Artificial Stupidity

[Miserable Old Git]

I can understand why the "AI" fails in a military sense; it uses if-then-else logic, when the permutaions are so great that only a proper AI could do the job decently.

But why, oh why, does it allow you to offer 1 luxury in fair exchange for 5 (on deity level); or expect you to pay to join a joint war that it (and not you) wants to wage?

 

[pokiehl]

It's not inconceivable to consider trading 1 luxury for 5 luxuries (which I've never seen). If the AI happened to be flush with extra luxury copies but really needed a few more amenities from a new luxury, why wouldn't they just trade? They lose nothing by trading excess copies and gain some amenities.

 

[dagriggstar]

My personal 'favourite' is pointless late game wars that predictably happen because the "mayhem" counter has been activated. The unwritten rule where if you're 'winning' you can't be at peace with every faction so a random neighbour will declare war on you even though they are two eras behind and pose no threat whatsoever.

 

[Linklite]

It's not inconceivable to consider trading 1 luxury for 5 luxuries (which I've never seen). If the AI happened to be flush with extra luxury copies but really needed a few more amenities from a new luxury, why wouldn't they just trade? They lose nothing by trading excess copies and gain some amenities.

Because they could trade them for something else? Or even, they're in competition with you, so every Luxury they give you, even at zero immediate cost to themselves, is helping a rival?

There are (somewhat niche) scenarios where it makes sense. For example, if the player is refusing more balanced offers, then sure, it makes sense. I'm not getting impression that's what's happened though.

 

[pokiehl]

Because they could trade them for something else? Or even, they're in competition with you, so every Luxury they give you, even at zero immediate cost to themselves, is helping a rival?

Maybe there’s nothing else they needed? Maybe their neighbors already had the same luxuries so they had no valid trading partners otherwise? What sense does a zero-sum competitive mindset make if they make a trade to avert a disaster? Maybe they’re so far ahead of the player it doesn’t matter that they get helped?

Point is there are plenty of valid reasons to make seemingly lopsided trades.

 

[Ordnael]

AI has so many production an gold bonuses in deity lvl that money is just laying around, so they don't care much.

 

[Oberinspektor Derrick]

AI has so many production an gold bonuses in deity lvl that money is just laying around, so they don't care much.

That's not the reason.
Even an AI that is virtually broke will often offer outlandish exchange rates (even all of their gold reserves, if available) for simple things like a luxury.

 

[BackseatTyrant]

I think huge part of it is that no one (not just at Firaxis, but anywhere) has really been able to figure out how to make computers context-sensitive. IIRC, pretty much all automated systems (that includes the AI in any video game) are coded to only react to the data they've just been fed, and that data is always going to be reacted to in extreme isolation

 

[Bösewicht]

Sometimes I think that the AI is actually coded to make certain deliberate mistakes for the sake of the overall unfolding story. Playing as the Aztecs, a scout will suddenly pause and continue to pause near an eagle warrior. After a focus on an early wonder and a subsequent lack of wider development, an unaccompanied settler appears, conveniently making up for the opportunity cost. Somehow the game feels a bit like a cheat afterwards. At emporer level.

 

[aieeegrunt]

They really, really need to, SAY IT WITH ME PEOPLE, make it like some of the older civs and make settlers non capturable. It either just dies or becomes a worker

 

[crocivfanatic]

I think huge part of it is that no one (not just at Firaxis, but anywhere) has really been able to figure out how to make computers context-sensitive. IIRC, pretty much all automated systems (that includes the AI in any video game) are coded to only react to the data they've just been fed, and that data is always going to be reacted to in extreme isolation

Yep, AI is not even accurate word for it, it should be named artifical pattern seeking . For example, when i played aoe4 against second level of dif, it was hard for me beacuse ai spawn units always faster. But then i noticed weakness - if i make a "hole" in wall, and put massive defense on it, AI will always rushing to that hole and completely ignore fact how that hole is heavily defended while rest of wall does not have any defense behind it. On civ, key to beat AI on harder level is fast attack, before it starting to utilize dif advantages.

 

[Miserable Old Git]

I'm confused. We're being bombarded with scare stories about how AI is going to take over the world. ChatGPT is apparently not far short of passing the Turing test. Yet people here seem to be saying it can only handle static as-is data, and not be able to adapt and learn from its mistakes, even in a relatively trivial two-dimensional game environment. Is current "true" AI just too expensive to be built into a game, or am I missing something?

 

[BackseatTyrant]

I'm confused. We're being bombarded with scare stories about how AI is going to take over the world. ChatGPT is apparently not far short of passing the Turing test. Yet people here seem to be saying it can only handle static as-is data, and not be able to adapt and learn from its mistakes, even in a relatively trivial two-dimensional game environment. Is current "true" AI just too expensive to be built into a game, or am I missing something?

The short of it is that AI doesn't exist, and any company who calls their tech AI, is using this obscure mass psychological trick known as 'lying'. And would you believe me, if I said that the apocalyptic scare stories were also part of the tech companies' deceitful marketing campaigns? Like @crocivfanatic said, what we call AI in video games are really just pattern-seeking algorithms that with enough time & dedication, the player can eventually learn how to exploit

 

[pokiehl]

The short of it is that AI doesn't exist, and any company who calls their tech AI, is using this obscure mass psychological trick known as 'lying'.

AI absolutely does exist and it is absolutely rapidly changing the world. Disputing the existence of AI is just going down the “no true Scotsman” fallacy.

“True” AI has to have some capacity to “learn” from inputs” and by that definition it’s true that videogames don’t really employ AI, since game AI is by and large static programming.

I’m not sure that an adaptive AI is even desirable for gaming. At the least, most of us can agree we want our artificial opponents to be able to play the game and use all of its systems.

 

[Miserable Old Git]

@pokeihl

Agree on all counts, with one reservation.

True AI wouldn't be desirable for gaming; it would win every game (Deep Blue beat Garry Kasparov 27 years ago). However, it could surely be programmed to replicate the current "difficulty level" system by turning off its learning mechanisms after x minutes (retaining what it's already learned), depending on the difficulty selected at the start of the game.

 

[Naokaukodem]

Actually... I feel like I'm playing like the AI, and that's why I struggle with it in Deity when I take care of what I'm doing. (otherwise I get completely eclipsed in research) Should I call myself Non-artificial Stupidity ? Sounds like just Stupidity heh ? Anyway all this to say that I find Civ6 pretty much complicated and forces me to THINK, thing I must say I'm not accustomed in video games at least. I guess also that I play the game as is intended to be played by the devs, I use basically 0 exploit in my games, for I don't even know them, eventhough they must be numerous in a game with such complexity, depending also of what you call exploits. But for example I never sabotaged ALL my enemy's spaceports to prevent him to reach the victory, seems like to much hassle and anyway I often neglect my spies, and when not, they fail miserably in stealing gold for XP or there is simply no known target to do so.

My gripes with the game are, non-exhaustively :
- bad production city spots
- not enough city spots due to c-S and barbarians (I desactivate them all now, just try to find regular trade partners) and mountains and deserts and water and toundra and ice.... and even rainforests are not so good as a starting location.
- early warfare in Deity nearly impossible, just I like I feel it is in Civ5 Deity ; this has not changed much, therefore no fun.
- Disappointing situations where you fixed you an objective at the start of a game and couldn't reach it. (for example, having a religion and keep it or even try to spread it abroad) Bad luck because AI sends tons of Apostols this time. And that's just an example : whenever I fix myself any objective, I generally fail due to bad luck. This game is so spread out that it creates the conditions for disappointment. And maybe unplayability in Deity too.
- But hey, famous youtubers beat Deity everytime so I guess it's fine... not. The most famous Civ6 youtubers are actually the luckier ones because they feel good and everybody wants to learn from them. But hey, it's just luck you know. Marbozir, PotatoMcWhiskey, the SpiffingBrit, etc. all guys with bad love situations IRL if you ask me. (j/k : "heureux aux jeux, malheureux en amour" = "happy in games, unhappy in love") They are all like les ravis de la crèche and play the game huh... in a surprisingly naive way. (damn I have so few gold... let's build markets, then ! And their problems disappear - I've tryed to do the same but it doesn't work because the more units you build, the more they cost, and the more you upgrade them, the more they cost, so dealing with dedicated districts that are supposed to give you gold, and very little and are very long to build, not talking about the pop cap, and the buildings you have to build inside on top of that, make the efficiency of such naive tactics debatable)
- Etc.
- To sum it up Civ6 is too spread out, and has too many factors for bad luck, including the whole "unfolding of events" that makes saves useless unless you take a save 100 turns before if it exists.

 

[aieeegrunt]

The various arguments like “well real AI would be unbeatable” or “true AI doesnt’s exist” or “AI for a game like Civ6 is impossible” are just copes to avoid the simple fact that this is easily the worst AI in the series, by a mile.

I’ve been playing Civ since it was literally a cardboard based life form, unmodded Civ6 AI is terrible. The experimental solitaire mod for the orginal boardgame is better.

It’s can’t even get basic things like where to settle correct; how many awful useless AI cities have we all had to raze in our time because the AI will settle in Tundra one tile away from fresh water and kiler just yields outside the basic city ring? Or idiotic disttict placement? Or spamming settlers while losing a war?

I swear just deleting the the fact that the AI can see all resource locations at start would help a lot, and five minutes of playtesting should have revealed that

It’s even more enfuriating because so much of (unmoddef) Civ6 boils down to deterministic numbers driven munchkin minimaxing; this is the exact thing a computer should excel at.

 

[Arcaian]

AI absolutely does exist and it is absolutely rapidly changing the world. Disputing the existence of AI is just going down the “no true Scotsman” fallacy.

“True” AI has to have some capacity to “learn” from inputs” and by that definition it’s true that videogames don’t really employ AI, since game AI is by and large static programming.

I’m not sure that an adaptive AI is even desirable for gaming. At the least, most of us can agree we want our artificial opponents to be able to play the game and use all of its systems.

The machine learning being used in what is currently being called AI is functionally as intelligent as autocorrect on one's phone is; it is capable of learning from inputs, but calling it AI is a huge marketing stunt to try and make people think it is somehow fundamentally different from algorithms we are more used to interacting with in our day-to-day life. That being said, I do agree that applying machine learning algorithms to a video game's AI is not the boon people expect it to be - people hugely underestimate just how many decisions are made across a game of civilization, and how difficult that would cause it be to make a ML-based AI for a game like this. It might be more appropriate for very specific circumstances, but it also runs into the issue of changing the game - you really don't want to have to retrain your whole AI model when you make significant balance changes in development/in patches. On top of that, I do agree that adaptive AI might not be what people want - if your AI is learning how to counter you specifically, it could very much feel like you're stuck on a treadmill where you're never improving.

Despite all that, the civ 6 AI is truly mediocre and is broadly incapable of using many of the interesting systems in civ 6. I really enjoy plenty of the design of civ 6, but it doesn't seem to be designed with writing an AI in mind, and I think it'd be worthwhile redesigning much of civ 6 for ease-of-AI-coding in civ 7. I hope they did that!

 

[aieeegrunt]

The machine learning being used in what is currently being called AI is functionally as intelligent as autocorrect on one's phone is; it is capable of learning from inputs, but calling it AI is a huge marketing stunt to try and make people think it is somehow fundamentally different from algorithms we are more used to interacting with in our day-to-day life. That being said, I do agree that applying machine learning algorithms to a video game's AI is not the boon people expect it to be - people hugely underestimate just how many decisions are made across a game of civilization, and how difficult that would cause it be to make a ML-based AI for a game like this. It might be more appropriate for very specific circumstances, but it also runs into the issue of changing the game - you really don't want to have to retrain your whole AI model when you make significant balance changes in development/in patches. On top of that, I do agree that adaptive AI might not be what people want - if your AI is learning how to counter you specifically, it could very much feel like you're stuck on a treadmill where you're never improving.

Despite all that, the civ 6 AI is truly mediocre and is broadly incapable of using many of the interesting systems in civ 6. I really enjoy plenty of the design of civ 6, but it doesn't seem to be designed with writing an AI in mind, and I think it'd be worthwhile redesigning much of civ 6 for ease-of-AI-coding in civ 7. I hope they did that!

Damn I had not thought of the retraining issue every time you patch the game or add a new feature

 

[pokiehl]

The machine learning being used in what is currently being called AI is functionally as intelligent as autocorrect on one's phone is; it is capable of learning from inputs, but calling it AI is a huge marketing stunt

Technological scholarship, the industry, and lay usage all disagree with this assessment. Not all AI is machine learning, but all machine learning is AI. Like I said, we’re just getting bogged down in the “no true Scotsman” fallacy here.

The various arguments like “well real AI would be unbeatable” or “true AI doesnt’s exist” or “AI for a game like Civ6 is impossible” are just copes to avoid the simple fact that this is easily the worst AI in the series, by a mile.

For what it’s worth, I wasn’t saying any of that (not sure if you were referring to me).

I agree and I don’t know what nitpicking at the definition of AI is really accomplishing in this thread anyway—we all knew what the OP was referring to when he said “AI.”

And we all know that whatever you call it, the way Civ 6 computer opponents play the game is just disappointing and could be so much better.