Can we do anything to make Firaxis devote more resources to improving AI?

[Gedemon]

I don't think this would be a good thing if possible, this kind of AI would use any available exploit in the rules to win, and the game would then be very frustrating for the players.

Or it should be used only at some levels, like tactical military.

 

[heinous_hat]

So what you're saying is Firaxis puts out a half finished game knowing that modders will complete the job for free?

Effectively yes... but, without the nefarious intention to dupe customers and exploit the modding community. It's a shortcoming in their development and release process that hurts the finished product.

IMO, it's impossible to deliver a well tuned experience (with regard to AI performance) until the project is feature complete, stable, and reasonably bug free. Due to the universal practice of releasing games piecemeal... via DLCs and expansions... you won't have 2/3rds of those conditions until several years after release. The problem is that they don't commit the additional time at the end of development to work on AI, and eventually provide a capable opponent for reasonably competent players. They move on to the next project, which will have enough fundamental changes to gameplay, that will preclude much continuity in AI development. AI for a game like this needs to be specific, not some abstract learning mechanism.

Solution? Make the Civ series a continuous cycle of incremental releases. In other words, don't throw out the baby with the bath water.

My point with source code is that it isn't very common to be granted this level of access. Some modern titles are coded in C#, which can allow for decompilation and deeper modification from the onset. But that isn't nearly as useful as actual source code and project files, along with documentation. Firaxis, despite the shortcomings of their dev cycle, provide this. That's a huge perk.

 

[glider1]

I don't think this would be a good thing if possible, this kind of AI would use any available exploit in the rules to win, and the game would then be very frustrating for the players.

Or it should be used only at some levels, like tactical military.

That is an excellent point. There are other issues too.

Is a neural net code locked into playing the game it saw when Firaxis rented time on the supercomputer? Or could the unaltered neural net adapt to the inevitable series of balance and new features that occur to the game afterwards? For example, AlphaZero learned chess in 24 hours playing millions of games against itself to become the strongest ever player in history including all other AI hand-coded engines. But what would happen if they used that same neural net unaltered to play Chess960 where the rules are identical to chess but the initial setup is randomised?

Then there is the hardware cost. The neural net can run on PC's, but not on CPU's but TPU's which only the latest GPU's will be capable of. That will mean having to upgrade GPU card for almost every potential purchaser of civ as well as the cost of buying the game.

 

[Thormodr]

Effectively yes... but, without the nefarious intention to dupe customers and exploit the modding community. It's a shortcoming in their development and release process that hurts the finished product.

IMO, it's impossible to deliver a well tuned experience (with regard to AI performance) until the project is feature complete, stable, and reasonably bug free. Due to the universal practice of releasing games piecemeal... via DLCs and expansions... you won't have 2/3rds of those conditions until several years after release. The problem is that they don't commit the additional time at the end of development to work on AI, and eventually provide a capable opponent for reasonably competent players. They move on to the next project, which will have enough fundamental changes to gameplay, that will preclude much continuity in AI development. AI for a game like this needs to be specific, not some abstract learning mechanism.

Solution? Make the Civ series a continuous cycle of incremental releases. In other words, don't throw out the baby with the bath water.

My point with source code is that it isn't very common to be granted this level of access. Some modern titles are coded in C#, which can allow for decompilation and deeper modification from the onset. But that isn't nearly as useful as actual source code and project files, along with documentation. Firaxis, despite the shortcomings of their dev cycle, provide this. That's a huge perk.

The solution sounds like Europa Universalis and I like that train of thought.

And, yeah. Having the source code is a huge gift from Firaxis.

 

[Krajzen]

The solution sounds like Europa Universalis and I like that train of thought.

And, yeah. Having the source code is a huge gift from Firaxis.

At what point of civ5 development did they release source code? BNW?

 

[Thormodr]

At what point of civ5 development did they release source code? BNW?

Don’t really know, TBH. I abandoned Civilization 5 due to it being horrendous.

 

[omega4]

Effectively yes... but, without the nefarious intention to dupe customers and exploit the modding community. It's a shortcoming in their development and release process that hurts the finished product.

IMO, it's impossible to deliver a well tuned experience (with regard to AI performance) until the project is feature complete, stable, and reasonably bug free. Due to the universal practice of releasing games piecemeal... via DLCs and expansions... you won't have 2/3rds of those conditions until several years after release. The problem is that they don't commit the additional time at the end of development to work on AI, and eventually provide a capable opponent for reasonably competent players. They move on to the next project, which will have enough fundamental changes to gameplay, that will preclude much continuity in AI development. AI for a game like this needs to be specific, not some abstract learning mechanism.

Solution? Make the Civ series a continuous cycle of incremental releases. In other words, don't throw out the baby with the bath water.

My point with source code is that it isn't very common to be granted this level of access. Some modern titles are coded in C#, which can allow for decompilation and deeper modification from the onset. But that isn't nearly as useful as actual source code and project files, along with documentation. Firaxis, despite the shortcomings of their dev cycle, provide this. That's a huge perk.

Yes but when Civ 6 AI can't even use aircraft, which is a fundamental unit of the base game itself, that speaks volumes as to who's left coding at Firaxis.

 

[heinous_hat]

At what point of civ5 development did they release source code? BNW?

It started mid cycle, after the first expansion (2012). I think source code for both the original game and G&K was released simultaneously (or nearly so). There were then updates for patches, and eventually BNW (2013). Each game version has its own corresponding library, and is its own project, but inherits its predecessors.

Civ4 went pretty much the same way... comprehensive source code releases followed the first expansion

 

[universecreep]

Who's "we"?



Honestly, the main problem here is the lack of competition, there just isn't any game that even tries to compete with Firaxis in the "run through history"-type of 4x genre.

This is a great point.

I always wondered why there are so many space Civ games (stellaris, gal civ, endless space, others) but no game even close to Civ. I mean Firaxis does a pretty good job of the game but that doesn't explain why there shouldn't be any competition. If there was, perhaps Firaxis would go the extra mile with the fans instead of ignoring them until they need to sell us something and perhaps really try to push the envelope with respect to Civ.

The best hope I think is with Stardock. They seem to like the 4X genre. Perhaps with Jon Shafer on board they may think about making their own Civ-type game.

 

[Patine]

Buy up shares in their parent company? Otherwise, I'm tapped for quick and easy leverage over the developers...

 

[theadder]

The best hope I think is with Stardock. They seem to like the 4X genre. Perhaps with Jon Shafer on board they may think about making their own Civ-type game.

He already left Stardock and Paradox. He's useless.

 

[Thormodr]

He already left Stardock and Paradox. He's useless.

His own game, “At the Gates” is about 3-4 years late now.

Civilization 5 and its disastrous launch makes a lot of sense now.

 

[steveg700]

This is a great point.

I always wondered why there are so many space Civ games (stellaris, gal civ, endless space, others) but no game even close to Civ. I mean Firaxis does a pretty good job of the game but that doesn't explain why there shouldn't be any competition. If there was, perhaps Firaxis would go the extra mile with the fans instead of ignoring them until they need to sell us something and perhaps really try to push the envelope with respect to Civ.

The best hope I think is with Stardock. They seem to like the 4X genre. Perhaps with Jon Shafer on board they may think about making their own Civ-type game.

Well, there's Endless Legend, which I think it's safe to say Civ VI liberally lifted its district system from (and is probably its biggest departure from Civ V). So, the question to my mind is, why aren't we all of playing Endless Legend instead of Civ? Force of habit?

I guess if it's really an issue of pseudo-historical theme, then they could reskin it, which is largely just a matter of subtracting elements until the game is more mundane.

 

[bitula]

As a programmer with some AI coding experience and active Civ 6 player I have the following impressions:

1) The AI itself is fine, but full of bugs, which makes it bad.

2) The bugs are not necessarily (or only) in the AI code but may also be present in other parts of the game which interacts with the AI code, making AI look dumb.

3) Having only one AI programmers is fine, this type of work is difficult to distribute between more than one person.

4) Neural nets and self learning computers are neither required nor needed to write a good AI for a game. This whole topic is so lame…

5) The negative effect of complexity of the game on the AI code as discussed here is a huge exaggeration. It’s not that difficult to write a competent looking AI for a complex game, and certainly not impossible.

6) In general I see no signs of bad AI per se, because the cases I see are all individual cases which can be associated with a particular bug or deficiency (eg.: low usage of airplanes). But together they add up to a general experience of bad AI.

 

[pietro1990]

As a programmer with some AI coding experience and active Civ 6 player I have the following impressions:

1) The AI itself is fine, but full of bugs, which makes it bad.

2) The bugs are not necessarily (or only) in the AI code but may also be present in other parts of the game which interacts with the AI code, making AI look dumb.

3) Having only one AI programmers is fine, this type of work is difficult to distribute between more than one person.

4) Neural nets and self learning computers are neither required nor needed to write a good AI for a game. This whole topic is so lame…

5) The negative effect of complexity of the game on the AI code as discussed here is a huge exaggeration. It’s not that difficult to write a competent looking AI for a complex game, and certainly not impossible.

6) In general I see no signs of bad AI per se, because the cases I see are all individual cases which can be associated with a particular bug or deficiency (eg.: low usage of airplanes). But together they add up to a general experience of bad AI.

A big example of a bug that makes a bad AI is the joint war. AI value's it as a trade deal wich results in them declar war constantlly.

 

[Morgawr]

I'm terrified that the AI will not know how to cope with the new loyalty system and continue to found far away cities and constantly lose them to other civs and free cities, same way they lose settlers. Effectively providing yet another handicap to AI and a boost to the human player.

 

[steveg700]

I think the AI will just happily continue sprawling and settling cities the minimum distance apart.

 

[Wingednosering]

The thing is, a lot of people don't actually want better AI. As one of the first posts said, Civ is more sandbox than strategy. Most gamers don't have 8+ hours to spend (and I know a lot of people play matches for far longer) to lose.

People's time is so valuable these days, that the trend is for longer games to become more casual and games with really short play cycles (roguelikes, racing, minigame, etc) contain more lose conditions, because it doesnt make the user feel like they wasted their time.

If they wanted to improve the AI, it isn't terribly difficult. The issue is that civ just doesn't need to (and perhaps shouldn't?) be difficult anymore.

 

[Hauptman]

The thing is, a lot of people don't actually want better AI. As one of the first posts said, Civ is more sandbox than strategy. Most gamers don't have 8+ hours to spend (and I know a lot of people play matches for far longer) to lose.

People's time is so valuable these days, that the trend is for longer games to become more casual and games with really short play cycles (roguelikes, racing, minigame, etc) contain more lose conditions, because it doesnt make the user feel like they wasted their time.

If they wanted to improve the AI, it isn't terribly difficult. The issue is that civ just doesn't need to (and perhaps shouldn't?) be difficult anymore.

Nobody cares about "a lot of people". They can go play something else. The type of person who plays Civ does want to spend 8+ hours on a game, and would love to get beat on occasion. The plebs can stick with simcity.

 

[RohirrimElf]

Would probably have played civ 6 for 400-500 hours more in its first 6-8 months after the release if the AI would have been more challenging.
I do think firaxis cares about total number of players that play civ 6 on steam. It would have been much higher if the dedicated civ players like myself would stick to civ 6.
At that time i only had pleasure in civ 6 if i would focus on what i was doing. Meanwhile the ai just moved units around pointlessly. Not participating in the game really.
This has changed a good bit since the use of different mods and the last 2 big patches.
Would have been playing civ 6 right now if the release of the expansion was not that close. I just cant bear playing the game now that i know what is about to change in 3 weeks time.
Really excited about the biggest new mechanics (governors, loyalty, era score/golden ages, and emergencies).