Does the AI cheat enough?

[acluewithout]

I don’t mind the AI as it is currently, although I appreciate it has some real limitations. Hopefully it will keep getting better over time.

A lot of people don’t like the cheats the AI gets, particularly on higher levels. But I wonder if the game might be more challenging or more fun if the computer actually cheated a bit more.

In particular:

- Perhaps each Civ should get some of its unique units for free in certain situations. That way, for example, you might fight a horde of longboats and beserkers if you’re next to Horrible Hysterical Harold. This would need to be fine tuned for eg Monty or
Gilgabro.

- Maybe Civs should spawn some defensive units if you attack them, to help them defend; or even spawn some units if they declare war; so the computer is more competitive in war. And or perhaps loyalty boosts when a Civ is on its last city.

- Maybe the AI needs some way to cheat with strategic resources and upgrades so it can get to eg musket men sooner.

- Maybe the AI should be able to buy tiles for free so it can place districts with more adjacency.

- Maybe the AI should get %bonuses to building certain wonders that are culturally appropriate for that Civ (so you know if you conquer Egypt you’ll probably get the Pyramids).

- Maybe we need emergencies which are more targetted at players. In particular: an Ideological Emergency, where if you’ve just go to space or are capturing all the great people, then Civs with a different ideology actively oppose you (via spies and or war; maybe their spies get +levels), or a Colonial Emergency, where if you settle or conquer on a foreign continent then the locals try to eject you (with the target getting loyalty negatives on the foreign continent).

 

[acluewithout]

On a related note, it looks like some bright sparks have found a coding mistake that may have been hampering the AI. See here. Sarah has posted to say Firaxis is investigating.

 

[Jackanape]

Interesting...you have to wonder how much this affects game play...

Im beginning to get tired of certain aspects of the AI...actually, most aspects. I've only ever had a peaceful game since R&F with the Cree...

 

[MarigoldRan]

AI needs to cheat more in late game and less in early game.

 

[Civ_Gumby]

On a related note, it looks like some bright sparks have found a coding mistake that may have been hampering the AI. See here. Sarah has posted to say Firaxis is investigating.

Kind of glad to see this, it is pretty obnoxious to see that most of the cities you conquer have a holy site aka useless. Someone on this forum actually identified this 3 weeks ago, just never caught any traction and it became a big deal on some other forum. I think the biggest thing that should probably be tweaked is AI intelligence, countless times I am attacking a city and they spawn a builder. There should be a switch that overrides other builds when being attacked like that. I'm not a big fan of situational buffs, random unit spawning. It is too unpredictable and would be extremely frustrating to deal with. Perhaps they should look into buffing stats again? Again I'm not really sure about this, the AI on deity has a +4 combat bonus and gets oligarchy very fast making it very difficult when your warriors fighting 20 strength vs 28 and getting slammed.

Either way I think its rather unlikely that they will make major changes. Hard to say how much of an impact the yield fix will have on the game yet.

 

[acluewithout]

... I'm not a big fan of situational buffs, random unit spawning. It is too unpredictable and would be extremely frustrating to deal with...

I think it would depend on how it’s implemented.

Playing Emperor, the AI seems to very rarely build its unique unit, and often when it does it seems to build only one. I had few games in a row where Ghandi declared war and unleashed his horde of, ah, one lonely Varu. Anti-climax.

I’d really like to see the AI get their special units up and running more. I mean, if I invade, say, Japan, I should be faced with Samurai. If gupta attacks me, there should be a horse of elephants coming over the hills.

Indeed, this is why Monty is such a distinctive opponent on higher difficulties - he starts with his UU!

 

[Chefofrats]

The AI cheating is superbly annoying and one of the main reasons I play on King and can't stand the higher difficulties in the game. Already on King it's unacceptable when the AI spams units out of its arse. In a recent game in its earliest stages I was attacking a Sumerian city with 4 archers and 3 warriors. The city kept vomiting a war cart EVERY SINGLE TURN for several turns. There is no way they had either the production or gold to do that. How do I prepare for an unpredictable cheat like that?

I understand it is probably hard to do, but Firaxis should look into making the AI properly use the existing game mechanics, like using adjecency bonuses and actually improving its terrain. In another recent game on King I conquered the Kongo capital around turn 250. Only several tiles around the city were improved and the districts had +1 adjecency, if any. That needs fixing, but cheating "mechanics" are not a fix, just a workaround.

 

[Rambo919]

A simple way is the AI getting free units with each age change... each successive age having more units (single-then corps-then armies?) with more variety. how exactly to balance that is the tricky part....

 

[Civ_Gumby]

I think it would depend on how it’s implemented.

Playing Emperor, the AI seems to very rarely build its unique unit, and often when it does it seems to build only one. I had few games in a row where Ghandi declared war and unleashed his horde of, ah, one lonely Varu. Anti-climax.

I’d really like to see the AI get their special units up and running more. I mean, if I invade, say, Japan, I should be faced with Samurai. If gupta attacks me, there should be a horse of elephants coming over the hills.

Indeed, this is why Monty is such a distinctive opponent on higher difficulties - he starts with his UU!

I think that they should add biases towards training unique units, some civs do. Some unique units are just bad, like the samurai. Even with the AI bonuses to production units are still costly to build. I think the big thing with varu is 120 production is 40 more than horseman, the AI doesn't chop for production. The AI is competent in wars during the first 50 or so turns of the game, when they have a chance to surprise you but after that they just have no clue what they are doing. I think the problem lies here more than anything else. I would like to see Varu movement points changed to 3 but thats a discussion for another day.

The AI cheating is superbly annoying and one of the main reasons I play on King and can't stand the higher difficulties in the game. Already on King it's unacceptable when the AI spams units out of its arse. In a recent game in its earliest stages I was attacking a Sumerian city with 4 archers and 3 warriors. The city kept vomiting a war cart EVERY SINGLE TURN for several turns. There is no way they had either the production or gold to do that. How do I prepare for an unpredictable cheat like that?

I understand it is probably hard to do, but Firaxis should look into making the AI properly use the existing game mechanics, like using adjecency bonuses and actually improving its terrain. In another recent game on King I conquered the Kongo capital around turn 250. Only several tiles around the city were improved and the districts had +1 adjecency, if any. That needs fixing, but cheating "mechanics" are not a fix, just a workaround.

The AI only gets +20% production on king difficulty, I really don't think that the AI "cheats" in the sense that they go above and beyond their production bonus. This would be very hard to code. I don't ever see cities spawning out that many units even on higher difficulties.

 

[Chefofrats]

The AI only gets +20% production on king difficulty, I really don't think that the AI "cheats" in the sense that they go above and beyond their production bonus. This would be very hard to code. I don't ever see cities spawning out that many units even on higher difficulties.

I have no idea how it was done, but the city spawned 3 or 4 war carts one after the other, turn after turn, no respite. Could Sumeria have had the gold to purchase those? Does not seem likely given the game's stage, even more so as they made two other war carts in quick succession in a city I conquered a few turns earlier.

 

[Peserwin]

I don't think AI needs to cheat more, it's just about choices it makes. I am using AI+ mod again and it's a difference between light and day on how well the AI is performing: I'm fighting huge armies which are almost always up to date, civs get wiped out, AI are spamming cities... it's a feast! And I believe most tweaks are done in that bias sheet.

Oh and I think your ideas, OP, about Ideological and Colonial Emergencies need to be implemented asap. Great idea!

 

[Civ_Gumby]

I have no idea how it was done, but the city spawned 3 or 4 war carts one after the other, turn after turn, no respite. Could Sumeria have had the gold to purchase those? Does not seem likely given the game's stage, even more or as they made two other war carts in quick succession in a city I conquered a few turns earlier.

It seems likely that they probably purchased one or two during the process. I'm just trying to rationalize the process.. Base production is 55 which goes down to ~40 with the 20% difficulty modifier maybe they had the 50% civic making the cost of a warcart 20-25 production. The wiki page says they cost 55 gold but I'm pretty sure thats wrong and the cost would be somewhere in the 200-300 range. Usually AI's have that kind of gold sitting around even pretty early on.

 

[acluewithout]

I don't think AI needs to cheat more, it's just about choices it makes. ...

Well, I really don't mind how the AI gets there! I just thought cheating might be easier!

Without turning this into a "things I'd like the AI to do better" thread, I guess the main things I'd like from the AI is (1) more defensive units when I attack the AI's cities (particularly ranged and anti-cav), (2) more, and more varied, units with the AI attacks me (including ideally a good whack of the AI's unique units, really just for the sake of fun), (3) a little more aggression generally, (4) better district placement, and (5) no building Petra, or Jungle Petra or Tundra Petra except where there are at least 4 or 5 workable tiles.

I think 1, 2 and 4 could be helped with some well done 'cheating', but happy for it to be achieved through better AI. 3 and 5 almost certainly require better AI no matter what.

 

[halfhalfharp]

Actually, can AIs just outwit me, instead of out-cheating me?

 

[Softly]

Use the AI+ mod.

 

[acluewithout]

That’s fair. I agree the AI should be better.

But that said, ultimately I don’t care how it’s done, I just want the computer opponents to challenge me and make the game interesting.

I’ve said this before: I want to be invaded by hordes of viking men with their longships and beserkers. (ah, “phrasing”?) When I capture a city, I don’t want it to have districts placed in absurd places, or with Chicken Pizza literally built on the only jungle tile. If the AI can do that, great. If it does it by cheating, that’s fine too - just don’t tell me. (Actually, you can tell me, I don’t care.)

Likewise, I don’t care if the AI stations archers or spearmen in it cities because the AI is thinking tactically, or because it just gets free archers and spearmen when it founds it’s first five cities; or if the computers buys a knight or ram because of war, or just got them for free when the war was declared.

Basically, if cheating meant the AI has an interesting mix of units, was never ‘short’ units, and had vaguely okay cities, I could live with a lot of the other AI silliness.

Actually, on the whole viking horde thing. I think it’s really important for the AI to build its things so each civ looks and feels different. One of the things I like about Monty in my games is that he has his eagle warriors. Likewise, I like that Gilgabro spams his Zigga-whatsits. His cities look different as a result. I mean, they look nuts, but they look different.

On the same note, the AI really need to focus on “their” wonders. If I pick a fight with Rome, I want to end up with the colloseum. Actually, scratch that. I want to deliberately pick a fight with Rome because I know that will get me the colloseum.

 

[Zuizgond]

Yeah let's give AI more bonuses instead of improving it....

/s

 

[Trav'ling Canuck]

AI needs to cheat more in late game and less in early game.

I wouldn't call it "cheat" so much as be provided with bonuses, but yes, I agree. I would prefer if, especially at higher levels of difficulty, the AI received escalating bonuses at each age. That could be easily tied to the new era system.

Right now, the difference in difficulty level between Deity and Prince is almost exclusively in the opening few turns. Once you catch up to the AI, the game plays essentially the same. If the AI at the higher levels got new bonuses with each new age (say, a % increase to culture, tech and gold yields), then you'd have to run faster on the higher levels to stay in front of it, which I think would be a good thing. As long as those bonuses are "behind the scenes", it should avoid the immersion breaking feeling some have complained about here.

Actually, can AIs just outwit me, instead of out-cheating me?

I'm pretty confident the answer is "no".

Yeah let's give AI more bonuses instead of improving it....
/s

To be fair, the AI has been improved. The basic design of the AI in Civ 6, however, as I understand it, has a fatal flaw for those who want it to be a tough opponent.

The AI is not designed to have an overarching goal or immediate strategic goals. It's designed to have an understandable approach to diplomacy so you can "role play" against it. And it's designed to have a decision tree on each specific decision that can be "tweaked" to make it more or less understandable from the perspective of the human player and/or more or less efficient during play (the AI+ and similar mods need to rely primarily on this, as I understand it). But it won't go into a war with a particular strategic objective in mind (defend, capture city A, etc.) and it will only achieve a victory condition by "backing into" it through the effect of its myriad decision trees.

I may be overstating the case somewhat here, as AI's will pick a victory condition to pursue, and I'm a layperson in this field so I'm relaying what I understand the experts to have identified and may have some of this wrong. But I'm confident that the AI was designed to be a foil to the player rather than to actively pursue victory, and the structure of the underlying AI decision making code reflects that design decision.