Is AI really brain dead?

[Deggit]

But in the end, as your analysis implies, opinions on the subject are also coloured by how much you desire Civ to be a production and numbers game. 1UPT heightens the wargame side, which is to some extent at odds with the production side. But if the 1UPT road had not been taken, I'm not sure how Civ's wargame side could've evolved beyond smashing stacks of doom against each other: if you keep warfare strictly strategic, its complexity has a (pretty low) ceiling, one Civ4 had already reached, from my point of view.

I agree completely, Lord Shadow...except that I think a complexity ceiling on the wargame subsystem is a feature, not a problem.

It means there is more room for complexity in other parts of the game, for the AI to calculate and for the player to keep in mind.

Sid Meier's Civ started the 4X genre but it always had an uneasy relationship with that concept because a pure production-based wargame is not an accurate or complete depiction of real history. Civ always wanted to be a little more multidimensional than that, which is why it started out of the gate in Civ1 with an alternate resource sink (spend on technology instead of just spamming more units) and an alternate victory condition to conquest (space race).

In Civ2-Civ6 they gradually added culture, religion, leaders, great people, civ traits, governments, civic policies and more. That evolution has been good to see, even if most of these added features just looped back into the 4X concept by buffing or upgrading your production capacity.

The great leap forward for Civ was when people right here on civfanatics started doing OCC games ("one city challenge") in Civ3. It was possible to deliberately lose the 4X portion of the game yet still win by culture. I think that woke up the developers' minds to the idea that you could have entire subsystems of the game that did not necessarily loop back into "make your cities better at churning out settlers and warriors."

 

[Heikki Mustonen]

Hah. Finally lost my first game after those four wins I told earlier. Deity, random leader, random map. Got France. Starting location 2 hexes deep in the tundra. I spend first two turn just to run out of tundra and then settled on turn 3. Normal 2-1 capital and had 3-1 horses and 2-1 stones next to it. Went scout, slinger, builder. First barbarian scouted the city before slinger was ready. The turn I got the builder out my capital was surrounded with three barbarian warriors and two slingers. Two turns later they captured my capital and I lost . I'm not sure if even scout, slinger, slinger would have helped.

So there is an AI that knows how to attack. Sadly it's just barbarians.

 

[Horizons]

The thing is that they *can* be pretty close in tech, so the AI clearly knows how to generate science/culture. In my Japan-game, Victoria had ironclads around 1000 AD (though she obviously beelined to steam power... perhaps she got a good combo of random boosts via a GS?).

However: They don't seem to "understand" the importance of having a modern army. In my conquest game as Trajan, England was still on a carpet of doom of archers when I invaded them with battleships around turn 280 on epic - and she did have much better tech than archery at that point.

Another critical error they commit is to completely ignore production-buildings. I've seen this in numerous games now where the AI simply will not build industrial zones. It's actually easy to check, since usually you'll be the only one competing for Great Engineers and there won't be many/any foreign trade-routes that give you production. In my (pretty ineptly played) science-victory game as Japan, only one AI civ (France) eventually started building these districts when I was already constructing my spaceports. I mean.. I know they get substantial buffs to their production-rate, but even those can't be enough to outweigh the disadvantage of not having workshops, factories or power-plants.

S.

I agree. If the AI would upgrade its units, it would be quite a formidable foe. That, and settling more cities and it would be quite decent. Not sure how these simple things got missed in all the pre-release testing?

 

[LDiCesare]

The great leap forward for Civ was when people right here on civfanatics started doing OCC games ("one city challenge") in Civ3. It was possible to deliberately lose the 4X portion of the game yet still win by culture. I think that woke up the developers' minds to the idea that you could have entire subsystems of the game that did not necessarily loop back into "make your cities better at churning out settlers and warriors."

I'm sorry but you are wrong.It'snot Civ 3 and likely not on cfc that OCC originated.
People did OCC on apolyton since Civ 2. See this for instance : http://forums.civfanatics.com/threads/one-city-challenge-how.42403/ or even older http://apolyton.net/showthread.php/14945-First-OCC-with-ToT which
Unfortunately, the link to the Paulicy is down, but there was a very very nice guide on how to play OCC in civ 2.
There's also this linke I found : http://apolyton.net/showthread.php/14945-First-OCC-with-ToT which is civ 2 pre-dating ToT. I didn't look for too long either, but OCC predates civ 3 by a huge lot, and was imo infinitely more fun in civ 2 than in any subsequent iteration. As in, really, infinitely. If you want to play OCC, play Civ 2, it's way better than the 3-6 versions.

 

[Sascha77]

*sigh*

New game.. Pericles on Emperor, Marathon and large fractal (hey, a man's gotta earn his achievements ).

First DOW came on turn 17 from Trajan, who is my nearest neighbor but still a pretty good distance from me. I had a warrior and had nearly finished a slinger at this point, remember: this is Marathon. Trajan brought two warriors and shuffled them around while they were adjacent to my capital. Only after two or three turns of this did he actually attack the city, and by then my slinger was ready. I killed his "invasion force" and he peaced out, granting me gold and GPT. After that, I had DOWs nearly every 15 turns or so. Some came from civs I haven't even met yet (and not as part of joint wars, unmet AI-players seem to be able to declare regardless).

All of these wars went like this:
AI declares war
AI sends nothing towards me
After the minimum war-length is reached, they offer hugely favorable peace-deals.

Plus some were constantly denouncing me for no reason, while others denounced me based on their agendas but at ridiculous points in time. Like Trajan who would denounce me because I "controlled not enough land" for his taste at a point when I couldn't possibly have built my first settler.

The funniest bit is this however: After having been the target of DOWs from all but one of the Civs I've met (plus two or three from ones I hadn't met), Trajan plops down a city right next to one of mine, gobbling up resources I wanted for my city. By this time I had a sizeable army (three hoplites, three warriors and three archers). The terrain around his new city looked favorable, so I decided to take it from him. I had him denounced, but that hadn't worked correctly for some reason... I couldn't declare formally (something the AI seems to be able to do without even denouncing me). So I declared a surprise war at moderate warmonger-penalty, took his city, killed a few of his units, stole a settler, then peaced out. And now *everyone* hates my guts and denounces me for being a warmonger - especially those guys who had constantly declared on me themselves for the entire duration of the game ..

S.

 

[Heikki Mustonen]

And now *everyone* hates my guts and denounces me for being a warmonger - especially those guys who had constantly declared on me themselves for the entire duration of the game ..

What I have seen they also hate each others for warmongering so it's not just you they denounce it is everyone looks like. But yeah, doing any sort of attack after classical era means you are hated rest of the game by everyone unless you get lucky and manage to meat both of their agendas.

 

[bladex]

you'll want this than

 

[Ammar]

One UPT is one main reason why the Civ 6 (and to a slightly lesser degree, the Civ 5) AI is bad. People point out to tactics simulations like Panzer General or Fantasy General working, but if you look at the main difference between them and Civ 6 one UPT you also can understand why one works and the other one doesn't: movement points per turn. In those tactical games units usually move a lot farther than the 1-2 tiles in Civ 6.

That is why the units so quickly start up crowding each other and making optimal army movement a pain. It's just not the AI, I also find it painful trying to get my units to move in the right order along a one tile wide road. If you could move theoretically 10 tiles using an optimal order UPT might mean you can move one unit 10 tiles & the other one nine. A minor loss. With two tiles moves you quickly start losing 50% of your max movement, which is absolutely insane in terms of efficiency. It is a genuinely hard (but IMO non-interesting) problem to solve these kinds of logistics optimally in Civ 6.

There is also one more problem with both Civ 5 and Civ 6 combat. Combats not being to the death, leads to a winner takes all system. In Civ 4 if 10 axemen fight 6 axemen the most likely price for the larger stack to destroy the smaller one is 3 dead axeman & 3 wounded. In Civ 5 & 6 the 10 axemen will probably be able two win without any losses, because wounded units can always withdraw. As long as you are able to achieve local numerical superiority you will likely crush the AI without any permanent losses. And given how bad the AI is with moving units you are always able to achieve this local numerical superiority.

 

[Deggit]

given how bad the AI is with moving units you are always able to achieve this local numerical superiority.

Yep, that's another good way of putting it.

People are saying "The AI was always bad" - this is true but it misses the point. In every game from up to V (and I assume in VI as well?) the AI is the exact same script at every difficulty level, it's just given lots of freebies and cheats at high levels. The AI is pretty terrible at the "equal to human player" level in every single Civ game.

The only way Civ designers, across 25 years of games, have found to make an AI that actually THREATENS the player, is

1) give the AI tons of production bonuses and then
2) make combat simple enough that the AI can achieve local numerical superiority.

Both halves of that equation are crucial. If the AI is not given cheats, it will lose wars because the human player will ALWAYS be better at achieving efficient production. If the AI can't achieve local numerical superiority, then giving it tons of production bonuses is pointless because the player can ALWAYS figure out a way to string out the AI and defeat it piecemeal.


Even if the AI played 1UPT like a champ (which it doesn't and never will) the essence of 1UPT is to "smear out" battles. Instead of a one turn decisive engagement between two massive stacks, it becomes a multi-turn slog of attrition, rotating units etc. This inherently favors the numerically smaller force because it gives them opportunities to win battles with local superiority, exploit tactical mistakes, rotate units, etc.

The stronger side always wants to force a decisive battle as soon as possible. 1UPT is actually punishing the stronger army. The war subsystem of Civilization is fundamentally about who brought more dudes to the battlefield, not how you manage them. The whole reason one side has 1000 dudes and the other has 300 is because the 1000 side settled better cities, developed them more efficiently, and took a risk by investing in military tech instead of production tech, and by building units instead of infrastructure or wonders. The 300 side deserves to lose if they can't figure out a way to avoid the battle altogether (hit and run tactics, defensive diplomacy, appeasement & tribute, etc).

 

[CaiusDrewart]

People are saying "The AI was always bad" - this is true but it mises the point. In every game from up to V (and I assume in VI as well?) the AI is the exact same script at every difficulty level, it's just given lots of freebies and cheats at high levels. The AI is pretty terrible at the "equal to human player" level in every single Civ game.

The only way Civ designers, across 25 years of games, have found to make an AI that actually THREATENS the player, is

1) give the AI tons of production bonuses and then
2) make combat simple enough that the AI can achieve local numerical superiority.

Both halves of that equation are crucial. If the AI is not given cheats, it will lose wars because the human player will ALWAYS be better at achieving efficient production. If the AI can't achieve local numerical superiority, then giving it tons of production bonuses is pointless because the player can ALWAYS figure out a way to string out the AI and defeat it piecemeal.


Even if the AI played 1UPT like a champ (which it doesn't and never will) the essence of 1UPT is to "smear out" battles. Instead of a one turn decisive engagement between two massive stacks, it becomes a multi-turn slog of attrition, rotating units etc. This inherently favors the numerically smaller force because it gives them opportunities to win battles with local superiority, exploit tactical mistakes, rotate units, etc.

The stronger side always wants to force a decisive battle as soon as possible. 1UPT is actually punishing the stronger army. The war subsystem of Civilization is fundamentally about who brought more dudes to the battlefield, not how you manage them. The whole reason one side has 1000 dudes and the other has 300 is because the 1000 side settled better cities, developed them more efficiently, and took a risk by investing in military tech instead of production tech, and by building units instead of infrastructure or wonders. The 300 side deserves to lose if they can't figure out a way to avoid the battle altogether (hit and run tactics, defensive diplomacy, appeasement & tribute, etc).

This is really well put, and gets to the core of why 1UPT is a poor fit for the Civilization series. Even if the AI could handle it well, 1UPT would not necessarily be a good idea. If tactical combat is important, than building a good empire automatically becomes less important. That's bad. And of course, it is universally admitted that the AI cannot handle it well.

Anyway, the problem in Civ VI is not merely 1UPT, since the Civ VI AI is clearly an easier opponent even than Civ V's was. Civ VI is a little different in that the freebies and cheats just do not seem to be high enough. The AI is really struggling to keep up in Science and Culture on high levels. Not enough expansion and struggles with the Housing/Amenities system might be part of the problem. This is easily fixable though--crank up the percentage bonuses a bit, and have the AI basically ignore those city growth limits. Those mechanisms should be there to challenge the human player, not to stunt the AI. Give the AI some help. I'm seeing AIs with 3 cities and a size 10 capital in the late game on Deity, and that should never happen.

 

[Lord Shadow]

I agree completely, Lord Shadow...except that I think a complexity ceiling on the wargame subsystem is a feature, not a problem.

It means there is more room for complexity in other parts of the game, for the AI to calculate and for the player to keep in mind.

Sid Meier's Civ started the 4X genre but it always had an uneasy relationship with that concept because a pure production-based wargame is not an accurate or complete depiction of real history. Civ always wanted to be a little more multidimensional than that, which is why it started out of the gate in Civ1 with an alternate resource sink (spend on technology instead of just spamming more units) and an alternate victory condition to conquest (space race).

In Civ2-Civ6 they gradually added culture, religion, leaders, great people, civ traits, governments, civic policies and more. That evolution has been good to see, even if most of these added features just looped back into the 4X concept by buffing or upgrading your production capacity.

The great leap forward for Civ was when people right here on civfanatics started doing OCC games ("one city challenge") in Civ3. It was possible to deliberately lose the 4X portion of the game yet still win by culture. I think that woke up the developers' minds to the idea that you could have entire subsystems of the game that did not necessarily loop back into "make your cities better at churning out settlers and warriors."

The warfare side's ceiling never was a feature. Civ iterated on the stack system until it reached its peak in Civ4, and then discarded said method. If 1UPT's tactical shift had been such a blasphemy to the nature of the series, I doubt the Civ5 team would've been allowed to go through with it. Sid made wargames once upon a time, and I don't see him being opposed to the idea, and as it was quoted earlier, even Civ4's Soren Johnson had been intrigued by the notion.

And as you said, the other subsystems kept evolving, so why couldn't the warfare one? As others have mentioned, it ultimately comes down to personal expectations about the franchise. Some would like to keep warfare simplified and focus on the production and numbers pseudo-puzzle game, like SimCity or other city builders such a Caesar, Zeus and The Settlers. Personally, I include myself in another group: while I'm not a warmonger, I consider warfare amongst the most fun parts of Civ even if I appreciate the importance of peaceful empire-building. I cannot condone it taking the backseat in favour of figuring out just how to get bigger numbers before an increasingly inflated AI beats me to whatever victory.

 

[LDiCesare]

The AI has trouble handling 1UPT but right now it can achieve some superiority. It simply has no idea how to exploit it (for instance it would be infinitely better if it attacked cities).
1UPT can be fun, but it requires lots of room so you can actually perform some tactical moves. Civ maps are too crowded for that. Except civ 1/2 actually, where the system was pretty close to 1upt since losing 1 unit ment losing the whole stack. Maps got smaller, particularly in 5.
One way to prevent the carpets of doom that prevent tactical moves would be to do like in civ 1/2, to limit the number of military units you could have. Right now, the gold cost is not a limit in any way. Civ 1's production limit was much more radical. Each city shouldn't be able to support more than 2 or 3 units. This means a vaster empire could attack in more places and win on one front even if locally outnumbered on another, ot it could just replace losses more quickly than the defender could build up?

 

[Deggit]

Civ maps have been shrinking for 10 years. The largest, in terms of # of tiles, maps were the Huge world maps from the Civ III era.

I still laugh when I see Civ 5 players debating "Tall vs Wide" and they mean a 4 city empire vs an 8 city empire. In Civ III you usually had 10-12 cities in the midgame on large/huge maps and then up to 24 or 30 after you conquered one or two nearby civs.

 

[CaiusDrewart]

Civ maps have been shrinking for 10 years. The largest, in terms of # of tiles, maps were the Huge world maps from the Civ III era.

I still laugh when I see Civ 5 players debating "Tall vs Wide" and they mean a 4 city empire vs an 8 city empire. In Civ III you usually had 10-12 cities in the midgame on large/huge maps and then up to 24 or 30 after you conquered one or two nearby civs.

I'm so glad to see Civ VI break from Civ V in this way. I mean, Civ VI has its quirks, but at least the game rewards making a real empire again. That makes a HUGE difference.

 

[genyl]

I was loving this game when I got into it because there are many good things in it. I like the general game design and mechanics. The music is fantastic and it looks generally polished .

However once I scratched the surface I don't want to really play it more at this point. The main reason is the AI is as bad as everybody is saying. I haven't seen something so bad and in a very expensive video game is outrageous.

I have checked the steam achievement and I was surprised to see only 4% of players have the deity victory achievement. I am not the greatest player and I still don't know the detail of the mechanics, however my first game in deity was disappointingly easy. Admittedly I played with germany and I had to restart because of the barbarian onslaught but once you are set up is a cruise. It doesn't matter if the IA has these massive armies because they simply don't attack you! they denounce you and parade up and down, I mean it is simply pathetic. I would be happy if the game was challenging even if the AI is bad but it is not. Even worse it is irregular and illogical. Diplomacy is like a lottery.. I suspect my next deity game I may be sweep off, this just seems like poor programming

This is exactly what happened with civ5 though so I won't lose hope yet

 

[Ferocitus]

Plus some were constantly denouncing me for no reason

It's more accurate to say that you did't know the reason why they denounced you.

And now *everyone* hates my guts and denounces me for being a warmonger - especially those guys who had constantly declared on me themselves for the entire duration of the game ..

So working like it does IRL then, where, for example:

"During its history, British forces or forces with a British mandate have invaded, had some control over or fought conflicts in 171 of the world's 193 countries..."
https://en.wikipedia.org/wiki/List_of_conflicts_involving_the_United_Kingdom

That hasn't stopped modern British governments squealing and denouncing the bellicosity of others.
Read up on Commodore Perry's exploits in Japan, or the Opium Wars, to see more full-blown hypocrisy.

If anything, Civs should have many more hidden agendas and always act according to de Gaulle's dictum:
"Nations do not have frends, they have interests."

 

[Jabulani]

A guy here in this forum perfectly summarized :

To devs : " you are seriously wrong in giving the AI 3 queens, you must teach it to play chess instead "

Such statement deserves a Sean Bean civ quote in the game

 

[ThERat]

Civ maps have been shrinking for 10 years. The largest, in terms of # of tiles, maps were the Huge world maps from the Civ III era.

Hell, yes. Have you seen those games we played on giant maps where we would reach 512 city limits. Those games were meant to played over weeks and huge fun.
We already debated whether Civ4 maps are too small. Well, there are some MOD like Realism Invictus, where the world map is pretty big on Civ 4 too.

1upt needs much larger maps. I have now played 3 Always War games and it's pathetic what the AI achieves. They have no chance in hell to win even on much higher difficulties. 1upt is simply not meant for a game such as civilization. But what does Civ6 do. It makes moving even more complicated thanks to 'must have full movement left to move into hills' and zone control. Are you kidding me? Have they actually played that mess themselves. It's dire and really painful.

 

[Deggit]

A guy here in this forum perfectly summarized :

To devs : " you are seriously wrong in giving the AI 3 queens, you must teach it to play chess instead "

Such statement deserves a Sean Bean civ quote in the game

The AI will never "play chess"! It's impossible!

The AI could hardly even manage Civ III's economy, and that one had things like ICS and lumberjacking where players exploited the economy in ways the developers never even anticipated. Then in Civ IV there was already things like great people points and cottage economy that players could scarcely understand, the forums were full of threads explaining & analyzing the system. Civ V and VI have only made the economy even MORE complicated.

The main reason we care about the AI anyway is how it interacts with the player. I have no problem with the AI's internal economy being fake (i.e. the game just gives the AI free stuff commensurate with what a player would have at that point in the game). It just has to look and feel fairly realistic.

The real problem is when the AI cannot apply the resources it is given to compete with the player i.e. beat the player to settlement spots, discover techs before the player, found religions before the player, build a strong culture engine before the player, beat the player to Alpha Centauri, and threaten to steal the player's cities with military. The AI must be able to do these tasks.

So let me turn your saying around -

"There is no problem with giving the AI 3 queens, the problem is the AI cannot even put the player in check."

 

[ProMeTheus112]

Someone wrote before that he made an AI for a 4x game similar to alpha centauri and that AI was hard to beat even without bonuses. Like the game had an AI similar to a civ game, and he turned it into something much harder to beat and they reduced its bonuses for higher difficulties. Like the average player would need to clock few dozen hours into the game before he could beat the AI without bonus. I think he said it took him a bit over a year doing this just by himself. So it seems it's possible to make a strong AI for a 4x game, I would say that's just what civ needs. A few people on it for 6months should manage a strong AI?
It's critical to the depth of the game in the long run I think, so that you actually have to expertly respond strategicaly (and tacticaly) using the systems of the game against your opponents and events. If the AI is rly bad, the best way to play is to exploit the weakness, and the hardest difficulty becomes where the bonus are so high and you just need to extremely exploit the weakness or something.