Why Sid Meier gave up on stacking units?

[Fippy]

For the same reasons they (everyone) do changes these days..they target a different audience.
Convincing serious gamers (in lack of a better word, can be difficult for me in english) has very little priority now, they want $$$ from peoples who feed themselves on social media and via Influencers. Those players prolly complained about the oh so evil doomstacks, or there was some other reason why they think 1upt attracts them more.

 

[TheMeInTeam]

Let me also be slightly cynical and suggest that the newer iterations of Civ are also designed so that you don't expand a lot, like you could in previous games. They don't want you to conquer half the planet by 1500 A.D. They don't want to improve the "AI" algorithms to deal with human expansion, so instead they've put mechanisms in place to limit that somewhat.

Civ 5 introduced a penalty for expansion, but otherwise this isn't really true (never has been in Civ 6). You can, simply by using a great general or two and good unit movement/attack choices, simply war-slap your way through deity in Civ 6 by killing everything and outscaling it.

As for AI, a major barrier to making it good is the constant introduction of new civilizations, new mechanics (major impact on AI), balance changes, and basic game design philosophy. By that last one, I mean that devs try to make the game something it isn't mechanically, but having the AIs (partially) role play...something borne of necessity when you shoehorn in multiple "victory conditions" while the overwhelming majority of mechanics/interactions in the game incentivize tech + military. Tech + military advantage will trump an advantage in other things. Hard to do much when your civ isn't on the game board any longer (failed the test of time).

So you're not just writing an AI (or using machine learning, though I haven't heard of a game ship where the dev team itself did that for its AI yet) for one set of rules and calling it a day. There's also the caveat that if you were to go full-bore and create something like AlphaStar, literally nobody could beat the AI. We'd then have a reverse of strategy games made to this point; players would have to have bonuses, just for the game to be winnable.

Those players prolly complained about the oh so evil doomstacks, or there was some other reason why they think 1upt attracts them more.

I wonder about the real motivations for retaining 1UPT. Though we've had 1UPT, stack size limits, and megastacks (with mechanical disincentives) all since at least the 1990s, in various games. None of these were new concepts, even at Civ 4 release 1UPT and stack limit had been around 10+ years elsewhere. Do casual players really prefer 1UPT? I suspect other reasons (UI incompetence, bad choices like how animations are handled, or who knows what else).

Dominions is probably the best turn-based game for "serious" play I've seen, though its AI is also quite bad and the vast majority of the value there is in MP. That AI is definitely still harder than Civ 5 or 6 deity on max settings (not a high bar), but I don't know about Civ 4.

Edit:

Also worth pointing out that Sid has not been lead designer in Civ games for years, so it's not really him that chose to move away from stack combat. He probably didn't push back strongly against it, since he's been pretty hands off with all of the Civ design successors from what I've heard. But I haven't heard that he particularly endorsed it either?

 

[innonimatu]

becauseeee it's not what the cool kids suppose to do. Stacking is something that's popularized by Civilization, Paradox perfected and balance it with attrition system, but Civ abandon it entirely try to make work on 1UPT system, plus the hexagonal mumble jumble, instead of perfecting what is already work, they create a new problem instead.

I don't know. They should keep the good old feature while adopting the progress of the new, but here the decision is just not satisfying at all. If tend right, the product quality should add up, not dumb down.

The most stupid thing with that move in civ V that they could instead have gone back to the old system of having the stronger unit defend the stack, and the whole stack destroyed if that unit failed. Which would be more realistic - historically, whole armies got slaughtered after they broke and fled, not in the battle itself. Better yet, give a few of the units on the stack some odds of withdrawing with damage, all the ones that fail the roll to withdraw get killed.

 

[H4run]

The most stupid thing with that move in civ V that they could instead have gone back to the old system of having the stronger unit defend the stack, and the whole stack destroyed if that unit failed. Which would be more realistic - historically, whole armies got slaughtered after they broke and fled, not in the battle itself. Better yet, give a few of the units on the stack some odds of withdrawing with damage, all the ones that fail the roll to withdraw get killed.

Civ already have variety of terrain with variety of modifiers (defense bonus, health, production, food, etc) why not add another modifier that dictate attrition and the effectiveness of the stack? Makes it relative also within the national border, allies border, neutra and enemy border, relate that with military ration that would impact overall unit health penalty per turn, combat effectiveness or even unit health that again can impact to health. Sound complex, but it can set to prioritize and simplified, but the main idea is that.

 

[GinandTonic]

The problem is that after it's free units the AI doesn't really build troops. After the units become more expensive the AI is incapable of building troops. Admittedly I've gone from Immortal/ occasional deity to deity over lockdown, but I've not seen an AI tank. I've not seen an AI infantry. I've not seen a AI ironclad or better.

It's not that the AI doesn't know how to use its army that is the problem - it's that it doesn't, ever, have an army.

 

[warpus]

There's also the caveat that if you were to go full-bore and create something like AlphaStar, literally nobody could beat the AI.

Ideally there would be some sort of AI behind the computer decisions though. Right now we are playing against a grandfather clock, pretty much. It relies on overwhelming you instead of out-strategizing you. Which in a strategy game is not a great approach. It works for new players, since they don't notice it until they gain a bit of experience, but it is quite noticable for those who have put enough time into the game.

I don't deny that programming solid AI is hard. It's pretty damn hard. My only complaint is that Civ devs aren't even trying. i.e. changing difficulty simply alters the bonuses & handicaps. That's what I mean by "not even trying".

And surely if you end up developing an AI that is able to beat humans each and every time.. you should be able to dumb it down for different difficulty levels.. Not by using handicaps and bonuses but by more effective means. See Chess. If you increase the difficulty your opponent doesn't start the game with 2 or 3 queens. It approaches the game differently.

 

[Lohrenswald]

Have only skimmed the thread tbh:

1. getting rid of 1upt wasn't Sid Meier's idea, I don't think he really worked on games beyond the first one. I even saw a talk he had about it where he briefly said he thought 1upt was bad design

2. Personally in my experience I've found 1upt combat, including ranged units, a bit more fun, but it's offset by the extreme tedium of trying to manage the units moving around the map, say pulling them back after a war or trying to redo the formation. So in the end I prefer stacks

don't know if I'll ever play civ again though tbh. I kinda just don't care anymore, and have become offput by strategy games in general

 

[H4run]

2. Personally in my experience I've found 1upt combat, including ranged units, a bit more fun, but it's offset by the extreme tedium of trying to manage the units moving around the map, say pulling them back after a war or trying to redo the formation. So in the end I prefer stacks

This 1upt, elevate me from a lowly Prince/King scum in civ 1, 2 and 4, to a powerful immortal. I just simply blast all my inferior enemies military skill till my units are overpromoted and unstoppable, wipe the map clean. Then I got bored.

 

[aelf]

Personally, I've thought a lot about capping the amount of units per tile, and in my opinion it's too arbitrary. Units in Civ are already an abstraction; a bit like unit counters in Risk (a board game, for the uninitiated). One hex of Swordsmen doesn't represent literally seven Swordsmen or however many it is. It represents a legion's worth, or however many is a fair amount to consider attacking a whole city. Capping it at n units isn't organic - it's saying "well okay 1 is too limiting, but infinite is too exploitable, so let's shove an arbitrary constraint that requires constant adjustment to work out". But we already have unit stats for that. So the problem is chiefly logistics, and that's why I really appreciate moving religious combat to its own layer (for example).

I've mentioned this somewhere, but it could have been a great opportunity to introduce combined arms. A tile could have 1 melee, 1 ranged, 1 siege and 1 support unit. So you can make a stack, but not of the same type of units, which would encourage players to diversify their builds.

It's kind of made the AI unable to pose any threat after the early game. In earlier iterations it was optimal to keep an eye on the sea, lest a doomstack appear from the fog. Now it's a joke. The AI is unable to launch any sort of offensive after the first sixty turns, perhaps 80 with a medieval UU they will actually build.

At deity, the AI is irrelevant once it's opening advantages are nullified. It cant cope with building districts, upgrading units or 1UPT. It just gets worse over time.

After a couple of thousand hours, I'm really ready for a sequel with functioning AI.

This. The AI is not coping well with 1 UPT, in my experience. It's only really harder for players when they get unlucky with the terrain.

 

[TheMeInTeam]

And surely if you end up developing an AI that is able to beat humans each and every time.. you should be able to dumb it down for different difficulty levels.

That doesn't follow. The kinds of AIs that actually beat humans with incomplete information like this are trained over millions of iterations playing the games by/against itself to get into a state where it makes the best decisions given the information/rules of the game it trained on. It doesn't think like a human or have some modular approach to being great at the game. You could pick earlier/weaker agents, but you'd be looking through millions of copies of mostly-similar things just to probably wind up with a still mostly-unbeatable AI with some obscure flaw, or going earlier something even worse than what we have now.

Simply the process of analyzing a ton of AI copies would get untenable/time consuming. Even more so if you wanted something that a) humans could beat and b) doesn't wind up being some garbled meme of a thing just to satisfy a).

You could brute force it with bonuses like we do right now, of course. And I suspect it would be similarly as engaging to win that way as now. But maybe I'm wrong, and the "how few bonuses can I get away with and still beat a literal machine optimizer" could be a more interesting question than "how many bonuses does this pile of junk need before I struggle to beat it?"

I've mentioned this somewhere, but it could have been a great opportunity to introduce combined arms. A tile could have 1 melee, 1 ranged, 1 siege and 1 support unit. So you can make a stack, but not of the same type of units, which would encourage players to diversify their builds.

I'm not sure "put the same combination of units in literally every tile you use for combat" would make the game's decisions more diverse, on average.

 

[warpus]

That doesn't follow. The kinds of AIs that actually beat humans with incomplete information like this are trained over millions of iterations playing the games by/against itself to get into a state where it makes the best decisions given the information/rules of the game it trained on. It doesn't think like a human or have some modular approach to being great at the game. You could pick earlier/weaker agents, but you'd be looking through millions of copies of mostly-similar things just to probably wind up with a still mostly-unbeatable AI with some obscure flaw, or going earlier something even worse than what we have now.

All you do is introduce an "error quotient" into each move it makes. Or every other move, or whatever. The lower the difficulty, the more mistakes it makes. The higher the difficulty, the less perfect moves it makes. It doesn't have to behave like an earlier version of the AI, all it has to do is make mistakes.

However, I really doubt it'd be possible to create an AI that can play Civ and beat humans each and every time unless you dedicated a lot of resources to this venture.

Either way, it's just been so depressing that each time a new Civ comes out.. The computer opponent is something they never spend any time on, in terms of improvements. At this point my post might as well go in the rants thread though

 

[H4run]

I'm not sure "put the same combination of units in literally every tile you use for combat" would make the game's decisions more diverse, on average.

Not in the sense that it's compulsory, but it's great way to:

1. Limit the stack without eliminating the stack.
2. Avoid (unit name) rush, that can dumb down and simplified or exploit the game.
3. Force player to build variety of units without increasing its price per production.

It's a great idea from Aelf. But I don't know how good the AI can follow it up. But perhaps this system can work good for AI, I imagine there would be no unprotected siege engine if the unit move in stack consist of variety of units. Promotion, tech and general perhaps can be use to fatten the stack, makes it bigger, so there will be more tactic and strategy variation.

 

[TheMeInTeam]

However, I really doubt it'd be possible to create an AI that can play Civ and beat humans each and every time unless you dedicated a lot of resources to this venture.

Machine learning AI have surpassed top professionals repeatedly, not just in chess but also in games like StarCraft 2 and Go. I don't know what you see in Civ that makes it special, because I sure don't see anything that a little extra training time wouldn't put to rest and allow it to do the same thing. Hardest part would probably be scaling the game into something that lets it quickly run the iterations.

Not in the sense that it's compulsory

When you have an obviously best choice, it's hard to avoid the conclusion that not doing it is a false choice.

but it's great way to:

1. Limit the stack without eliminating the stack.

I don't see an obvious benefit to doing this for its own sake. And games like Warlords 2/3 did this better with fewer constraints on what went in the stack, more than 20 years ago.

2. Avoid (unit name) rush, that can dumb down and simplified or exploit the game.

Actually it's not clear you'd avoid this, especially if we take for granted that the "put one of each unit type on each relevant tile" isn't true. This also isn't true (on average) in previous civ games, unless target didn't invest properly to stopping the attack. Besides, we already have a Civ game model that forces some combined arms (Civ 4).

3. Force player to build variety of units without increasing its price per production.

Forcing the same pattern w/o exception does not add variety, it reduces it. If you create a scenario where each tile only has one slot for each unit type, then unless production is a serious bottleneck it is strictly optimal to fill each slot. I hold this would necessarily fail to create a game state requiring more thought on which military units to produce, and would likely require less thought in practice.

But perhaps this system can work good for AI, I imagine there would be no unprotected siege engine if the unit move in stack consist of variety of units.

If you force the same stack combination to be optimal because it is the only choice, then yes it would be relatively simple to make the AI also do the only (non-false) choice consistently w/o hurting its performance. It would attain that goal. But let's not pretend this would somehow "diversify builds" or "add tactic and strategy variation". It would almost certainly reduce them, unless the current meta is to spam one unit type.

 

[H4run]

When you have an obviously best choice, it's hard to avoid the conclusion that not doing it is a false choice.

Generally might be right, but there can be a condition where moving the unit in smaller or incomplete stack can be beneficial, like to take the full benefit of cavalry movement, or to minimize the damage from collateral damage.

I don't see an obvious benefit to doing this for its own sake. And games like Warlords 2/3 did this better with fewer constraints on what went in the stack, more than 20 years ago.

I don't follow your argument here, never play the game, if the game able to make stack work better, why not demonstrate how civ's stack can benefit with borrowing some of the Warlords 2/3 stacking system/rule.

Actually it's not clear you'd avoid this, especially if we take for granted that the "put one of each unit type on each relevant tile" isn't true. This also isn't true (on average) in previous civ games, unless target didn't invest properly to stopping the attack. Besides, we already have a Civ game model that forces some combined arms (Civ 4).

Eh, axe rush? I don't know why you dismiss this clear problem.

Forcing the same pattern w/o exception does not add variety, it reduces it. If you create a scenario where each tile only has one slot for each unit type, then unless production is a serious bottleneck it is strictly optimal to fill each slot. I hold this would necessarily fail to create a game state requiring more thought on which military units to produce, and would likely require less thought in practice.

Good argument. We can make a variety perhaps by putting more variety of what combination of stack you may have. Lets say you can stack either 2 cav, 1 infantry, 1 pikeman and 1 archer. Or 1 siege unit, 1 pike man, 3 infantry. Or with certain tech, building, civ or policies you can also have different stack, like Mongolian for instance can stack 3 heavy cav and 2 light cav. So on so forth.

Not going into very much detail here, because I can't really argue about its technicality, because I never give it a well thought. I just give the general ideas.

 

[warpus]

Machine learning AI have surpassed top professionals repeatedly, not just in chess but also in games like StarCraft 2 and Go. I don't know what you see in Civ that makes it special, because I sure don't see anything that a little extra training time wouldn't put to rest and allow it to do the same thing. Hardest part would probably be scaling the game into something that lets it quickly run the iterations.

That's interesting, actually. I looked it up and you're right, they were finally able to write an AI that is able to consistently beat humans in GO.

This effort has taken years of effort though. It has been an unsolved problem for quite a while. I also assume that whatever methods they are using to "teach" the AI take a while and would not be easily transferrable to something like a board game being simulated on the PC (like Civ). It also seems that Civ has so many different ways to play, whereas GO has one set of rules and that's it. Which makes me wonder how the StarCraft 2 bot that can beat humans works? Do you have a link to this thing in action?

And I mean, that's a part of the reason why Civ doesn't use any AI principles when presenting us with computer opponents, right? Actual AI is hard and takes a while, and different approaches are needed for different situations. Most games don't try to figure out this problem by introducing machine learning, there's simpler approaches possible. Imagine having to re-teach the AI each time you add a new unit, a new map style, or a new game dynamic. Releasing DLCs would be a lot more work and mods would probably confuse computer opponents, depending on how they modify the game.

I was not asking for a pure AI opponents in Civ. I just don't want them to be so damn stupid and predictable. I want difficulty levels to be actual difficulty levels and not just handicaps. I want computer opponents having different strategies that go beyond "This guy likes to build wonders"

 

[uppi]

1UPT makes for a better war game. The tactics of a doomstack are much simpler than the tactics required for 1UPT. Whether or not Civ is supposed to be a war game is a different discussion though.

In my opinion, a lot of the problems with 1UPT in Civ 6 could be alleviated by giving every unit an extra movement point. Why they went with such a low movement speed is the more baffling design decision to me.

 

[TheMeInTeam]

I don't follow your argument here, never play the game, if the game able to make stack work better, why not demonstrate how civ's stack can benefit with borrowing some of the Warlords 2/3 stacking system/rule.

Because a lot of what made that system good was fantasy elements and a much wider array of units with +stack bonuses (many of which were situational outright or more impactful against certain things). No era in Civ has even close to that level of mixing/matching available to stacks, and if the game wants to keep any semblance of historical inspiration it probably can't. Warlords had an economy side, but it was more combat/war focused. Civ's degree of abstraction is a bit higher-order, and thus we wouldn't expect quite that degree of complexity in one element of its gameplay.

Eh, axe rush? I don't know why you dismiss this clear problem.

Assuming axe vs axe with anywhere near similar investment, defender wins the war and both players probably lose the game. Axes w/o spears will get absolutely rinsed from the map by chariots, but if they invest in spears they lose to pure axe at less hammer investment.

All of this gets more important as you add collateral damage, crossbows/gunpowder, and counter promotions.

Good argument. We can make a variety perhaps by putting more variety of what combination of stack you may have. Lets say you can stack either 2 cav, 1 infantry, 1 pikeman and 1 archer. Or 1 siege unit, 1 pike man, 3 infantry. Or with certain tech, building, civ or policies you can also have different stack, like Mongolian for instance can stack 3 heavy cav and 2 light cav. So on so forth.

Why constrain what can go in the stack, at all? Could offer different abilities/limitations depending on what's in it.

This effort has taken years of effort though. It has been an unsolved problem for quite a while. I also assume that whatever methods they are using to "teach" the AI take a while and would not be easily transferrable to something like a board game being simulated on the PC (like Civ).

It would be easier to look up machine learning. The gist is that training the AI on Civ doesn't seem to have any barriers that weren't already solved for the other games.

It also seems that Civ has so many different ways to play, whereas GO has one set of rules and that's it.

Civ has one set of rules and that's it, too. You have constraints, and victory conditions. But those victory conditions are all part of one rule set, and which victory condition doesn't matter, the AI just has to attain one of them. What I suspect you'd find is that Civ does not, in fact, actually have "many ways to play". And that instead some strategies/approaches strictly dominate others. Whether or not that's true we'd get to see play out if an AI were trained on it. But I don't see many "alternative strategies" that do anything meaningful against "technological advantage + military units". Maybe if the defender's advantage is too large for optimized AI that don't make mistakes and tech similarly, space would be the default win? Give it enough imbalanced game scenarios and I bet we'd still see it conquer weaker starts though. Would have to see it play out to know.

Which makes me wonder how the StarCraft 2 bot that can beat humans works? Do you have a link to this thing in action?

When saddled with extra restrictions, it "only" beats top competition most of the time. When it still has some restrictions, but much fewer, it went 4-1 against Serral (reigning world champion at the time).

https://deepmind.com/blog/article/A...t-II-using-multi-agent-reinforcement-learning

Whether the literal best player in the world could take a game or two off it in Civ wouldn't matter, much. Nearly every person in Firaxis' target market would have an unwinnable opponent in SP w/o massive bonuses.

And I mean, that's a part of the reason why Civ doesn't use any AI principles when presenting us with computer opponents, right?

Machine learning is pretty new (these AlphaStar developments are mostly post-Civ 6 release IIRC). I'm not familiar with it being used by game developers, and the people who handle it the best are usually working outside the world of gaming.

I think that's part of the reason. The other part of the reason is that the AI in Civ 6 (and earlier Civ games), in addition to being bad in general, is also intentionally bad to mask design flaws with the game. As in, I strongly suspect that if you are really trying to win against opponents who are also really trying to win, and everyone is playing close to optimally, that a significant chunk of the game's options and victory conditions might as well not exist, and Civs that emphasize those options would be strictly dominated by those that emphasize viable strategies.

Even going back to Civ 4 times, the devs went on record saying they designed their AI to intentionally not play the same game humans who are trying to win play, because they believed that approach to be more "fun". That practice/choice clearly remained in play for Civ 5 and 6.

I still hold that this is an admission of a design limitation, a flaw with the core rules of the game. But maybe they couldn't do anything about changing the rules/engine to work around that and instead compromised the AI to get something that could ship and play passably enough to have a good game otherwise. I can't even say they were wrong for making that choice, for accepting the design flaws in the name of producing a game at all (and one that was nevertheless quite successful).

I was not asking for a pure AI opponents in Civ. I just don't want them to be so damn stupid and predictable. I want difficulty levels to be actual difficulty levels and not just handicaps.

This generally involves writing multiple different AI, however. And Civ is in a state where it's not clear they can write one AI .

1UPT makes for a better war game. The tactics of a doomstack are much simpler than the tactics required for 1UPT. Whether or not Civ is supposed to be a war game is a different discussion though.

People like to claim this, but I haven't seen it demonstrated as true even once. You move a simple doomstack into the territory of a human in PvP at equal tech, and you better be damned sure his army is somewhere else. Otherwise, you're the one who will be out of the game soon.

In my opinion, a lot of the problems with 1UPT in Civ 6 could be alleviated by giving every unit an extra movement point. Why they went with such a low movement speed is the more baffling design decision to me.

You need more tiles/space between cities relative to production. Wouldn't hurt to make cities less ridiculous rather than more so too, since that would mean survival actually takes meaningful investment.

 

[warpus]

The goal with an AI-based opponent in video games is not to create an opponent that can always beat you. That part is a great challenge to be sure, but for video gaming what you want is opponents that seem human, both in their unpredictability and adaptability, but also ability to mess up and make mistakes and not always win.

In terms of deathstacks, to get back to the purpose of the thread.. I do like how you are presently able to stack multiple units (of different types) as well as merging them to form corps and so on.. I really think Civ should open up to the idea of expanding this functionality a bit more. So far we've experienced 2 extremes: as many units as you want in one spot.. and only one unit, with certain exceptions. Why not something a bit more balanced? Stacks of 5 max? 7? 10? Would that throw the balance of the game off?

Maybe my cynicism has lead me to the reason they have not done this yet - the computer opponents would have to be hardcoded to deal with all the new situations that pop up when you allow stacks. Or maybe we will see such modifications in the next iteration of Civ after all

 

[MrCynical]

Sid Meier wasn't the designer by the time Civ 5 came along. You can blame Jon Shafer for the whole 1 unit per tile thing.

The main problem with 1UPT in my opinion is that the maps in Civ simply don't have the room for the kind of tactical maneuvering that they were trying to base it on. They cited games like Panzer General as the inspiration for this, but those have far more empty tiles than units. In Civ 5 and 6 combat always turns into a horrible traffic jam because after you allow for mountains, water etc, there's often only a handful of tiles units can actually move to, and they'll likely all have a unit on them except in the very early stages of the game. Naval combat is noticeably better, as at least there tends to be enough open water tiles to do some actual maneuvering.

And the AI didn't have the faintest idea how to deal with it in 5 and is only slightly better in 6. It's not that this kind of tactical approach is inherently bad, but it doesn't fit on the scale of maps that any of the Civ games use.

I suspect a compromise of a limited number of units per tile (but a fair bit greater than 1) would work a lot better than 1upt, as well as opening up options for some combined arms strategy. At the very least it would effectively increase the amount of "space" for units and reduce the congestion issue.

 

[Snerk]

Stacks were lame and I'm glad they axed them. 1upt was an improvement but Civ5 favoured way too many units in play at the same time which made manoeuvring super tedious at times. Civ6 improved this.