• Civilization 7 has been announced. For more info please check the forum here .

Likelihood we'll get a decent AI

ThunderLizard2

Warlord
Joined
Oct 6, 2010
Messages
271
How likely do people think it is that Firaxis delivers a decent AI. In the age of Chat GPT, is it too much to expect? And if no decent AI is provided, will they at least open up dll so we can get modders to fix the AI like in CIv V?
 
ChatGPT can't even play a legal chess game, what makes you think it's existence would be beneficial to the quality of a Civ AI?

Machine learning might be able to benefit the Civ AI, if properly applied, but it is questionable whether this is financially feasible, as well as whether it's desirable. Most likely, there won't be any machine learning for Civ VII or even a successor of it.

The question is more how much work they put into the AI they do create, as well as of course the exact mechanics the game features, which may impact the strength of the AI because some aspects may make it easier or harder to code a competent AI.
 
Well, for ChatGPT and similar technologies, there are usecases https://www.galciv4.com/article/520331/galciv-iv-supernova-dev-journal-20---how-people-use-aliengpt .
I think similar (non-hallucinating) technologies could be very interesting for the diplomacy menu.
For the underlying tactical decisions, that depends.
Researchers have before trained an AI for FreeCiv on its manual, see http://groups.csail.mit.edu/rbg/code/civ/ . For a more complex game this would have been prohibitive, but with the current options... it might be interesting to see if Firaxis will integrate a way that you may provide your game as feedback to have an AI trained. Having a "community trained AI" as separate option after deity would certainly be interesting.
 
Well, for ChatGPT and similar technologies, there are usecases https://www.galciv4.com/article/520331/galciv-iv-supernova-dev-journal-20---how-people-use-aliengpt .
I think similar (non-hallucinating) technologies could be very interesting for the diplomacy menu.
For the underlying tactical decisions, that depends.
Researchers have before trained an AI for FreeCiv on its manual, see http://groups.csail.mit.edu/rbg/code/civ/ . For a more complex game this would have been prohibitive, but with the current options... it might be interesting to see if Firaxis will integrate a way that you may provide your game as feedback to have an AI trained. Having a "community trained AI" as separate option after deity would certainly be interesting.
Well, you speak of two things:
- Having AI that can (somewhat) converse with us, more than the usual pre-recorded message we had so far would be nice, if the AI stay in the correct caracter (a Pierre II that with an aristocratic tone, etc...). It shouldn't be too difficult to do, but we would have automatic voices, thing that some find rebarbative...
- An AI that can act strategically: day one, the chance to have such AI are next to null... Machine learning takes a helluva of data, correctly labelled (you have to tell it "bad behaviour", "good behavior", "big/little mistake", and so on...). And some actions are bad short term but allow you a great return on invest later. Hard to indicate to the IA which action led to benefits...
However, if players allow Firaxis to have access to their games, we may have a addaptative AI by the thirst or second expansion (which would be a huge commercial argument).
 
Generative AI like ChatGPT has very little to do with game AI, as far as I'm aware. Still, I think the advancements in the field should make it much easier to create good game AI.

Unfortunately, good AI has never been a priority for Firaxis with Civilization, and I have no reason to think that this will change. But there's no way to know at present.

However, if players allow Firaxis to have access to their games, we may have a addaptative AI by the thirst or second expansion (which would be a huge commercial argument).
Dont give them any ideas, please.
 
Last edited:
It likely depends on what you mean by "decent AI".

I'll share the 3 things I always indicated I wanted when we had this conversation about Civ 6, and a 4th I've added after the experience of playing Civ 6:

1. I want the AI to act as a pace car. The speed of the pace car should depend on the difficulty setting, but I should have a reasonable idea, at the outset of the game, that if I don't win within X turns, one of the AI is likely to have won before me.

2. I want the AI to act as a speed bump. I don't want the AI to passively ignore what I'm doing, I want it to take steps to mess with my plans and make me adapt.

3. I want the AI to enhance flavour and re-playability. Having Shaka as your neighbour should be a different experience than having Gandhi as your neighbour.

4. I want the AI to avoid player-obvious, immersion-breaking incompetence. I don't care if the AI can't play all the game mechanics "well", as long as it doesn't blatantly display the things it does poorly continually and regularly to the player.

I'd like to think these are reasonable goals for Civ 7, but I'm not sure any of these will be priorities for the current design team, so who knows.
 
It likely depends on what you mean by "decent AI".

I'll share the 3 things I always indicated I wanted when we had this conversation about Civ 6, and a 4th I've added after the experience of playing Civ 6:

1. I want the AI to act as a pace car. The speed of the pace car should depend on the difficulty setting, but I should have a reasonable idea, at the outset of the game, that if I don't win within X turns, one of the AI is likely to have won before me.

2. I want the AI to act as a speed bump. I don't want the AI to passively ignore what I'm doing, I want it to take steps to mess with my plans and make me adapt.

3. I want the AI to enhance flavour and re-playability. Having Shaka as your neighbour should be a different experience than having Gandhi as your neighbour.

4. I want the AI to avoid player-obvious, immersion-breaking incompetence. I don't care if the AI can't play all the game mechanics "well", as long as it doesn't blatantly display the things it does poorly continually and regularly to the player.

I'd like to think these are reasonable goals for Civ 7, but I'm not sure any of these will be priorities for the current design team, so who knows.

It's hard to say, especially because often those basic points contradict each other. Just looking back at the past games of the series, in broad strokes:
In civ 4, the AI had heavy personalities, but they were predictable. If you kept Hatty with the friendly icon as your neighbour, you could literally have no units in your empire and you were 100% certain she would never attack you.
In civ 5, the AI played the game a lot more. So while they were more of a speedbump, you would often have cases where you were friends forever and then suddenly late in the game your friendship would break down and they would eventually declare war, because they now acted to prevent you from winning.
In civ 6, they sort of took a middle ground, which in a lot of ways just made personalities weak. I mean, sure, it's nice to see friendships return and all, but it almost feels random. Maybe I just never paid close enough attention to AI agendas, but it seems like AI sometimes just hate you on turn 1 no matter what the situation.

To me, the harder part is making sure that the AI is at least competent. Even ignoring personalities, I want to make sure if an AI declares war on me, they can physically pose a threat to me. Early on in civ 6 the AI would do dumb things like constantly shuffling catapults around, moving them into the water while sieging a city, etc... or come at your city with 6 ranged units and like one scout, so as long as you kill the one unit that can capture the city, you can let them sit there forever. Or if I declare war and am invading them, I want them to be fighting back and posing a threat to my units. Even at the end of the cycle, it feels like sometimes the AI just sits there without actually shooting back, and doesn't pose a real threat.
 
I'd say that there's a 0.01% chance that we are getting a better AI. The AI is most likely a copy'n'paste of the civ6 AI modified to work with civ7 features. I don't know if they are working under some internal target regarding the turn times, but they have certainly been unwilling to improve civ6 AI in its long life time. I would personally prefer longer turn times, if only the AI got better and used fewer cheats.
 
In civ 6, they sort of took a middle ground, which in a lot of ways just made personalities weak. I mean, sure, it's nice to see friendships return and all, but it almost feels random. Maybe I just never paid close enough attention to AI agendas, but it seems like AI sometimes just hate you on turn 1 no matter what the situation.

My Civ VI experience with the AI does seem rather leader-independent, but I wonder how much of this is simply a result of playing on Deity. The first meeting bonus/malus depended on difficulty level, and even the most positive variance in the malus pretty much ensured initial dislike on Deity. That basically led to 'ignore the AI for a while, then when the frowning face disappears, make friends with them', or alternatively, as I eventually found out, on the turn of meeting they'll almost always accept a delegation, you can then also gift them open borders (gift + having open borders both providing a relationship boost) and perhaps gift some money to get an immediate declaration of friendship. And then, AI will (almost) always accept a request for a declaration of friendship if you send it on the turn the previous one ran out.

I guess the summary is just that the system was too rigid, at least on Deity (more variety may be possible on lower difficulties, which might be leader-dependent for all I know), and too easy to game.
 
The only way to get a better AI is to build a new one, from the ground up.

avoid already existing AIs - including programming AI bots. They make it easier for the programmers, but only if their skill level os low. The truth is that a human will always do a better job than an AI if they're trained well. Invest in programmers that know what they're doing, guide and train them, test, test, test, and you should be good.
 
My Civ VI experience with the AI does seem rather leader-independent, but I wonder how much of this is simply a result of playing on Deity. The first meeting bonus/malus depended on difficulty level, and even the most positive variance in the malus pretty much ensured initial dislike on Deity. That basically led to 'ignore the AI for a while, then when the frowning face disappears, make friends with them', or alternatively, as I eventually found out, on the turn of meeting they'll almost always accept a delegation, you can then also gift them open borders (gift + having open borders both providing a relationship boost) and perhaps gift some money to get an immediate declaration of friendship. And then, AI will (almost) always accept a request for a declaration of friendship if you send it on the turn the previous one ran out.

I guess the summary is just that the system was too rigid, at least on Deity (more variety may be possible on lower difficulties, which might be leader-dependent for all I know), and too easy to game.

Civ 6 definitely front-loaded the challenges when playing on Deity. If you survived the early game, after that the game didn't play that much differently from lower difficulties. I'd prefer a new approach with Civ 7, with the higher difficulty level challenge focused on out-racing the AI in the mid- to late-game, rather than on how well you can turtle and play tower defense in the early game. Giving the AI an additional bonus every time it enters a new era would, I think, be better than just giving the AI more and more at the start of the game.
 
I think that the goal for Civ 7 is an AI that can challenge and win against deity level players without needing bonuses.

What is a deity level player? The deity player knows the advancement trees very well. The better they are familiar with the trees, the better they can plan their cities and empire. The deity player understands how diplomatic relationships are influenced and make moves that exert some degree of control over who their allies and enemies are. They understand what poses threats and how severe the threats are, and what is threatened. For example, a barb camp close to horses is a different threat than a barb camp that is not close to horses. If it is close enough to a non capitol city, the city center is threatened. Infrastructure is threatened. Builders are potentially hampered.

There is probably more to it even than that.
 
The Computer Adversaries (I'm using this term to differentiate from the art, writing amd programming AIs that have become a fad recently) will probably be just like they've always been: crap at launch, improved over time, but always ever designed to provide entertainment to the average Civ customer who bought the game because they saw Cleopatra in a YouTube video thumbnail and vaguely recognize her name from their middle school world history overview.
 
What is a deity level player? The deity player knows the advancement trees very well. The better they are familiar with the trees, the better they can plan their cities and empire. The deity player understands how diplomatic relationships are influenced and make moves that exert some degree of control over who their allies and enemies are. They understand what poses threats and how severe the threats are, and what is threatened. For example, a barb camp close to horses is a different threat than a barb camp that is not close to horses. If it is close enough to a non capitol city, the city center is threatened. Infrastructure is threatened. Builders are potentially hampered.
They need skilled beta testers. There are plenty of really good civ 6 players out there that they can approach to help them with that.

I believe they did exactly that during the development of Civ4. Which makes sense to me since, Civ4 is by far the hardest Civ game.
 
They need skilled beta testers. There are plenty of really good civ 6 players out there that they can approach to help them with that.

I believe they did exactly that during the development of Civ4. Which makes sense to me since, Civ4 is by far the hardest Civ game.
A few of the testers from Civ 4 are still in the testing group and also tested 5 and 6.
 
This is irrelevant. Aside from that, you and me as consumers and players do not want a good AI. Studies show we automatically think a good AI is unfair and/or cheating. ;)
No, we definitely want a good AI. Speak for yourself but when the game is basically just SimCity with the AI being little more than comedic relief & actors to simulate immersion it creates pretty boring gameplay. It would be nice to feel like there were other civs actually in the game world with plans and working toward them.
 
  • Like
Reactions: Xur
Top Bottom