Does AI cheat?

[AutomatedTeller]

Humans do tend to remember failures far more often than successes, even if the successes are more improbable. I know I've occasionally won a fight at 25% odds, but I care a lot more about the fights I lose at 75% odds*.

That said, the AI does have a number of handicap benefits simply based on the nominal difficulty level they play at. Full information here (no idea how up-to-date that is). Among other things -- and keep in mind, this is assuming you're playing a difficulty above Noble:

* They pay less for research, units, city maintenance, etc.
* They get better starting sites compared to the humans' sites.
* They have a built-in diplomacy penalty towards humans.
* They'll refuse to trade techs with humans as often (without changing how readily they'll trade techs with other AIs)
* Their cities grow faster.
* Their unit upgrade costs are lower.

And so on. There are also some AI-only benefits -- for example, their unit sight range is boosted by their movement range, to help make up for the fact that they don't remember where they've last seen units. I don't know what all of those modifiers are.

* Strictly speaking I probably shouldn't have started either of those fights...but enh, the unit was disposable.

I had no idea either of those things were true.

How does the game do the first one? And the second is pretty funny - i guess my attempts to hide my units gathering to attack are all in vain

 

[Spectrite]

I'm conviced that the AI gets a combat bonus that isn't reflected in the combat odds.

I got tired of it last night and saved, fought, lost, reload, fought, lost, reload, etc... On 67% combat odds, I lost 13 times before I finally won. That's 7% combat odds, mine and the AI's were both knights. It was ridiculous.

If you reload and fight with the exact same units, the combat odds won't change. They will, however, change if you do something different: attack with another unit first, give it a different promotion, wait one turn before attacking, etc.

 

[ParadigmShifter]

You only get a different result if you invoke the random number generator, so I don't think promoting will do that.

 

[Derakon]

I had no idea either of those things were true.

How does the game do the first one? And the second is pretty funny - i guess my attempts to hide my units gathering to attack are all in vain

I would assume for site selection that the game does something along the lines of generating a set of civ starting locations -- either by generating a map and then looking for sites that match the minimum requirements for a starting capital, or by generating a map sans resources and then placing resources such that certain areas are valid capital locations. Then it ranks the sites based on their available resources, presumably with all improvements built / all strategic resources discovered and some kind of equivalency to compare food, hammers, and commerce. So for example, a site with a wet corn, two fish, horses, and iron would rank higher than a site with a floodplains, a plainscow, and a hill sheep. Once they're ranked, each civ is assigned a site, with the player's position in the rankings depending on their difficulty level. Presumably civs at the same difficulty are randomly assigned sites from the available selection.

This is all speculation, of course, but speaking as a software guy it's how I'd do it. And again, remember that the normal way to view this is that as you go up in level, your starting site gets worse -- not that the AI sites get better.

 

[Spectrite]

You only get a different result if you invoke the random number generator, so I don't think promoting will do that.

Well, in my experience, it has led to a different result. Makes sense, too, since it changes the combat odds.

 

[Jerrymander]

Your result might have been in the range between unpromoted and promoted.

Suppose you have a 63.4% chance to win umpromoted, and a 73.4% chance to win promoted. With a basic assumption of a RNG, there (roughly?) exists a 10% value where you lose the first and win the second. So promoting can affect even non-random reloads.

 

[Deckhand]

* They get better starting sites compared to the humans' sites.

I don't think this is true.
I have never seen this claim outside of this thread. Until it is confirmed by a "code diver" I won't believe it.

Spoiler :

Or at least annecdotal consensus.

 

[raxid]

I don't think this is true.
I have never seen this claim outside of this thread. Until it is confirmed by a "code diver" I won't believe it.

Spoiler :

Or at least annecdotal consensus.

I'm pretty sure i've actually seen a code diver disprove this once before (this was long long ago). I could be wrong though.

 

[Nick Carpathia]

I said it earlier and I'll say it again. There's enough variance in start sites that some will be better than others. You, the sentient human capable of forming value judgements, are going to fall somewhere within that range. The AI, on the other hand, will not complain about a plains cow start, you'll just roll over them without a thought.

 

[revengeofmakno]

So you lost 13 times in a row with a 33% chance to lose? Aside from not having Random Seed on Reload enabled (which would mean that the outcome of the battle does not change by reloading) you had a 0.000018% chance of losing 13 times in a row.
God hate you.

This assumes that each 33% throw of the dice was an honest throw. But I am almost convinced it is not. I once had a game save and my very next combat was at over 99%. I lost it.
Ok, that can happen. As an experiment, I went back to the save and again did that very same combat with no intervening combats. It lost again. I did it twice more and lost again each time. That is 4 consecutive losses at over 99%. I didn't try a 5th time. That might stand up as the greatest long shot in history (depending on what the exact odds really were).

I don't believe there is random number generation going on for the combat resolution. Perfect random number generation is likely a theoretical impossibility but it is not difficult to get something close. However, there is something fishy about the way CIV is doing it.

One possibility is that a set of numbers is periodically generated for the combats and then that set of numbers will be used for the next series of conflicts. If you run the exact same combats over and over, you will get the same results. So, a 99+% that loses would lose again if run in the same order, time and circumstances after a save. One day, if I really have nothing else to do, I might test the hypothesis out.

 

[mbtshoes]

I thoroughly convinced that the AI cheats, but I can't really prove it. I once heard somewhere that the AI has a certain number of 'hidden traits" or whatnot. I'm not sure what that means, but it's the reason I can't actually prove the cheating.

A good example: I'm researching Liberalism, and according to my espionage, I should get it. But just 1 or 2 turns from it, a random AI will take it. In a massive rage, I check the diplomacy, and the AI that got it doesn't have all the perquisites! The only explanation for this is either flat out cheating, or hidden tech advancements.

Moderator Action: Spam ad link deleted.
Please read the forum rules: http://forums.civfanatics.com/showthread.php?t=422889

 

[Deckhand]

I don't believe there is random number generation going on for the combat resolution. Perfect random number generation is likely a theoretical impossibility but it is not difficult to get something close. However, there is something fishy about the way CIV is doing it.

There is definitely an RNG invoked for each coin flip of a combat. Last year (?) there was a thread discussing the particular RNG that Civ uses and (IIRC) an analyst noted that it is "streaky".

Spoiler :

and as already noted, unless you have New Random Seed selected, you will get the same result with each reload (your post doesn't mention your setting)

 

[Nick Carpathia]

Ahahaha, I've never seen this kind of spambot before. Congrats Nikolai Dneppen, your post was so out there that even a computer script called you on it.

 

[the343danny]

This assumes that each 33% throw of the dice was an honest throw. But I am almost convinced it is not. I once had a game save and my very next combat was at over 99%. I lost it.
Ok, that can happen. As an experiment, I went back to the save and again did that very same combat with no intervening combats. It lost again. I did it twice more and lost again each time. That is 4 consecutive losses at over 99%. I didn't try a 5th time. That might stand up as the greatest long shot in history (depending on what the exact odds really were).

I don't believe there is random number generation going on for the combat resolution. Perfect random number generation is likely a theoretical impossibility but it is not difficult to get something close. However, there is something fishy about the way CIV is doing it.

One possibility is that a set of numbers is periodically generated for the combats and then that set of numbers will be used for the next series of conflicts. If you run the exact same combats over and over, you will get the same results. So, a 99+% that loses would lose again if run in the same order, time and circumstances after a save. One day, if I really have nothing else to do, I might test the hypothesis out.

Yes, that is exactly the case. In the custom game menu there is an option called "generate new seed on reload" or something. That makes it so you do not get the same result upon reloading. BTW read the thread. Maybe you replied to one post and skipped a lot of posts, but what I just said has been repeated several times already before you posted this.

 

[Jerrymander]

This assumes that each 33% throw of the dice was an honest throw. But I am almost convinced it is not. I once had a game save and my very next combat was at over 99%. I lost it.
Ok, that can happen. As an experiment, I went back to the save and again did that very same combat with no intervening combats. It lost again. I did it twice more and lost again each time. That is 4 consecutive losses at over 99%. I didn't try a 5th time. That might stand up as the greatest long shot in history (depending on what the exact odds really were).

I don't believe there is random number generation going on for the combat resolution. Perfect random number generation is likely a theoretical impossibility but it is not difficult to get something close. However, there is something fishy about the way CIV is doing it.

One possibility is that a set of numbers is periodically generated for the combats and then that set of numbers will be used for the next series of conflicts. If you run the exact same combats over and over, you will get the same results. So, a 99+% that loses would lose again if run in the same order, time and circumstances after a save. One day, if I really have nothing else to do, I might test the hypothesis out.

You just didn't turn "Random Seed On Reload" on. You are seeing the same result every time, not a random result.

 

[Spectrite]

One possibility is that a set of numbers is periodically generated for the combats and then that set of numbers will be used for the next series of conflicts. If you run the exact same combats over and over, you will get the same results. So, a 99+% that loses would lose again if run in the same order, time and circumstances after a save.

You are exactly correct. I can confirm, this is the case.

 

[rogue131]

philosphically speaking, I don't think a program is capable of cheating. . . However,

 

[rogue131]

BTW @Nick Carpathia, are you running A Global Community in Civ 4 ;-)

 

[Notenoughtrebs]

I don't think this is true.
I have never seen this claim outside of this thread. Until it is confirmed by a "code diver" I won't believe it.

Spoiler :

Or at least annecdotal consensus.

I'm with you Deckhand.

I don't believe humans get bad starting positions. I play on Emperor and you get good, bad and average positions. If you take an AI capital, they don't seem to me at least to be different.

 

[revengeofmakno]

You just didn't turn "Random Seed On Reload" on. You are seeing the same result every time, not a random result.

Also @spectrite

Thank you. This makes sense. I don't really care too much about losing the high odds combat but the fact that it repeated begged for an explanation.


@Deckhand

That bit about "streaky" was part of what I was saying. Random number generators are such in name only although some are much better at maintaining the illusion than others. I remember a bunch of us in a programming class long ago struggling with trying to create good ones.