From Easy to Deity. A new way to code the Ai difficulty?

[Lord Shadow]

Oof. We've discussed this at length over the past 15 years. I bet I can find older posts of mine that are still as relevant as ever.

My argument in its most distilled form is that truly smart ("good") AI requires a very desproportionate amount of resources, for little gain, given Civ doesn't have a competitive scene and hardcore players are always a niche. Assuming the same, finite amount of resources (people, time, money), you'd have to compromise a lot of the gameplay, its systems and its depth, just for the sake of AI. And the result may not be satisfactory even if the AI turned out more competent than the alternative.

Civ is among the most complex strategy games out there, light years ahead of chess or simpler computer games like StarCraft. It's also constrained by its turn-based nature, and having to strike a balance between AI and turn processing times. Nobody wants to wait minutes between turns. Hell, even 20-30 seconds would be grating, especially for the vast majority of the playerbase, who isn't looking to be really tested from a tactical/strategic standpoint.

Warfare in particular suffered the most due to the introduction and dogged adherence to the 1UPT model, which come Civ5 made a challenging task to the AI suddenly extremely hard.

Finally and unfortunately, there's one of Sid Meier's design tenets which rather gets in the way of truly prioritized AI in a complex systems context: computer opponents are meant to support the player's experience and give an illusion of challenge without really endangering the human civ's journey. I mean, as long as it's easy and cheap (Civ1-4), sure, make the AI challenging, but as soon as that task threatens the implementation of other parts of the game...


PS: Consoles and mobile gaming have, by and large, little to do with the 1UPT decision or the issue of AI. Nor does, really, any concept of "planned obsolescence".

 

[Acken]

The biggest issue with civ5 warfare was how overpowered range units are against an AI that over invests in melee units while you sit there with double strike promotions mowing them down. But the AI was never coded to reduce its flavor of melee vs range. Sure its pathing and coordination was far from great but just correcting how imbalanced units were would have gone a long way to make warfare against AI a better challenge without having it be more intelligent. And that investment is not huge.

 

[Ebon-Hawke-]

I watched boesthius civ 7 video, he was playing on immortal and it seemed quite competitive, or at least AI seemed very aggressive in regards to warring and were beating him in some of the non war pathways as well. He ended up getting attacked on both sides and pretty much giving up on the playthrough.

 

[Lord Shadow]

The biggest issue with civ5 warfare was how overpowered range units are against an AI that over invests in melee units while you sit there with double strike promotions mowing them down. But the AI was never coded to reduce its flavor of melee vs range. Sure its pathing and coordination was far from great but just correcting how imbalanced units were would have gone a long way to make warfare against AI a better challenge without having it be more intelligent. And that investment is not huge.

The Civ5 AI's inability to effectively use ranged units, the concept itself yet another new rock in the AI's rucksack, was definitely a factor.

The arguably greater one that stuck with me, however, was the fact that at its peak, while Civ5 AI could manage to compose a fighting formation (if the war started on their terms), it was a house of cards. The moment you managed to break the back of that singular army, a moment of fleeting satisfaction, that whole empire was wide open, up for grabs. The challenge evaporated. There was no way that opponent could reform an effective defensive force under pressure and active attacks before it had lost large swathes of territory. By which point the conflict was most likely decided.

 

[Ornen]

a lot gets lost in the AI discussion, but there's a difference between highly intelligent AI and competent AI that plays to win.

in base Civ 6, the AI does not generally try to win. this is why a leader like Pericles is much more dangerous than the competition: he builds culture districts in every city, which is the most fundamental strategy to win a culture victory.

most AIs simply do not do this. with a handful of exceptions, an AI will not build science districts across the board, or culture districts across the board, or go on conquering sprees to take over their weaker neighbors and snowball towards a conquest win.

programming an AI to do these things is not that complicated. in fact there are existing mods for Civ 6 that change the way the AI behaves, and in my experience, they raise the difficulty by 2-3 levels.

the current system — give the AI big advantages out the gate — has its issues. in addition to what has been said, it makes the early game very challenging, but mid-to-late game very boring. once you get ahead of the AI, the win is yours.

I don't think the AI playing to win is too much to ask... and done right, it can make the game feel more balanced across all difficulty settings.

(there is an additional level on top of 'playing to win', which is 'playing to stop opponents from winning'. Civ 4 AI actually did this as well. but this is harder than just playing to win, and players are likely to enjoy it less)

 

[Taefin]

Optional patches with clear indications of their scope and purposes could help identifying dangers, although I had a hard time interpreting your concerns, some of your's could be interpreted as advantages?

Wow, yeah, that was a confusing sentence. I suspect the biggest challenge the player will face will be having an AI snowball who they don’t border. If AI pursues the victory conditions and starts its endgame project, this would often require the player to intervene against that AI. I think this might be more likely than late game AI threatening the player’s empire militarily. I think this will create fun games.