Does the AI actually play by the same rule set as the human player?

[Planktonic]

Or is there some leeway they are allowed to take?

This has happened to me twice: I am trying to forestall a runaway civ from a science victory (deity/standard). I manage to get a carrier/bombers to the city where the spaceport is building the rocket (visible to me from the spy in the city, who is also trying to disrupt rocketry). No other city has a rocket on dock in the spaceport- I think the project can only be run in one site in any case, right?

I air pillage the spaceport/rocket; the blackened icon shows a successful pillage.

But in the first game, 2 turns later the moon landing was launched in any case- presumably from a different spaceport?

In the second game, 2 turns after the air pillaging, I could see a different spaceport loaded with a nearly fully assembled rocket where none had been present before, as if the AI had just transferred the partially completed project from one city to another. In this case I was able to air pillage the second spaceport, and for good measure nuked the parent city.

But it leads me to wonder if the AIs are able to take measures behind the scenes beyond the rule set available to us as human players, and beyond the bonuses they are given.

And I also wonder about the bonuses- multiple times I've encountered AIs with science or culture yields that seem far beyond what even a 32% bonus would allow once the full measure of the civ is taken in exploration.

The first game was particularly galling- I was going for a culture victory and wanted to prevent the culture burst from the moon shot. I thought I had done so, only to have the rocket launch. (I still won, but many turns later and only after nuking the capital city, which doesn't feel like much of a cultural victory at that point....)

Has anyone encountered this or other circumstances where the AI seems to be behaving outside the "rules"?

 

[Gedemon]

Except the bonus, it uses mostly the same ruleset, if it doesn't in a case like that it would be a bug IMO.

 

[Abaxial]

Although the AI can demand you move your units, and you the player can't. What is particularly irritating is when the AI suddenly forward settles you and then demands you move.

 

[Linklite]

Allegedly, yes, the AI has pretty much the same rule set (the other responses cover the types of exceptions you get, which don't really affect abilities as such). They also get bonuses from difficulty.

I have a strong suspicion that in reality that it doesn't, though. Two experiences that tell me otherwise:

1. My wife and I were playing as a team. Between us, we controlled all sources of uranium (confirmed both by a visual inspection of the map and the Global Resources tool) bar two which were uncontrolled. One civ had 4 GDRs walking around and another 2 being built. How, when they don't have any uranium to build them in the first place? One could come from a GA dedication, but the rest?
2. I was playing a game where I was fairly consistently 2.5x yield for Science and 4x for Culture of the second best civ for each category respectively throughout the game. Towards the end, the AI was still beating me to Techs and Civics. How? I can understand at the beginning due to beelining and one offs like huts and pillaging, but towards the end, those yields should be enough to give me an absolute lead.

Neither of those are absolute proof, of course. Perhaps I'm just not understanding or missing something, and there are legitimate things that mitigate my lead (increased costs for getting.ahead of the Era, for example). However, they can't account for that reversal in totality and it does make me suspect that the AI gets help behind the scenes.

 

[Atlas627]

I know in Civ5 they were allowed to ignore some of the movement restrictions when swapping units because they couldn't plan as well as a human, so they might still do that in 6. Also I know in 5 they could buy a unit when the city was already occupied by a unit, so they might be able to do that in 6 also.

 

[CromulentRomulan]

To add to the anecdata of not knowing whether the AI plays by the same rules I would add the faith/gold acquisition of great people. Sometimes I'm neck and neck with an AI for a great person and they will be able to acquire them despite having a lot less in their faith/gold balance than me

 

[InsidiousMage]

Here is a video by Soren Johnson talking about AI design and Civ4 that talks about the fact that the AI does need some boosts that the player doesn't have access to and why it's necessary.

 

[Planktonic]

Here is a video by Soren Johnson talking about AI design and Civ4 that talks about the fact that the AI does need some boosts that the player doesn't have access to and why it's necessary.

Thank you for posting this very interesting video. It explains a lot.

I was curious to see that he admits that they coded it so that the AI will never deploy nuclear weapons against the human player (for Civ 4, but I imagine it's the same for Civ 6). Having had a recent playthrough where I was forced to use nuclear weapons (for only the second time in my experience, which is admittedly not that large), it seems to me that this is a real conundrum. On the one hand, at higher difficulties the AIs will always reach this tech first, and could therefore always obliterate the upstart human, which would ruin the game. On the other hand, I see now that there is no real downside to using nukes if there is no risk that the AI will retaliate in kind (the so-called "nuclear emergency" is toothless), which kind of ruins the game in the other direction....

 

[Vizurok]

A semi-related idea came to my mind as I just finished a Hungary SV game on Deity. At around T160, I became sovereign of all city states that somewhat concerned me - obviously, the levy game gave me a huge amount of extra envoys but the sheer surplus of envoys made me wonder whether the AI participates in the CS quest mini-game. I couldn't find any concrete info on this, so my question stays open:

Are CS quests limited to the human players? It is absolutely sure that it's not a first-takes-it situation as objectives do not disappear unless the human player clear them. But is it possible for the AI to grab the envoy for completing a task (that obviously goes unnoticed by the player)? If the AI is ineligible to play this game, then there we have a huge assymetrical design point that gives significant advantage to the human player.

 

[Lily_Lancer]

One thing I noticed is that AI can move his purchased unit instantly, or use it for an attack.

I haven't found other differences. （Maybe they always get fair trade despite their relationship?)

 

[InsidiousMage]

AI will never deploy nuclear weapons against the human player (for Civ 4, but I imagine it's the same for Civ 6)

Unless it was a bug I've been nuked by Germany before, multiple times. Given that he kept nuking the same city though I'm wouldn't be surprised if it was a bug.

 

[UWHabs]

A semi-related idea came to my mind as I just finished a Hungary SV game on Deity. At around T160, I became sovereign of all city states that somewhat concerned me - obviously, the levy game gave me a huge amount of extra envoys but the sheer surplus of envoys made me wonder whether the AI participates in the CS quest mini-game. I couldn't find any concrete info on this, so my question stays open:

Are CS quests limited to the human players? It is absolutely sure that it's not a first-takes-it situation as objectives do not disappear unless the human player clear them. But is it possible for the AI to grab the envoy for completing a task (that obviously goes unnoticed by the player)? If the AI is ineligible to play this game, then there we have a huge assymetrical design point that gives significant advantage to the human player.

I have to imagine the AI has quests available to them, since I do see them have more envoys in CS than you would be able to get neutrally from the civics tree. That is, unless if they cheat and do something like don't give them quests, but when they trigger a civic free envoy, they secretly get 2 or 3 instead.

I imagine though that the AI basically ignores those quests completely, so would be really easy for them to get stuck on a specific one for a CS and simply never complete it. But that would also obviously explain why it seems like I can ally with about 2/3 of the CS on the map in many games.

 

[Dotsworthy]

One question I do have. Are the AI aware of tribal villages in areas of the map the player has not discovered? I feel I've seen several games where the AI has tribal villages near their cities but only collects them once a player discovers them on the map.

 

[OmegaDestroyer]

The reason why we have weak AI is due to weak programming. And instead of trying to fix it, they just give the computer cheats to be "competitive" or put in systems designed to only hurt players (always banning luxuries the player has, etc.). I have sunk about 1,800 hours into the game and it gets harder and harder to come back to it.

 

[c4c6]

The reason why we have weak AI is due to weak programming. And instead of trying to fix it

I agree, that would be the best solution. And the most expensive. So some general annotations:

1. If they would release the source files of the core DLL, modders could improve the AI like this was done for civ4 & civ5. Strangely enough not a lot of players are asking for "release the source files of the core DLL".

2. I think the keyword cheats doesn't help us at all. It has only a meaning for an objectively GoodAI (cf. to Soren Johnson's GoodAI vs. FunAI video), and to expect we could receive anyhow a strong GoodAI is less than an illusion.
More important: I want to win most of the time! The AI players shall be challenging, sure, moderately to very challenging; but to be honest, only VERY, VERY few players want to loose more often than very seldom.
[Developing a strong GoodAI and then dumbing it down to customers wishes isn't cost effective at all]

For a FunAI the word cheats has no meaning: the AI players shall play by rules so that it is fun, challenging and fun.
We humans like exhausting complex & complicated rules (in increasing numbers). Don't expect the AI players to play by those overly difficult rules, because it is a nightmare to implement. The AI should play the same rules in general, but with (a lot) less details. Of course this must be done during gameplay in a inconspicuous way, ie. the human player has to "cheat" himself to reveal it.


Boardgames epilogue: Differently strong players can agree to play with handicaps on one side - nobody calls that cheating. Just adjust to differently strong players.
Ie. in Chess the stronger player can forego 1 or 2 pawns or a knight. In Go this is more smooth and done regularly by the weaker player starting with a couple of extra pieces ...

 

[Linklite]

I agree, that would be the best solution. And the most expensive. So some general annotations:

1. If they would release the source files of the core DLL, modders could improve the AI like this was done for civ4 & civ5. Strangely enough not a lot of players are asking for "release the source files of the core DLL".

2. I think the keyword cheats doesn't help us at all. It has only a meaning for an objectively GoodAI (cf. to Soren Johnson's GoodAI vs. FunAI video), and to expect we could receive anyhow a strong GoodAI is less than an illusion.
More important: I want to win most of the time! The AI players shall be challenging, sure, moderately to very challenging; but to be honest, only VERY, VERY few players want to loose more often than very seldom.
[Developing a strong GoodAI and then dumbing it down to customers wishes isn't cost effective at all]

For a FunAI the word cheats has no meaning: the AI players shall play by rules so that it is fun, challenging and fun.
We humans like exhausting complex & complicated rules (in increasing numbers). Don't expect the AI players to play by those overly difficult rules, because it is a nightmare to implement. The AI should play the same rules in general, but with (a lot) less details. Of course this must be done during gameplay in a inconspicuous way, ie. the human player has to "cheat" himself to reveal it.


Boardgames epilogue: Differently strong players can agree to play with handicaps on one side - nobody calls that cheating. Just adjust to differently strong players.
Ie. in Chess the stronger player can forego 1 or 2 pawns or a knight. In Go this is more smooth and done regularly by the weaker player starting with a couple of extra pieces ...

A good breakdown. I'd add though that even fun AI needs to have maintain certain unbreakable mechanics. I was really disappointed when I realised that I couldn't use besiegement in AoEII - the AI only appeared to collect resources, but in reality had it's own internal infinite supply. So if you slaughtered all their villagers, they could still maintain a constant supply of soldiers and villagers.

When you cut corners like that, it can really ruin the player's experience if they build their strategy around an ostensibly indispensable mechanic that the computer doesn't use or even need.

 

[c4c6]

yes, it has to be done in a sincere way, not quick & dirty alias cheap on purpose ... you can use endlessly words, still it is easy: FunAI means _FUN_ AI - you cannot exaggerate it like in your example! (not all players are careless or dumb)

I was really disappointed when I realised that I couldn't use besiegement in AoEII - the AI only appeared to collect resources, but in reality had it's own internal infinite supply. So if you slaughtered all their villagers, they could still maintain a constant supply of soldiers and villagers.

when modding civ2 @ terraforming (mountains > hills > plains > swamps > coast > ocean & vice versa) I was inconsolable to find out, that the AI players did not use the special resources at all - just the standard yields from terrain (in that case ALWAYS changed low yield jungle into irrigatable grassland even when destroying precious bananas on that jungle tile)

 

[Planktonic]

Another instance of weird AI behavior. Following up on my earlier game, I had air pillaged 4 German spaceports to prevent them from launching the exoplanet mission. I kept all spaceports in view (with spies) and they remained pillaged. ~ 10 - 15 turns later I was informed that Germany had launched the exoplanet mission. Apparently in that period the AI built 2 (!) spaceports in other cities and launched the rocket.

So I determined that I had at most 50 turns to eliminate the German civ if I was going to salvage the game. I resorted to nuclear attacks, and was able to start conquering the cities in good order (I had GDRs to enter and occupy the blasted cities). After I had captured about half of them, including the ones with the new spaceports I thought to check up on the progress of the spacecraft. I was surprised to find that the launched spacecraft was travelling at *zero* light years per turn- screenshot is attached.

I didn't think it was possible for the exoplanet to travel at zero, meaning no progress at all. Is it the case that if you eliminate the originating spaceport city that the exoplanet rocket progress is halted? Would that hold true for the human player as well? Or is this just some weird bug?

 

[PiR]

Thank you for posting this very interesting video. It explains a lot.

I was curious to see that he admits that they coded it so that the AI will never deploy nuclear weapons against the human player (for Civ 4, but I imagine it's the same for Civ 6). Having had a recent playthrough where I was forced to use nuclear weapons (for only the second time in my experience, which is admittedly not that large), it seems to me that this is a real conundrum. On the one hand, at higher difficulties the AIs will always reach this tech first, and could therefore always obliterate the upstart human, which would ruin the game. On the other hand, I see now that there is no real downside to using nukes if there is no risk that the AI will retaliate in kind (the so-called "nuclear emergency" is toothless), which kind of ruins the game in the other direction....

Unless it was a bug I've been nuked by Germany before, multiple times. Given that he kept nuking the same city though I'm wouldn't be surprised if it was a bug.

I've been nuked on different cities in a 1v1 game. Also one or two of my CS were nuked. But the AI never sent anyone to take the nuked cities (even though it had a few GDR).

 

[Abaxial]

Boardgames epilogue: Differently strong players can agree to play with handicaps on one side - nobody calls that cheating. Just adjust to differently strong players.
Ie. in Chess the stronger player can forego 1 or 2 pawns or a knight. In Go this is more smooth and done regularly by the weaker player starting with a couple of extra pieces ...

In chess, this used to be commonplace in the 19th C. Nowadays, in my experience, people really don't like games at odds.