Install Steam
login
|
language
简体中文 (Simplified Chinese)
繁體中文 (Traditional Chinese)
日本語 (Japanese)
한국어 (Korean)
ไทย (Thai)
Български (Bulgarian)
Čeština (Czech)
Dansk (Danish)
Deutsch (German)
Español - España (Spanish - Spain)
Español - Latinoamérica (Spanish - Latin America)
Ελληνικά (Greek)
Français (French)
Italiano (Italian)
Bahasa Indonesia (Indonesian)
Magyar (Hungarian)
Nederlands (Dutch)
Norsk (Norwegian)
Polski (Polish)
Português (Portuguese - Portugal)
Português - Brasil (Portuguese - Brazil)
Română (Romanian)
Русский (Russian)
Suomi (Finnish)
Svenska (Swedish)
Türkçe (Turkish)
Tiếng Việt (Vietnamese)
Українська (Ukrainian)
Report a translation problem
Discussions Rules and Guidelines
Report this post
So, how can the AI do better at defending against my steamroller attack. Clearly just throwing units in piecemeal is not going to hack it. The AI needs to embrace the same two principles I use for the attack, namely concentration and mutual support. My attack strategy relies on fires - ie the ability to degrade units and defences with ranged fire from MG units, bombardment from siege, battleship and bomber units and strafing from fighter units.
The best approach for the defenders is to degrade/destroy my fire units. This can be achieved a number of different ways. Defending fire units need to concentrate their fire on one of the attacking stacks till it is forced to withdraw and repair or it is degraded to a point where it can be assaulted by counterattacking tank and inf. A defending force organised on similiar lines to my attacking stacks could be deployed adjacent to the targeted city. Then it is a battle of attrition between opposing fire units. Long range bombers flying from rear bases can attack attacking airbases to neutralise offensive fires.
But for the AI to be successful in defence it needs to maintain a stronger military of the right composition. In all my testing I have yet to see the AI produce enough forces. I suspect the AI is not taking into account that I have ruled out all paths to victory other than domination. It still prioritises cultural and scientific pathways.
Assuming this gets addressed, AI nations need to assess the military capabilities of any potential threat. It's all good and well to have a force of AT units against an enemy heavy in tank and cav, but these AT units won't degrade enemy fire units protected by inf before they themselves are destroyed. A good defence required combined arms like a good offence. More fire units are needed.
The AI needs to assemble and concentrate its forces and move in a coordinated manner. It should aim to form lines of mutually supporting units. It needs to know when to switch from an attrition battle of fires to a breakthrough assault. The trigger for this should be the strength of enemy inf protecting their fire units.
There are a lot more options to canvas. But as lunch is ready here I'll leave this for others to discuss.
https://steamcommunity.com/app/289070/guides/
It is certainly true that the programs the AI civs follow could be tightened up. Yes, absolutely, the situation you describe where you have disabled all victory conditions except domination, could and should have been programmed to make the AI focus pretty ruthlessly on building plenty of units, at least in the late game as you describe. That's a pretty clear and categorical condition -- late game with military techs all unlocked already, and domination the only way to win -- upon which the programmers could write a set of responses for the AI that is better at keeping you from winning than what the AIs do now, continue to devote resources to culture,science,etc.
Of course, before the end-game, neither you nor the AI can at all afford to ignore science, culture, and all the etceteras, because this is a well-designed complex game, and every single etcetera system in the game can be brought together to produce by the end-game any of the victory types. There's absolutely no simple program that categorically prioritizes building military units all through the game that is going to lead the AI anywhere but to disaster, and being even easier for you to beat than it is now. Even in the very late game, unless you create the very special condition of limiting victory to domination only, a reasonable program can't dictate spamming units as the only priority, or some other civ will sneak by with one of the other victory types.
So, sure, the devs failed to put into the program for the AI players a good response for the very late game in which all victories but domination have been disabled. Their program also fails to let the AI players do a good job of simulating strategy in all sorts of other edge cases the game doesn't really focus on. An inherent drawback to all the modes is that the AI players' programs have pretty clearly not been adjusted to make reasonable accommodation for the rules differences the modes introduce, so playing with any of them tends to give the human player an unfair advantage. And some of the civ/leaders you can play as have mechanics that the AI also clearly can't change responses to counteract at all sufficiently.
In general,the idea that an AI in this game should be able to assess your intentions and your capabilities is quite beyond anything any AI in the world right now can do in any context. All we have right now are programs of various complexity and sophistication, and the current Civ6 program is not sophisticated enough to simulate strategic thought at all well in the end game with only domination victory enabled.
If you want a challenge from the AI you have to give it a handicap, because you can do strategy, you can assess a game state and your opponents' intentions and capabilities, and all the AI has is a program. Give it all the Deity advantages, and you will sometimes see both adequate defense and even successful conquest, at least sometimes, especially if you don't luck into a good starting location and haven't dealt yourself an S-tier civ. I suspect you are playing at a lower difficulty, because the conquest method you describe relies on a brute force advantage, in quantity and/or quality of units, and you don't have that luxury against Deity until and unless you have already clawed your way to the top, and your victory by any of the victory types is pretty much assured. Why worry about strategic challenge after you have already effectively won?
If you want a challenge from players who can actually do strategy and therefore don't need a handicap, play against fellow humans.
plaguepenguin,
Thank you for your detailed response. It's most appreciated.
You are right, you can't ignore science and culture etc as they all give your civ benefits direct and indirect for winning a domination victory not to mention any of the other victory types. I'm advocating that the AI should scale these back, not eliminate them, when the sole victory mode is domination.
But I do contest your assertion that "that an AI in this game should be able to assess your intentions and your capabilities is quite beyond anything any AI in the world right now can do in any context".
But first a disclaimer. I am the designer and AI programmer for a WW2 operational wargame called Command Ops. It's sold on Steam. I have been writing AI for this game engine since 1995 - so almost 30 years now. I've also done consulting for the military in this space.
Now, Command Ops (Cmd Ops) does in fact assess intentions and capabilities of the opposing side. It's pretty smart but not perfect. No AI ever will be. But it is possible to write an AI that does assess. I contend that Civ VI does that already, albeit at a very basic level. It clearly assesses when enemy forces move close to their borders and reacts. It send the human player a back-off message and if the border is breached it sends its troops to face them.
What I am asking for is a little more sophistication. I'm not calling for it to be as detailed as Cmd Ops is but I would like to see it better.
As a gamer I hate the practice of allowing the AI to cheat by gaining perfect intel on the enemy and I loathe lazy AI that require advantages like improved resource rates and yields. Good AI should eschew these practices. But for that to happen the AI needs to be good enough to do a reasonable job of making decisions - not perfect decisions just reasonable ones.
Well, in those two cases there is a easily defined set of parameters in the overall game state that can be used to trigger a change to different AI behaviors that unequivocally meet the AIs' need to keep you from winning.
What's the game state that would prompt ignoring all the etceteras to focus on producing units to the exclusion of all else? The only relatively hard input factor to the decision to stop building more campuses and doing more research grants, would be that the AI has unlocked the last GDR improvement. For one thing, that's way late. The game is almost always already effectively won or hopelessly lost by then. And even if you and the AI find yourselves in one of those one-in-a-million games where it's still on a knife's edge after you both have unlocked fully promoted GDRs, there's still possibly all sorts of economic/development game considerations that might be at least as important as the pure wargame, even at this late point in the game. Who's got, and can retain control of,more uranium? How important are naval units at securing uranium supplies? What other units might be more or less needful in neutralizing or bypassing the enemy GDRs? What other civs than your main opponent can and should be invaded, to secure uranium, aluminum, or just more productive capacity? What other civs help you more by remaining allies or at least trading partners?
That's just the simplest case, the rare end-game case in which the game is still at all close. For all games, the only victory type that should influence your decisions, at all, until the mid-game at the earliest, is religious victory. For that one victory type only, you have to get a religion, and the competition for the limited number of religions means you can't count on getting your religion unless you act very early and very aggressively to that end. Other than that, before the mid-game at the earliest, every victory type dictates that you expand as much as possible as your only priority. A consequence of that is that you move up the tech and (less urgently) the civics tree, because moving up the trees unlocks every good thing in the game, from more productive tiles to mor distircts and their buildings, to better units,to better policy cards, etcetera, etecetera.
Early conquest is a wonderful help for any type of victory, and not especially for domination, because it is one of the modes of expansion. The fact that only domination victory is enabled should not, even a little bit, change what the AI should be doing,until we reach that endgame state as discussed above. Having the AI prioritize building units before the end-game would only make it easier for humans to beat the AI. Military units are only helpful to winning if produced just when they are needed to conquer another civ, or prevent conquest by another civ. Just exactly when x number of y unit types will make it possible to get a conquest done in z turns, before your victim can respond with an effective defense, is a constantly moving target as the game progresses.
There isn't even any one game state in the late game you can specify for the switch in the program away from the eteceteras to spamming units. The human player actually has to be attentive to the possibility of such game states all through the game, because they arise many times during the game, but each time the opportunity passes if not seized in time. At lower difficulty it is easier to go for a very early conquest (conquest before walls), or an early conquest (before crossbows), but you don't need to, because humans are better at economic.development strategy, and if the AI is not given a handicap, you can count on getting ahead in tech and production by the Medieval or Renaissance era, then building an army on the lines you advocate, something that can predictably and safely deal with all defenses. Higher difficulty removes that safe play, because however smarter you are at development, you reach a point by Deity at which you can't stay ahead of the AI by superior development strategy alone, and you have no choice but to risk conquest as early as that's not clearly suicidal. You have to be attentive and open to bringing in all sorts of side-hustles to succeed at this, all sorts of mechanics in the game that aren't core.
No doubt you can write a program that lets an AI do a reasonable job of simulating actual decision-making based on the enemy's intentions and capabilities within a space that only has one dimension of strategy. I can see a pure wargame making that possible. As a pure wargame, Civ 6's AI doesn't do a horrible job, at least if you keep the level down to pure wargame and a limited space. It's competent at counter-punching within one theater of war, and I learned the "tactics" to use in this game from the AI when I was just starting out. It gets worse and worse the wider the sphere. Just spatially, it can't manage allocating forces between different theaters. It can't assess the level of risk of an actual invasion presented by all the civs that have declared war on it (as opposed to the many purely notional wars the AI civs often declare on their own, or that you can get them to join in on). Clausewitzian means and ends in deciding to go to war? Forget about it. The idea that the AI could analyze the total game state to the point of being able to identify when to spam units vs when to continue development to get better and more units in the future -- hopeless. Okay, totally hopeless until the end-game,and even then pretty marginal. I think you have to keep intentions and capabilities really limited and one-dimensional for AI at present to simulate humans' use of strategy,at least in a very open-ended game.
Of course I might be wrong. i will try out some of the games in the Command Ops series. i started out in gaming as a 6 year-old on the 1961 version of Avalon Hill's Gettysburg, and pure wargames are what I played for a long time, a long time ago. We didn't have AI at all back then, of course, just older brothers until at least high school and you could find other weirdos outside your family who liked wargames. I'll see how your AI does compared to my older brothers.
As a final thought, to really get this wide open game to cone down to only playing as a pure wargame, you have to to do much more radical surgery than just shutting down all victory types except domination. Take a look at the War Machine scenario. It only covers 40 turns (each one a week if I recall), cuts out all the etcetera systems except science, but then only has 6 researchable techs, all of which only do buffs rather than new units (aside from one that unlocks biplanes). You win by capturing or defending Paris. Sadly, the devs made it MP only, no doubt because they would have had to write an entirely new program for the AI to allow single player. Never having done any modding, I don't know if it's possible for a modder to do the AI rewrite, but I would definitely be interested in playing the scenario if anyone ever undertook that task. I suspect that this AI for War Machine would look a lot like your above suggestions for offense and defense,which seem to be quite sound for a version of Civ that tossed out everything but its wargame dimension.
I just don't think that any version of Civ that preserved its many dimensions could allow a program to simulate at all well strategic analysis of intentions and capabilities. The full game makes the whole competitor civ the object of that analysis, not just enemy units. If it's just a wargame, and it's just the enemy units the program has to come up with a response to, that's one thing, but if its units are just one manifestation of a civ's total threat package, I don't see it. Could the AI be tweaked to counterpunch better? Sure. But I don't see it made able to respond decisively better to the totality of the threat a rival civ poses. You have to really simplify the threat by removing all the mechanics except just the wargame to come up with a tweak that will improve its chances dramatically, but that simplified game is not this game