I was chuntering over this again this afternoon, and realised that it’s a cyclical thing: I felt divorced from Civ 4 because I bonded strongly with 3, and doubtless fell for 5 because I lacked that relationship. I wonder if that means that I will hate 6 while everyone else goes gaga over it?

I think the reason why I’m fighting this one so hard is because a) I’m a Civ V nerd, granted but also b) all of my favourite moments from the 400 odd hours are related to the AI. Like, the three or four games where I have had a reliable ally who has backed me to the hilt, from beginning to end, so that I felt genuine affection for them at the end.

Oh, and finally: you can press M for move mode, and hold down the click button to see the path and move time estimates. Fnar.

You know, I think you’re largely right. While not exactly sour grapes, this did come about because I was trying to play Civ V the way I did Civ IV. But then I think I’ve mentioned before how Civ IV’s game design was an iterative process building on what did and didn’t work in previous games, and I was kind of expecting Civ V to continue that trend rather than throwing the baby out with the bathwater.

(The baby is a documented diplomacy system.)

Also your SI point: it might surprise you to learn that that RPS game *was* my Solium Infernum learning experience. The reason it was so awesome was because literally every other person playing it was in exactly the same boat and we were all making it up as we went along. If I’d been playing with five seasoned veterans who skipped straight to using the puzzle box to teleport legions to Pandemonium I wouldn’t have anywhere near as high an opinion of SI as I do, and also it would have made for a really boring write-up.

Also also, up until about 8pm last night I was telling people in Teamspeak that Civ V was a game that could sit comfortably alongside Civ IV as something separate and worthwhile. I didn’t go into it with a hostile mindset at all.

Okay, I mean, I think I’ve been really clear and explicit throughout that having this system being undocumented is problematic, so I don’t think you’ll be surprised to note that your point is not exactly risky, for all that it’s a world away from the point you made in the actual post.

That said, you’re still pushing your position further than it supports. Yes, I am a 400-hour fanatic, but that doesn’t mean that it takes anywhere near 400 hours to grock the system (I didn’t, for example, only start intuiting invasion threats ten hours ago – if I had then Civ V wouldn’t have made it this far on my hard drive either). And, at the other extreme, we have an example of a counterfactual: Kenti with Solium Infernum, whose professed desire is to not have to learn systems by failure, repetition and learning, and whose position you broadly disagree with. Given those two extremes, my question is this: if you are content to learn SI by failure and repetition, but not Civ, then where’s the sweet spot? Where’s the middle ground? How much on-the-job learning is enough, and how much is too much? (Frankly, given that you can play a game of Civ V at Prince with no greater understanding of its diplomacy structure than a babe in arms, and still win handily, Civ’s diplomacy seems easier to grasp, to me, than SI’s mechanics – but that’s just me, and as suggesting that you may possibly have come to Civ V, on this occasion, looking for flaws rather than possessing the charity of an open mind would be unreasonably hostile I will have to put that down to the vagaries of personal taste.)

But your comment about difficultly levels is apt, and got me to thinking while walking home. Two points arise from it. Firstly, Civ V’s difficulty level is whack, on much the same grounds as Civ IV’s was. There is graduated AI capability – Chieftain AI is noticeably stupid – but it peaks at around Prince or King, and further difficulty levels are managed purely though hidden advantages. As a result, I’ve never even attempted to win at Deity level. It’s just no fun. And that’s genuine bullshit – as is the fact that the game needs to give its AIs a few advantages to be even moderately playable (Prince, the so-called default, is laughably easy).

The other is that, of course, your whole argument misses the point hugely, in that new players are going to start at Chieftain, not King, and are going to develop and understanding of how to play the AI organically as they work up. This is purely a Civ IV purist problem, I’m afraid. If you come in at King then you will get turned over.

(Aside: On Saturday, while watching over my shoulder, I had occasion to use the Denounce button on Washington, and you chortled and said something to the effect that it was just spite. I kind of mentioned it at the time, but when you say “There’s no reasoning with an AI, though. Negotiating simply isn’t possible except in so far as the “What will you give me for this” button allows you to get a reasonably fair exhange for your goods” it suggests that you haven’t explored this part of the system yet. Because denunciations and declarations of friendship are actually very precise tools in Civ V’s internal diplomacy. A Declaration of Friendship is more than flavour; if an AI feigns friendship, signs a DoF and then attacks then that is a meaningful betrayal that some of them will baulk at doing. Denouncing an AI will mean that their friends like them less and your friends like you more. Either of these measurably affect the probability that war will occur, and, again, it’s something that you would learn organically while trying to master the game at Warlord. There are negotiating tools; there are ways of dissuading an attack, and encouraging one. They aren’t documented and that’s a shame, but they are there, and they work as input methods.)

Finally, I don’t think you’ve addressed one of the key points of my last comment, which is: how do you distinguish between a “smart” AI which plans in the way described above, and a “dumb” AI making random moves that appear the same as the smart AI?

I mean, yeah, you’re courting the foul here, and that’s fair enough, but I’ll take the hit. The obvious answer is still the answer: “play for 400 hours and you’ll just know.” Your Napoleon / Suleiman gambit is one that I have seen a hundred times, and fallen for more than once. I could write a comment about Civ V AI War Prep Indicators but it would be boring and long-winded. But you can’t play a game for 400 hours and not come away with an idea of what the AI is doing. If it was random I would know.

This goes back to the difficulty post I did on Saturday. Difficulty and teaching the player how to play the game are inextricably linked. Civilization V makes no effort to tell the player that the AI lies. It makes no effort to tell them anything about the diplomacy; a year after release and you’re still plonked down and expected to figure out research and trade agreements and what they’re worth to the AI all on your lonesome. If it were a human, I could use empathy to try and put myself in the other player’s place. I could try to reason with them, negotiate with them, to try to get them to accept a given deal. There’s no reasoning with an AI, though. Negotiating simply isn’t possible except in so far as the “What will you give me for this” button allows you to get a reasonably fair exhange for your goods. Since the numbers are all locked away behind the scenes it’s impossible to even begin to understand how the AI is thinking without the oodles of experience you mention — and it’s thinking in a way fundamentally different from that of a human, so empathy is no help there.

Finally, I don’t think you’ve addressed one of the key points of my last comment, which is: how do you distinguish between a “smart” AI which plans in the way described above, and a “dumb” AI making random moves that appear the same as the smart AI? One is about to attack you, the other is the standard Civ troop movement AI behaviour. This isn’t a problem so much with human players since there’s always a person at the controls. With the AI, there’s just an algorithm.

PS This is getting rather thin. Start a new comment thread if you reply!

Consider this: you would recognise the pre-betrayal behaviour of Josh or Feet not because all humans have exactly the same betrayal patterns, but because you know Josh and Feet and know that if Josh is bothering to build an army it’s because he thinks he can use it. Consider also this: that after 400 hours with the game, an experienced player can tell when he’s about to get hit by the AI. Again, I’m keen to accept that it’s almost impossible to gather that understanding without considerable play, and that’s problematic, but will you at least accept that it is possible that there is a whole layer of nuance that you just haven’t got to yet? Because I can assure you, it’s there.

Numbers are only necessary if it is totally impossible for the player to get feedback through any other route. It isn’t, so they aren’t.

And I wasn’t being unfair, because even if you did expect Napoleon’s betrayal, too still allowed yourself to get caught up in a war of choice. Motivations can be argued back and forth, but the thing is that his plan worked, and that’s awesome, because it’s exactly what AI is supposed to do.

*That* is why the numbers are necessary, and that is why coding an AI that tries to produce a crude facsimile of a human attempt to lie was a really bad idea — it is fundamentally different from a human being, it’s not capable of nuance or subtlety or any of the other myriad other things that might tip me off that something was up, and so I really need to be told what it thinks of me if I want to be able to plan for the future beyond “Wipe out every AI player because they might not be telling the truth about this peace treaty they want.”

(Also I think it’s slightly unfair to characterise me as a rube, there. I *didn’t* trust Napoleon precisely because of his sudden chumminess, hence the border forts + citadel garrisoned with a third of my army.)

This, though, is incorrect:

It’s just that if the AI can declare war on you at the drop of a hat, with no build-up and with a diplomatic opinion that’s read “Friendly” for the last fifty turns, then there is no way to plan for it.

Because the AI is programmed to lie, sometimes, but it is still AI, so you can tell when it’s lying.

In the example you cite in the piece, to be quite frank, the problem wasn’t the AI, it’s that you were a putz. If you’d been at war with Napoleon, and he suddenly switched to friendly and asked you to get enmired in a war with a third party, why on earth would you go along with that? That is actually an example of the diplo AI being awesome, because to be honest it saw you coming and took you for a rube.

The point you would have, if you had made it, is that the information that the attitude markers give is opaque if you don’t know how to read them, and the lack of documentation means that they only yield to trial and error. But the AI does not act at random and the leader attitude markers are not useless. They are information. They are just not trustworthy. They respond to stimulus in a logical manner. If you have been at war with a guy, the war ends, and then he switches to friendly, here’s what you do:

1. Don’t trust him. You’ve been at war with this guy, unless you’ve bought him off what is he doing being friendly?
2. Immediately denounce him. That will a) garner you some international sympathy and a boost in relations with your allies, and b) lead to the following two steps.
3. Observe the relations marker for the next few turns. If he drops to guarded then it worked; he’s probably deterred from attacking you again soon.
4. If he stays friendly despite the war and despite being denounced then get ready for round two of the war because it is coming.

The developers tried to create an AI that simulates, to an extent, humans who want to be deceptive. That is a noble aim: Civ IV was deeply annoying for its need to document every +4 modifier and reduce all interpersonal issues to an algorithm. It made balancing relations a tickybox exercise rather than something genuinely responsive. Of course, you can’t create an AI that replicates human mistrust. It’s a fudge, and play it for long enough and that becomes obvious and annoying in its own way. That’s not what the problem you’re describing is, though. The problem you have is that you’re expecting the AI to act like an AI, and expecting your enemies to give you reliable information. You wouldn’t expect it if the other player was human so why should you expect it from the AI?

Also the instantly heal in the middle of combat thing is monumentally stupid.

But hey ho. You don’t like it and that’s a pity! At least Mass Effect 3 is nearly out.