Why Sid Meier gave up on stacking units?

[Angst]

It's honestly quite simple, doomstacks were never intended.

Here they note that they initially intended a wide, WW2-style line similar to old wargames, but Civ1 was balanced wrong. Over the years they've slowly made attempts at removing doomstacks, and they now think that they've met that goal (not directly saying it, but yea, not allowing stacks at all fulfils their vision I guess)

 

[Ajidica]

I think Civ abandoned stacks of doom because they wanted to make individual units more important, so the player has use to them intelligently through interesting decisions, rather than just flood the map. That's also fairly well seen in the reduction in number of cities a player can reliably get. Reduce the number of decisions the player has to make so the single desicsion is more interesting and more meaningful to better highlight trade-offs.
It isn't a bad idea, just one they are struggling to balance well; a situation I feel isn't helped by -in by very limited experieince - that Civ6s desire to replace 'paint the map then choose how to win' with 'choose how you are going to win on turn1 and then focus on that'.
Like, with cities, because of how important district bonuses and wonders are, it feels you have to plan out your city all the way into late game before you settle your second city. I've lost track the number of times I realize 50 turns in that I placed a district in the wrong tile so now I either miss out on major adjacency bonuses or can't build that wonder. It isn't a great feeling.

 

[EnglishEdward]

Last I heard, he was still alive.

https://en.wikipedia.org/wiki/Sid_Meier

Has anyone actually asked him?

 

[Angst]

Last I heard, he was still alive.

https://en.wikipedia.org/wiki/Sid_Meier

Has anyone actually asked him?

I'd refer to my above video. He states his position directly during it. I don't know the timestamp sadly, but he notes that doomstacks were never intended and was more a result of wrong balancing on the developers' part, compared to the design intent. Wide lines were always the thing they wanted to do. It's not a "Civ5 came in and changed Sid's vision" but rather that Sid didn't manage to achieve what he wanted to do in earlier versions.

Now, I must stress I prefer stacks myself. I'm not defending 1UPT. I'm just saying that it was always their vision. They always wanted to emulate wargames and make wide lines, and after some balancing and back and forth, Civ5 just went "screw it" and did 1UPT to mechanically enforce what they always wanted to do.

 

[EnglishEdward]

Thank you.

I'll have a look when I get a spare hour.

 

[aelf]

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.

Yes? Range and melee/cavalry dominate the early to mid game phase. In fact, sometimes people build almost exlusively ranged units. And that's because of 1 UPT.

There are reasons people don't build siege units much, but that's partly because they don't offer much for the space they potentially take up. Now, they do offer something - they let you maximise the use of a single tile in combat. At worst, they can move through other units that are jammed up because they can't be on the same tile. You can still don't build them. You can let your 1 UPT cavalry run deep into enemy territory. It could still work, but you could also run into an enemy combined-arms stack, which will be more effective than your unsupported cavalry in a straight-up fight.

That doesn't lead to more decisions in combat?

 

[Angst]

Thank you.

I'll have a look when I get a spare hour.

It's a good watch! Very fascinating.

 

[aelf]

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.

One value it might have for the AI is it prevents a traffic jam. No more melee and ranged getting in the way of each other, for example. That AI archer can suddenly pop up in wall of its warriors and hit your units from safety.

 

[Lohrenswald]

I even saw a talk he had about it where he briefly said he thought 1upt was bad design

It's honestly quite simple, doomstacks were never intended.

It was this one

Very briefly somewhere in here as an aside really, he says something like 1upt being bad design (I don't fully remember, but I think he said something like it ruined it)

 

[TheMeInTeam]

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.

Aside from childish intentional game-throwing, which we probably don't want the AI to model, most humans will at least broadly try to win, and be some combination of unhappy/complaining/adjusting when they lose.

The "fun" AI, as Soren called them, role played in a way that most humans would not be happy to do, because it reliably loses the game. Even players that allegedly don't care for conquest or optimizing tend to complain when they consistently losing while trying to do what they want.

Presumably, the game is good because its mechanics lead to interesting choices when trying to win. When this isn't happening, it's an indictment on the game's designed mechanics. You can only somewhat patch over that hole with a plaster of suboptimal AI. "Our alternative strategies are only viable when opponents aren't trying" isn't a convincing position for those alternative strategies being something other than a false choice/challenge run.

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.

Alternative would be to simply increase number of tiles/space between cities. We're far enough into 64 bit times now that this could probably be accomplished. Though Firaxis' performance optimizations might not be able to handle it, there's no reason in principle you couldn't have 1000's of tiles running fine, especially with fewer cities/tile per rules.

Ranged being very strong in 1UPT is an issue Firaxies has yet to fully solve, in general. Ranged was used throughout history for good reason, but its balance compared to melee is clearly too strong in Civ 5/6.

Movement in 1UPT is still awful though, would be nice to find a compromise there with UI so shifting fronts isn't a chore.

I think Civ abandoned stacks of doom because they wanted to make individual units more important, so the player has use to them intelligently through interesting decisions, rather than just flood the map.

Players got away with this while playing vs the AI, but every single Civ game has punished "flooding the map" to differing degrees. Investment into more units than you need + margin of safety means you are leaving other productive outputs (science, gold, etc) on the table and either slowing down your win or letting other opponents close the gap. If you really want to see this taken to its logical extreme, read up some of the team competition threads where each team plays the same map. Particularly the ahead-of-time micro planning for unit production, which units, where they will go, and when. Even in SoD era, if you have opponents doing that while you try to just flood with units, you will lose and lose badly. Both in time rush against AI, and in direct combat vs such a player/team.

And their micro conclusion isn't just "build the same unit a lot", there is at least some deliberation of what to produce and in what ratios in a lot of cases.

Yes? Range and melee/cavalry dominate the early to mid game phase. In fact, sometimes people build almost exlusively ranged units. And that's because of 1 UPT.

Range, melee, and cavalry are three different unit classes, and in the early game siege doesn't exist as a combat unit (just support) until catapults.

There are reasons people don't build siege units much, but that's partly because they don't offer much for the space they potentially take up. Now, they do offer something - they let you maximise the use of a single tile in combat.

The alternative would be to make siege do what they did historically: strike cities from ranges they could not trivially return fire. Though it's also worth noting that IRL siege units pre-cannon weren't nearly as common as we'd expect based on Civ, and a pretty common way to take a city was to starve it out (can't do that at all in Civ).

Anyway, going from "never build this" to "always build this because it's the only thing that can fill a slot" does not increase meaningful choices in the game.

You can let your 1 UPT cavalry run deep into enemy territory. It could still work, but you could also run into an enemy combined-arms stack, which will be more effective than your unsupported cavalry in a straight-up fight.

That doesn't lead to more decisions in combat?

In Civ 6, the cavalry will tend to just die to city wall shots, even with some pillaging.

And no, you do not get more decisions. Every tile would have the same combinations of units. Even in your quotes, you're saying "almost always" wrt unit construction, which implies that right now, there are exceptions. If each unit type has one slot per province, then lack of production is literally the only thing that would make you even consider any configuration aside from "everything in each tile". Civ 6 combat isn't exactly a deep tactical experience, but IMO it offers more than that right now, and so did Civ 4.

 

[warpus]

Aside from childish intentional game-throwing, which we probably don't want the AI to model, most humans will at least broadly try to win, and be some combination of unhappy/complaining/adjusting when they lose.

I mean sure, but who wants to play against an opponent who always wins? As an academic exercise in futility, sure, but as a form of entertainment?

Maybe I'm just not understanding what you are trying to say. My position is simply that the goal of any AI opponent designed to be used for a video game should be to make the game more fun. This means that the opponent should feel like a human, but not actually be a human. Aside from that the actual ability level of this opponent is up for debate and would likely depend on whatever difficulty setting you set in the game. Makes sense, no?

 

[Gorbles]

I mean sure, but who wants to play against an opponent who always wins? As an academic exercise in futility, sure, but as a form of entertainment?

Maybe I'm just not understanding what you are trying to say. My position is simply that the goal of any AI opponent designed to be used for a video game should be to make the game more fun. This means that the opponent should feel like a human, but not actually be a human. Aside from that the actual ability level of this opponent is up for debate and would likely depend on whatever difficulty setting you set in the game. Makes sense, no?

In essence, what you're asking for is "please build a functioning AI that can compete with a human, and then hobble it in a believable manner to aid the goal of human fun". I'm not trying to ridicule you here - you're asking for what a lot of people want out of game AI. But making that happen is very, very hard. It's why game AI is so often complemented by "cheating". All it has to do is fool the player enough. To that end, it doesn't actually have to be a good AI. Firaxis don't necessarily need to commit the amount of work you think they do, to make this happen.

This whole tangent kinda segues from / alongside what I was discussing with you. Certainly, at the moment, for various types of player, the Civilisation AI isn't capable of making a game enjoyable. But to improve or address that, "make a better AI" isn't the end goal either. That's technically the wrong way to go (as well as being very difficult).

 

[TheMeInTeam]

I mean sure, but who wants to play against an opponent who always wins? As an academic exercise in futility, sure, but as a form of entertainment?

Maybe I'm just not understanding what you are trying to say.

Most players would not play an unwinnable game, so we'd be stuck with bonuses to player.

Though it seems that there's a "diamond-level" agent version of AlphaStar, so maybe I was too hasty in concluding that it isn't viable to make flawed AI that still plays pretty well using machine learning (diamond league is a reasonably high percentile, but lots of players are better than that). I again emphasize that doing this is probably over most devs' heads, but it's interesting to see that it's possible in principle and wouldn't be surprised if it's more common in the future.

I don't think many games with complexity will have good AI written conventionally. Too many changes to the game itself, too much complexity for AI written that way to handle. There are probably people out there that could do it, but they're being paid more than they'd get in game development, and writing a good game AI requires at least understanding what makes for good gameplay in said game. Most devs are adequate at their own games, but it's rare that they're actually legitimately good at their own games. I think the Celeste dev is, but cases like that are exceptional.

 

[aelf]

Anyway, going from "never build this" to "always build this because it's the only thing that can fill a slot"

No, that's not correct.

 

[warpus]

Most players would not play an unwinnable game, so we'd be stuck with bonuses to player.

That's what I've been trying to say this whole time though - I want us to get away from bonuses and handicaps, and instead have a game where each difficult level plays differently. i.e. if you select a higher difficulty level, your opponent increases in skill and ability, strategic planning, etc. And not simply bonuses for the computer and handicaps for the human player.

Whether we use AI techniques to get there or not is another question. However, the computer players in civ and other games are called "AI" by default, so the conversations tends to head in that direction..

In essence, what you're asking for is "please build a functioning AI that can compete with a human, and then hobble it in a believable manner to aid the goal of human fun"

Not necessarily, that was simply one potential scenario. Give us an AI that feels like you are playing against a human, no matter how that is accomplished. Use AI techniques or not, that is beside the point. The only reason AI ever came up is because that is the default name for computer opponents in video games.

Whether you design the perfect computer opponent who will be able to beat virtually anyone in almost any situation, and then dumb this down for different difficulty levels.. or take another approach.. is of no consequence to me. My point is that the current way difficulty levels are built into the game is lazy and gets boring if you are an experienced enough player.

This whole tangent kinda segues from / alongside what I was discussing with you. Certainly, at the moment, for various types of player, the Civilisation AI isn't capable of making a game enjoyable. But to improve or address that, "make a better AI" isn't the end goal either. That's technically the wrong way to go (as well as being very difficult).

The problem I've pointed out initially were the difficulty levels and how they're implemented. One way of fixing this would be to improve the computer opponents and the way difficulty is handled. Surely this is a logical train of thought?

Technically the game makes no use of AI concepts whatsoever, so there is no AI there to improve in the first place... I threw out "AI" as an idea as to how better difficulty levels could be accomplished. There are likely other ways to get to a better game, not necessarily using any sort of AI concepts at all.

 

[Lexicus]

Did 1UPT make the game less military oriented or at least point it in that direction?

Yes and no. The AI can't threaten you in 1upt conditions, so it's less military-oriented in that sense, but the AI also can't stop you from simply rolling over it whenever you want, so it's more military-oriented in that sense.

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.

This. It also, thematically, makes no sense. Archers firing over the English Channel and all.

I think Civ abandoned stacks of doom because they wanted to make individual units more important, so the player has use to them intelligently through interesting decisions, rather than just flood the map. That's also fairly well seen in the reduction in number of cities a player can reliably get. Reduce the number of decisions the player has to make so the single desicsion is more interesting and more meaningful to better highlight trade-offs.

If they were trying this, they got the opposite of what they wanted, at least in Civ 5. Changes to the underlying terrain-improvement-yield system (to say nothing of the removal of the slavery whip) made city-management and worker-turn decision far less meaningful in Civ 5 than they were in Civ 4.

but rather that Sid didn't manage to achieve what he wanted to do in earlier versions.

Is there a kinship between Sid and George Lucas then? Both geniuses who made masterpieces and didn't even understand why those masterpieces were good?

 

[TheMeInTeam]

No, that's not correct.

Care to clarify how I'm mistaken? Are you also significantly messing with unit stats/costs, effectively reworking not only stacking but also how units fight, period? You haven't written anything that leads to a conclusion other than "stack as many things in each tile as production allows". I guess if a map is big enough, you could avoid that issue by having surrounding threats or something...but that's also true with unconstrained stacking or no stacking as well.

That's what I've been trying to say this whole time though - I want us to get away from bonuses and handicaps, and instead have a game where each difficult level plays differently. i.e. if you select a higher difficulty level, your opponent increases in skill and ability, strategic planning, etc. And not simply bonuses for the computer and handicaps for the human player.

At least historically, this has been too costly/impractical to be viable, and when it comes to the highest difficulties far beyond the developer abilities...both as programmer and player. To my knowledge, there's not a single person who has ever worked at Firaxis who could play Civ at a level even kind of close to someone like Rusten or Unconquered_Sun. Even if they had perfect programming knowledge and made no mistakes, they couldn't write an AI to outperform those players w/o bonuses. They're not good enough at Civ, either. Similarly, when the devs for FTL: Faster than Light first made the game, they envisioned a ~10% winrate for players, on normal. Now, with hard mode added (much less scrap income) I have a > 80% vanilla winrate and a few different double-digit streaks, and the best players are > 90% with streak records > 100.

With that kind of disparity between developer and player, I don't think they can possibly make an AI competitive with strong players w/o using bonuses, absent reliance on machine learning.

Whether you design the perfect computer opponent who will be able to beat virtually anyone in almost any situation, and then dumb this down for different difficulty levels.. or take another approach.. is of no consequence to me. My point is that the current way difficulty levels are built into the game is lazy and gets boring if you are an experienced enough player.

I agree, but it's non-trivial to solve this, per above. You're asking more than there is skill (in programming, play knowledge or both) to deliver, and more than is needed to produce successful games (so insufficient financial incentive to recruit stronger programmers alongside the designers).

 

[Fippy]

Civ is too complex, in IV there are lightyears between deity players and those who struggle with Emperor.
Nobody could expect Devs managing to create equal (or even close) AIs

So i think AIs that can play a game very good just means we talk about easy games anyways.
Unless it's something like chess, but the rules of chess are much simpler and super computers just need to be fed all possibilities, openings etc.

One nice example of why doomstacks worked: when i was still less experienced with deity, i felt strong and ran into enemy territory blind with my large stack of advanced units.
But what's that after ending my turn..attack after attack after attack, and when the dust settled i lost 75% of my units. AI threw all their siege & main stack on me.
They ofc couldn't do that if they were all spread out with 1upt.

So it's not that i defend doomstacks, i defend a sense of danger even when i feel strong that can catch me off-guard.
No shiny new other features can replace that feeling, if the AI struggles with defeating 4 units.

 

[aelf]

Care to clarify how I'm mistaken? Are you also significantly messing with unit stats/costs, effectively reworking not only stacking but also how units fight, period? You haven't written anything that leads to a conclusion other than "stack as many things in each tile as production allows". I guess if a map is big enough, you could avoid that issue by having surrounding threats or something...but that's also true with unconstrained stacking or no stacking as well.

I wouldn't make units in a 1:1:1 ratio all the time. I can't imagine that to be an effective strategy, given that production is limited. I would only build siege if I really need to maximise space usage due to unfortunate terrain or if I need to take down walled cities. Overall, I'd still favour building ranged early to mid game and build only as many melee and siege units as I need.

The difference between this and 1 UPT is I'd build more siege and now I would have no problems bringing up reinforcements from the rear as long as they're a different type of unit. That is way better than the 1 UPT situation.

 

[TheMeInTeam]

I wouldn't make units in a 1:1:1 ratio all the time. I can't imagine that to be an effective strategy, given that production is limited.

That isn't a given in Civ 5 or 6 though, because of crowding. Player and AI alike can trivially fill up way more than the available frontage. This is part of the reason ranged is so dominant in the first place right now, and a major factor in the problems with 1UPT in general.

Nobody could expect Devs managing to create equal (or even close) AIs

I do want to see a good machine learning AI trained on Civ 4. Those things are nightmares in the context of a single task/set of rules, once optimized for them. Would be fun to see an AI w/o bonuses outperform current deity AI.