tiles from the engagement point (yes, like Endless Legend but you control the units like in Civ). Each side arranges their units and combat plays for 2-3 turns. This setup is more AI friendly as you can teach the AI to build different stacks made of mixed units, and teach it basic formations (melee in the front, ranged in the back, etc). Plus better performance as you would have much less units moving around. AI vs AI combat can be autoresolved. It's quite a big change, but I think it would definitely improve the AI on the tactical front without much performance cost.

    It's funny to have all these AI talks since that's what we've been working on these past days on our new game. Being a turn-based RPG with a more limited amount of combatants (5-6 enemies at most) makes it much easier to have an AI that performs well and doesn't make obvious blunders like ending a turn out of cover (https://store.steampowered.com/app/648410/Colony_Ship_A_PostEarth_Role_Playing_Game/ <- shameless plug if you want to wishlist it :p).
     
    Alexadamz and Trav'ling Canuck like this.
  10. anandus

    anandus Errorist

    Joined:
    Oct 27, 2005
    Messages:
    3,859
    Location:
    Amsterdam, Netherlands
    It does seem like a lot of work and hardly doable if you have, say, a 60 second or 90 second turn timer. Especially if you have more than one fight on your hands.
     
  11. Elhoim

    Elhoim Iron Tower Studio Dev

    Joined:
    Nov 3, 2004
    Messages:
    2,524
    Gender:
    Male
    Location:
    San Isidro, Argentina.
    Yes, I agree it's a system definitely more geared towards single player than multiplayer.
     
  12. Krajzen

    Krajzen Deity

    Joined:
    Oct 23, 2013
    Messages:
    2,921
    Location:
    Poland
    I'd quite possibly enjoy that if it was done well, but it can backfire in a lot of ways and when I think of it I'm not sure if it could be done in civ format. You have to optimize size of battlefields, stacks, size of armies possible for average civ (some sort ouf unit limit), solve the issue of several stacks clashing at once, units of several parties clashing in one place (such as civ - civ and city state ally - barbarians)... Also, Endless Legend doesn't have aircraft or navy and they'd have to be integrated as well...
    How would that system do modern era battle of four civs in one land-and-sea area, involving aicraft, navies, land units and embarked units, land-ship combat etc?
    How big tactical range of battle would have to be allowed to make mobile units useful?
    It's a lot of issues that would end up changing half of the game systems to accomodate such combat in civ game.

    On the other hand I am slightly dissatisfied with the current combat/unit mechanics andwould like to change them in some way.

    I dislike how unwieldly current system is for AI for logistics micromanagement of many units and presumably for performance; also, while it simulates 20th century era "fronline" conflicts well, it still seems awkward for me regarding early era open field battles.

    I'd simply tweak 1UPT system in some ways. Personally I'd like less movement-punishing terrain and less units overall on the map.
    Honestly, I have my theory that ranged units may be the main problem for AI.
    I think it'd be significantly easier for AI to grasp the system if all units (except aircraft) had to be adjacent to another to attack
    , and "ranged" units either wouldn't exist or were ones with no enemy retaliation. Imagine how many problems it'd solve for AI regarding unit movement, positioning, rough terrain, unit fragility, army composition, city sieges etc.

    I could sacrifice existence of 2-3-tile-shooting units for much more dangerous AI. It could also solve balance problem of ranged units always being dangerously stronger than others and immersion issues such as eternal problem of "archers archetype" breaking in gunpowder era, gunpowder units weirdly being melee, weird modern ranged units such as machine guns (still totally capable vs tanks) etc.
     
    Last edited: Feb 13, 2019
    Alexadamz likes this.
  13. wilcoxchar

    wilcoxchar Warlord

    Joined:
    Oct 29, 2005
    Messages:
    150
    Meh. I play SP but I'm not a minmaxer so don't care, the AI's fine.
     
    pgm123 likes this.
  14. steveg700

    steveg700 Deity

    Joined:
    Feb 9, 2012
    Messages:
    3,551
    To no one in particular I posit that if you don't care about something, then you probably don't set a high standard for it. Then again, if you really don't care, why come into a thread just to declare your apathy?

    To no one in particular I would also posit that if you identify with a game to the point of reacting to criticism of it as if it were an attack against a family member, then you lack objectivity. "Beating a dead horse" is just a way to deflect criticism when all rebuttals are exhausted but you don't want to concede the point, as if there's some imaginary window of time in which a person gets to complain about something, and if they persist after that window closes....well, roll your eyes at them. They should just give up and stop harshing the buzz. And if you want to complain further, we'll consign you to some subforum ghetto. That's the politics of overidentification.

    Personally, I think it's perfectly reasonable to expect the AI to make fair use of the tools at hand, and if it can't it begs the question of what that tool is for. I think it's fair to say the tactical AI is lousy and that the lack of noticeable improvement means it's not receiving developer priority. I don't think the AI ever can make effective use of 1UPT, given that there simply aren't enough hexes on any given map (and that calls it into question as if it should be there). I also don't think it can ever make effective use of combined arms, given that it beelines in the tech tree in a fashion that doesn't allow for effective unit diversification.

    But the thing about the AI not using air units is a real puzzler. It's a way for the AI to get around the IUPT quagmire. It's all fire-and-forget. What's up with that?
     
    Last edited: Feb 13, 2019
    Prima Italia, Krajzen and Aristos like this.
  15. Gorbles

    Gorbles Load Balanced

    Joined:
    Nov 24, 2014
    Messages:
    5,583
    Location:
    UK
    Ah yeah of course, that's a good point. Turn-by-turn performance is an important metric that people forget (especially when comparing with CiV, don't people recall how bad that can get?).

    Oh, in either case, my points apply. Designing a base from which to work with is just as important as programming on top of that base - especially when it comes to defining a framework for a mod to exist within (depending on the game, there could be some structure you can borrow / sit within, or you have to devise it completely from scratch). I have a decently-long background in games modding, and the answer, again, isn't "have the developer consider X when developing Y". Development is done (usually) at a breakneck pace with harsh deadlines. Your concerns are a) stability, b) getting it done, c) not causing regression issues and d) maybe performance.

    To expand on my agreeing with Ferocitus (and why I put his quote first), in any game (real-time or turn-based) there's normally a window for executing the AI. A certain amount of time that can be spent processing the AI without impacting on anything else being done in that time slice. Increasing the complexity increases the time spent, but obviously, that window doesn't shift. Making a change in the design, there's an immediate impact on performance once-implemented. Sometimes it can be a positive effect on performance, but not usually.

    I mean, if the AI in the Gathering Storm can use the new features (World Congress, etc), Firaxis obviously are considering the AI when implementing mechanics. The existence of bugs - however severe - doesn't mean that they're not. See my original post where I point out that just because there are issues, that this doesn't mean Firaxis are okay with the status quo. Increasing consideration for the AI during development would have a knock-on effect elsewhere. Ideally, we'd get it without the knock-on effect, but hey, business, etc. I don't like that side of the equation either (as a software developer), but it exists.
     
  16. PhilBowles

    PhilBowles Deity

    Joined:
    Nov 20, 2011
    Messages:
    5,323
    Those of us who aren't familiar with the coding side aren't using the phrase "AI-unfriendly" to refer to difficulty in modifying the AI's coding, but to the degree of complexity the AI is required to deal with. A system is AI-unfriendly if it offers a large number of decision points where the opportunity exists for the AI to make suboptimal decisions; if it is so context-dependent than 'one-size-fits-all' coding algorithms can't reliably produce optimal results; or if the impact of poor decisions on gameplay success is disproportionate.

    For instance, Civ IV stacking was much more AI-friendly than the systems in Civ V and VI: stack size mattered more than unit composition, so if the AI had the wrong mix of units it was not heavily punished for it, while in Civ V and VI having the wrong unit type for whatever its opponent is using strongly disadvantages it. The ability to move multiple units at once and the fact that units did not obstruct one another made pathing decisions less complex. Everything in the stack attacked at once, with no opportunity for the AI to make suboptimal targeting decisions with multiple units, and there was no ranged combat system that demanded different rules to follow or specific sequencing to ensure the archers fired first.

    On this basis, Civ VI is exceptionally AI-unfriendly. It is crammed with decision points - the AI should be favoured by the fact that in almost all non-military cases, individual decisions are of very low significance and even in aggregate often don't substantially affect game outcomes, but that's assuming a human opponent who isn't optimising their own play or exploiting game systems in ways the AI cannot.
     
    Gorbles and Aristos like this.
  17. ashendashin

    ashendashin King

    Joined:
    Sep 19, 2016
    Messages:
    985
    I'm going to respond to this specifically because I don't want to get into the performance aspect(I highly doubt the AI is optimized).
    A knock-on effect is the idea. In all likelyhood approaching any area of Civ with the idea of fixing it is going to take time because the series never got much polish in the first place. Personally, I do not care for Firaxis to make a game as polished as VP themselves, but that's only with mods in mind.
    I wouldn't even bother reading these threads if the DLL had a set release date. However, that doesn't seem to be happening, and the state of Civ5s AI was so poor that I have trouble accepting excuses for it in Civ6, since the gameplay is overall similar.
    Of course, it is normal to ship out barebones AI across the entire game industry. Nothing about these AI threads is unique to Firaxis. If anything they set bad priorities more than they screwed up any one area.
    If people don't like the game because of the AI, I don't see how the producer's side matters. They set their priorities and fell short. If people want a game with good AI, that's a desire derived from the design. FXS/2K could always switch things up if they want to cater to a different audience with all the complaints that follow. Rushing these hack jobs ain't gonna shut down discussion/whining if the game still caters to those interested in higher quality.

    I'll make it clear that I do not hate any individual developers. I don't even know their damn names. The product doesn't satisfy and as such, the company behind it doesn't either.
     
    TheBigFundamental likes this.
  18. Gorbles

    Gorbles Load Balanced

    Joined:
    Nov 24, 2014
    Messages:
    5,583
    Location:
    UK
    I appreciate the breakdown here, and that makes complete sense.

    I'm not sure how to resolve this, without designing the game for decisions that the AI can process more effectively. I don't want to say "dumbing down", because that's a bit of an overused and misused phrase (or get into 1UPT vs. MUPT, that rather endless discussion), but you are talking about complexity in design. Both "soft" decision making (weighting thing that are hard to balance with raw numbers, that require human guesswork in some cases) and also compositional balance with units (and things like even City-State delegates, though they're a bit more straightforward) for the most effective force (which also gets messed up on the regular; a decent composition against one other player or AI could be completely invalidated by another player or AI, wrecking the AI's current set of decisions, which is probably where most of the breakdown in military AI logic happens).

    Do you have any ideas, even completely idealistic personal ones? I mean, other than "further investment in AI". Personally, I'd like to see that. I'd like to see more titles get the polish they deserve in general. But the industry doesn't seem to cater to that kind of setup, at least at the moment.

    By saying "I highly doubt the AI is optimised", you're getting into the performance aspect :p

    The whole point of what I and others have been talking about is that the gameplay being "similar" doesn't mean that the AI is anywhere nearly so similar. That's a layman's opinion, it's basically saying "I perceive these things to be similar, therefore the related systems must be as well". You're free to it, but it is what it is.

    I mean, what are you trying to argue at this point? Your ability to discuss these kinds of topics? Like I said at the start, you're free to it. But people are free to reply, as much as they disagree, however they disagree. That's kinda the point. You're not being shut down nomatter how Firaxis address the situation, and nomatter how people reply to you.
     
  19. ashendashin

    ashendashin King

    Joined:
    Sep 19, 2016
    Messages:
    985
    Yup I admit I'm baiting the performance part when I shouldn't. I'd have to do some parroting...

    Again, I concede that the underlying systems aren't necessarily the same. It ties into my desire for a knock-on effect from increased AI prioritization in the absence of the possibility of a VP-like mod, since rushing out hack jobs could easily lead to entirely different systems but similar gameplay.
    I was arguing my lack of interest in supporting the company's side of the argument more than anything. But I do find it distasteful how often that side is taken.
     
    Last edited: Feb 13, 2019
  20. Miravlix

    Miravlix King

    Joined:
    Jun 28, 2012
    Messages:
    645
    Civ seems to have turned into some weird Sandbox, there is no game in it anymore.

    I play for 30-50 turns, look at my position and feel like I've won already and there is nothing left to do than hit next turn until the game is won.

    I love sandboxes, like Factorio, Oxygen Not Included and many others, they have zero game in them, it's more like taking AutoCad and pretending you are an architect and trying to build something and I love that, but Civ is a really horrible sandbox, so I for one need some kind of game, where I compete against the AI, but the AI is nothing but a random number generator with mostly no weighting towards optimal play.

    I just don't get how anyone can enjoy the game, it takes like a handful of playthroughs to learn the mechanics and after that it's just the same and the same and the same, because the AI never does anything interesting.

    I've watched the press release players, but I'm not impressed, it doesn't seem like any of them is enjoying playing and no one seems even remotely interested in the ambiance of the game, it's just watching someone trying to click next turn as fast as possible with no appreciation of the game. I know they most likely make money from the videos and I think that is the only reason they bother.
     

Share This Page