The AI on Sovereign Difficulty

[ggalindo001]

Playing a VI game now, had a near-death Horseman next to a capital. Next turn, AI shoots at an Archer with an Archer in the city, and 2 of his horsies swap places, neither of them finishing off the easy picking Horseman, nor finishing the easy to reach Archer.

Sure, the AI may deliver troops to the front quicker and at one time with Commanders - but if the actual tactical battles keep using VI or VI-like code, it will be a total joke once again. I saw on the livestream that the foreign ships sailed away from danger. Definitely tells us nothing...

It is this that is the most frustrating aspect about Civ 6 and possibly Civ 7 if replicated.

Unless there is an overwhelming strategic objective that otherwise dictates, finishing off a unit has a massive tactical advantage in the Civ 5-6 series. The opportunity cost is tremendous.

I can understand if I have an equally fatigued unit that I might want to get my unit to safety -- so in that case, it is understandable not killing off an enemy unit.

But having the ability to do so, while fully protected in the city center -- unless the fire was aimed at a unit that needed to otherwise be damaged or would kill me outright -- not firing at this horseman while fully protected is just plain idiotic. Or not using the city strike blast to kill that unit as well.

I see this all the time with Civ 6 AI. You can get very lazy with fatigued units in Civ 6 -- in Old World, they would get annihilated.

I somehow think that the AI doesn't factor in opportunity costs appropriately into it's tactical decision making. The ability to kill a unit has such a massive advantage -- I (or the opponent if another AI) now have to invest time. production and potentially strategic resources to rebuild and replace the unit -- if heavily damaged, you only need to invest time (and have available the strategic resources)). The AI tho may be so worried about losing a unit that it just decides not to fight -- without realizing that it is safe from destruction. I don't know.

And what's worse -- the Barbarian AI usually DOES tactically fight more appropriately - maybe because there isn't a corresponding calculation of the cost of attacking, which seems to be in tremendous error here. Or maybe because Barbarian units don't heal, and as a result, they just go for broke.

The more I think about the responses and the messaging from the Firaxis team -- the more I think that they are focused on the new gameplay mechanics AI (so, the "points" system) and not as much on the tactical combat and/or military side. Which is a shame -- and it also makes me think that the AI around the commanders and deployment will also not be solid -- leading to easy pickings of vulnerable units because they are deployed incorrectly.

I still would want there to be a multiple unit per tile concept -- limited early to maybe 1 per category (so no stacks of doom) and later to 3 per category per tile. This could help the AI not be so fickle in open combat. There is a mod that I use for Civ 6 that allow 1 per category per tile, the AI is much better as a result IMO.

 

[Josephias]

They made a (poorly done) stab at this in 6 with a negative modifier if the player was ahead ("They hate us because we are winning.") but it didn't do much and was '4th wall' breaking.

I agree 100% on a toggle/slider - essentially "AI plays like a competitive player" on one end and "AI roleplays on the other". You think it could be done with a better implementation of that sort of modifier like in 6.

Since the essentially question for that is basically - If you and Gilgamesh have been allies since the ancient era, should he turn on you and start nuking you because you are getting close to victory? I suspect a lot of people do not want that.

Yes, that's a quite bland and unreasonable interpretation of the rule, but very used in many games (looking at you, Total War series...)

With "if they fall to much behind victory, they change their objective to disrupting as much as possible the players they have most grievances with" i proposed something more nuanced. First, it's rage does not need to be directed to the player if he had good relationship with them (nevertheless you could expaind quite a lot the causes of grievance to not abuse this - maybe you don't helped, maybe you don't wanted to trade, and all that created tension...). If the AI is attacking other player, at least it is generating chaos and movement in the game, and that is what we look for (and that other player may fall behind and it may indeed have grievances with you - or if the other player is you ally you may be forced into the war).

 

[King Jason]

Civilization hasn't had good A.I. in over a decade and that's probably not going to change.

Old World is on sale right now, just try it. It's genuinely better than any civ-style game I've ever played and it has amazing, interactive dev support and a strong discord community with great multiplayer.

It's challenging, nuanced, in depth, and doesn't feel like a mobile game.

Civ 5 and civ 6 had pretty terrible A.I. and I have very little reason to suspect civ7 will be any different.

 

[ggalindo001]

Civilization hasn't had good A.I. in over a decade and that's probably not going to change.

Old World is on sale right now, just try it. It's genuinely better than any civ-style game I've ever played and it has amazing, interactive dev support and a strong discord community with great multiplayer.

It's challenging, nuanced, in depth, and doesn't feel like a mobile game.

Civ 5 and civ 6 had pretty terrible A.I. and I have very little reason to suspect civ7 will be any different.

Completely agree on Old World. One of the best things about OW is even on the moderate levels, you get punished if you get sloppy with tactical strategy by the AI.

 

[Core Imposter]

It's basically impossible to create an AI for a complex game like this. And that's not changing. Sam Atman thinks we will have a superhuman AI in a thousand days and it will solve every problem in physics. None of that is real. You have to build out the AI situation by situation. To get what you want you'd have to increase the budget by 10 and dedicate 90% of it to AI development. And that might get you a competent opponent.

The smarter way is to build the game with flavor but with very limited complexity. Multiplayer and calculable paths.

 

[King Jason]

It's basically impossible to create an AI for a complex game like this. And that's not changing. Sam Atman thinks we will have a superhuman AI in a thousand days and it will solve every problem in physics. None of that is real. You have to build out the AI situation by situation. To get what you want you'd have to increase the budget by 10 and dedicate 90% of it to AI development. And that might get you a competent opponent.

The smarter way is to build the game with flavor but with very limited complexity. Multiplayer and calculable paths.

Not to beat the drum some more but respectfully, this is such a cop out and it's been prevelant in this genre and industry for years.

Genre too complex = good A.I. isn't possible.

This just isn't true. Will A.I. ever be perfect? No, probably not. As good as a human? Nah, no way.

Can it be **good**? Yes. Yes it can. Alex Mantzaris had demonstrated this twice. Soren Johnson also demonstrated this; twice.

I'm not just fanboying here, either - for the first 4 months of playing Old World I repeatedly got the absolute crap beat out of me because the A.I. is so relentlessly tactically sound in that game it's absurd.

Are there still weird moves, the occasional suboptimal play? Do bugs pop up now and again? Of course. But Old World has **good** A.I.

I would also rank Old World as more complex than civ5 and civ6. So while the game might not be everyone's cup of tea, I just think it is objectively untrue that strategy games, civ-style games, can't have good A.I. when there are games out there actively demonstrating that you can.

The broader issue is one of statistics.

Civ5 and civ6 have been the most popular in the franchise to date. Internal tracking stats in both games demonstrated repeatedly that the vast majority of players who play these games don't even play on Noble, the standard normal difficulty - they play on EVEN LOWER difficulties.

The reality is, but most strategy gamers don't actually want to admit it: Most players aren't looking for a tough, challenging experience from a genre of games that takes an entire evening or an entire weekend to play one single game.

Simply put: Firaxis just has zero incentive to make the civilization series have "good a.i." - because the majority of players who are going to play their game don't want it or won't utilize it. They don't care.

Civ is basically a beer and pretzels game and yet every 10 years people are shocked to rediscover that when they begin to think the game might be too easy.

Old World has a fraction of the budget of any civilization game, and lightyears better A.I. - This idea that firaxis somehow needs MORE money to fix this problem is just absurd: For their company and their goals - it's not a problem that needs fixing; Their customers aren't looking for good A.I.

 

[gdr_willter]

Well I'm not going to play CK Civ so let's talk about *The Civilization*

 

[King Jason]

Well, I am talking about civilization - it's probably going to have the same level of A.I. that the franchise has maintained for the last two iterations; mediocre at best.

But not because coding good A.I. is this elusive, ethereal, unattainable thing; not because the genre or the game is too complex - but because Civilization isn't a game that's designed for hardcore strategy gamers - but hardcore strategy gamers want it to be because we all grew up on it.

 

[King of Prussia]

Its very much possible for Civ to have good AI. It takes a lot of time and money to do so. You have to make the AI know how to use a system. You spend way more time improving how they use it, fixing weird edge cases, on top of usual bugs that will appear. Its very tedious work. Sometimes its a lot of work(and time) for a very small gain. Over the course of time, all those small gains add up. The performance cost for making the AI "smarter" also has to be considered. Is it worth having long turn times so the AI plays slightly better?

Civ 7 did change a bunch of core aspects, primarily ones the AI struggles most with. The new army commander system should make it a lot easier for the AI to get its troops into positions. This is one of the biggest issues 5 and 6 had. I disagree with Ed when he says the combat AI was fine, it was just the getting into position part that they struggled. They made some weird combat choices too. Other games do a lot better at this.

The change to how tiles are improved is also huge. They no longer need to make builders, and send them off to improve a tile. Instead its all managed from the City. Now you only have to focus on AI building management and using gold to fix tiles. Its also one less thing for the AI to factor when it does combat due to it loving free workers. Its also less units on the map to clutter it up.

The latter change is something i would of considered doing as well, if i made a strategy game. Of course, it wouldnt have the history of Civ for the last 30+ years to consider.


As for Old World. Its a very good game. Ive had a lot of fun with it. The AI puts up a good fight and makes some strong nations. The AI seemingly goes through every possible move its units can make before it does something. It can lead to them doing some weird unfun stuff too. The turn timers, for the few nations playing at a given time get longer and longer. By the late game, you are waiting a very long time between turns. Imagine how much worse it would be with out the action cap point system. This would not work very well with 8, 16, or more civs playing on a huge map.(assuming we get that post release) Im sure some of the more hardcore gamers would love it, and dont mind waiting. Would your average person? I dont think yout average civ player would, personally. How would this impact consoles and what not?

Modding would be a good way to help the AI too, but that requires those areas be open and easy to change. Probably wouldnt help those on non steam versions sadly.