AI cheats at combat odds

[aicheats]

Everyone who pays any attention to the combat odds has noticed that the AI doesnt roll fairly. This has been debated on the forums many times, IMO if you study it yourself at all you know it DOES IN FACT CHEAT. Of course you have to have a basic understanding of probability to know.

I just had a combat, 3 catapults attacked a city. Odds of the 3 combats was 91%, 90%, and 90%. I lost all 3 combats..

The probably of losing all 3 combats are:

.09 * .1 * .1 = .0009 = .09%

1 in 100 times is 1%
1 in 1000 times is .1%
1 in 10,000 times is .01%
9 in 10,000 times is .09%

The odds of that happening is INCREDIBILITY LOW IF THE AI ISNT CHEATING. It is mathematically possible to happen... but... I SEE TOTAL CRAZY ONE IN A YEAR CHANCES GO AGAINST ME SEVERAL TIMES EVERY SINGLE GAME.

Firaxis... I know this game is old.. but Civ5 sucks IMO. Please make a patch to Civ4 to take out the AI roll hax.

Thank you.

EDIT: People reply on basically saying 'the ai doesnt cheat because I read a thread where some guy did a test and he said it doesnt cheat so it doesnt cheat'. Or.. 'do a test, see if it cheats'.

These replies are mindbogglinly frustrating. I JUST TOLD YOU what I saw... so, my test showed either I am the most unluckiest person ever or it does cheat. You can choose to believer I'm lying I guess. What I really think is happening is that people dont truely appreciate the mathematics of how rare a 9 in 10,000 event is.

If you bought a lottery ticket every single day, and it had a 9 in 10,000 chance of winning, it would take you 3 years on average to win. If you won the lottery every other day...that would be an impossibly unrealistic level of luck. This is what is going on when I play Civ4. About every game I see this sort of total 1 in 1000 or worse odds.

You can choose to believe it or not... you choosing not to believe it doesnt stop it from being reality.

 

[pob]

There was a bug with some combat odds for units that retreat in that sometimes it didn't take into account the final roll to withdraw. Are you fully patched up? (Other than that, combat odds have been shown to be accurate.)

 

[GGracchus]

It's actually .09 *0.1 *0.1 odds of losing, or .0009 chance of losing all three. However, assuming BTS 3.19, the odds are fair. WB yourself and a non-barbarian Ai like 100 infantry or whatever, make sure you both are on flat ground, turn on "new random seed on reload", turn on stack attack, and attack his with yours. Around 50 should always survive(assuming both have no promos. Don't try this against Toku with Gandhi, so don't do this with an agg/pro leader unless using horse units.).

 

[aicheats]

Yes I have the latest patch.

And yes the 'conventional wisdom' on these forums is that some dude created 100 units with the editor and attacked 100 times and saw the odds were what was to be expected so everyone just assumes that is true.

I know what I see. When I see the odds happen just like I saw in my OP, I KNOW it is cheating. I assume others that post here either dont pay attention to it or dont understand the higly unlikely probability of that event occuring. Now if that was the first time I've seen totally BS odds in the last 6months... sure its an outlyer. When I see that sort of crap every single game... I know its cheating.

My assuption is the cheating algorithm cheats based on military strength, not just a pure fudge of the roll. Meaning... if the human player is much stronger than the AI opponent I think it will cheat the rolls much more than it would if the 2 nations are equal in power. If that is true, that would make creating battles with the editor uneffective.

When you are on settler level, you get better goodie huts. Its my assumption when you are stronger than the AI you are fighting, you get worse AI roll haxs. years of playing support this. The probability math of my OP... a 7 in 10,000 chance... supports this.

 

[GGracchus]

Think about it as if you won a 9% and two 10% battles. The odds would be equally impossible, but you'd be much happier. People have lost 100% odds(rounded), and won at equally low odds.

 

[pob]

Well do a proper test where you are stronger than the AI and show that it is true.

edit: this comes up fairly regularly on the forum (linky1 linky2 linky3)

 

[aicheats]

GGrachacus, ty for correcting my math error.. sloppy on my part.

The point of the thread is the same; .0009 is an incredibly low chance of happening. As I stated in my first reply; I beleive the AI cheated based on human vs. AI strength; I don't trust just making 100 units in the editor.

However... you both are glossing over WHAT SPECIFICALLY I SAW. I had a hugely unlikely event happen. Do you think I'm lying? If not, why are you dismissing it. And again, if this was the first time I saw something this unlucky I would chalk it up to an outlyer... but really bad luck happens almost every game... suggesting I'm the unluckiest person ever and I have been hit by lighting 25 times or the AI cheats teh roll.

 

[aicheats]

Well do a proper test where you are stronger than the AI and show that it is true.

Fascinating.

I JUST DID. My original post told you exactly what happened... I don't understand why you are dismissing it.

Yes I know this subject comes up regulary. The reason it does is because the AI is cheating. The people who defend the game never seem to use concrete logic... they just some variation of 'no it doesnt, because some guy tested it and it doesnt'.

I've tested it.. for years... and it does. My post (if im not lying) proves it does cheat. We can have an arguement about if I'm lying, argue about whether 1 in 1000 event should happen multiple times in a 2 hour game... but just saying 'it doesnt cheat cause I said so, what you have observed I will just dismiss because it doesnt fit what I am told to believe'... is frustrating.

 

[aicheats]

Think about it as if you won a 9% and two 10% battles. The odds would be equally impossible, but you'd be much happier. People have lost 100% odds(rounded), and won at equally low odds.

I'm thinking about it in a pure mathematically way.. the only way to think about it. The odds of that event as I descibed happening is incredibly low

Do you agree the odds of that happening is incredibly low? If you disagree then perhaps the subject needs to be changed on a discussion of statistics/probability.

If you agree that it is incredibily low... I don't understand why you are arguing with me. Unless you are saying I am lying and don't see this sort of stuff happen way way more than it should happen mathematcially.

 

[pob]

You only started sampling after you lost the first roll so that one must be discounted.

You can use the WB to test this situation, just load up a file from where you feel this has happened and do a statistically relevant test with many repeats.

 

[DesertPfox]

I ran into this problem just the other day too, playing on emperor. I was attacking a city with cavalry and my first three guys to attack were 3%, 3% and 6%. And all three survived! I should never win three battles in a row under 10%, the AI cheats!

 

[Yoshiegg737]

I ran into this problem just the other day too, playing on emperor. I was attacking a city with cavalry and my first three guys to attack were 3%, 3% and 6%. And all three survived! I should never win three battles in a row under 10%, the AI cheats!

I get this too. Especially on Multiplayer

 

[pob]

I can't wait for the day that the one player who has randomly never lost a battle on civ registers on the forums to complain that the game is too easy on the player.

Anyway... the key arguments against "the AI cheat rolls" are:
1. It's perception bias
2. The number generator has been tested
3. The code is available!

 

[aicheats]

You only started sampling after you lost the first roll so that one must be discounted.

You can use WB to test this situation, just load up a file from where you feel this has happened and do a statistically relevant test with many repeats.

I'm trying to be a non-insulting as possible here... What you said is illogical. An event happened. That event had a % chance of occurring. The fact that someone was running an official test or not before hand has zero to do with the odds of that event occurring.

You keep ignoring what I'm tying and sticking to you what you understand. I see this happening. Its factual to me because I see it. You can choose to believer I'm lying. You can also choose to believer I'm too stupid to read the combat text to see what the odds were b4 I attcked, lost, see the odds again, attack, lose, see the odds agin, attack, lose.

I... dont.. need to run a test.. I've already witnessed it happening.

Please answer my questions if you are going to reply to my thread.

Question: If what I said is true... if I lost 3 combats in a row when I had a 91%, 90%, and 90% chance to win... do you agree that the odds of that happening are extremely low? Lets start there. If you agree with me then you are saying my observations are wrong. If you disagree you have a misunderstanding of statistics. We need to know where you stand to continue.

 

[aicheats]

The odds of a 9 in 10,000 event should be super rare... if the power went out at your home 9 in 10,000 days, it would take on average 3 years for the power to go out of your home.

Its utterly fascinating to me no one here talks about the pure facts... the statistics of events occuring. I guess its my fault for being frustrated about not finding people who have a grasp of statistsics on a game forum.

 

[pob]

I'm saying you only started paying attention when you lost the first battle, so it can't be included in your sample.
The odds of you considering something was wrong becomes only 1% (0.1*0.1) which isn't that significant. i.e. after any battle you lost at ~90% odds, you get angry and pay attention to the next couple of rollls. In 1% of these circumstances you hit another 2 losses.

(I'm not outright refusing to listen to you. Post a save where you are far ahead of the AI and I will test it fairly. I don't have any saves like that.)

 

[aicheats]

I ran into this problem just the other day too, playing on emperor. I was attacking a city with cavalry and my first three guys to attack were 3%, 3% and 6%. And all three survived! I should never win three battles in a row under 10%, the AI cheats!

The odds of that happening is .03 * .03 * .06 = .000054 = .0054% =

5.4 in 10,000.... or 1 in 1851

If that happened you would be incredibly lucky.

I guess you dont have an appeciation for how rare these events are...

No one studies math anymore I guess.

For the 30th time.. this event is very rare.. but possible. What I'm saying is I see these type of events most every game I play. It simply isnt mathematically possible to occur randomly, the AI has to be cheating. You may not grasp the mathematics of it and choose to not believe... or you may beleive I'm lying... if you did grasp the math of it... you would know the AI is in fact cheating.

 

[pob]

Maybe you're just the random person who loses more than average?

 

[aicheats]

I'm saying you only started paying attention when you lost the first battle, so it can't be included in your sample.
The odds of you considering something was wrong becomes only 1% (0.1*0.1) which isn't that significant. i.e. after any battle you lost at ~90% odds, you get angry and pay attention to the next couple of rollls. In 1% of these circumstances you hit another 2 losses.

(I'm not outright refusing to listen to you. Post a save where you are very ahead of the AI and I will test it. I don't have any saves like that.)

Ok I'm done talking to you... your logic is wrong. You are confusing the fact that a previous event does not effect the outcome of a future event (roll 2 has nothing to do with roll 1) with the fact that there are certain odds of several event happening.. whehter they happened before you were there or not.

Last attempt to get though to you:

Walk into a casino, see the roulette wheel scoreboard, notice the last 3 rolls are black. Of course the next roll to be black is still 18/38... the previous 3 rolls do not effect the outcome of the next roll. But, the odds of that the last 3 rolls would be black is (18/38 * 18/38 * 18/38). The fact that you were not in the room to see it doesnt change the odds of it happening. That event did happen, it had a certain chance of happening.

Maybe another shot... a baseball player has a batting avg. of .300. That was his odds of getting a base hit. Period. If his avg was .999 would assume there is some funny business going on, you wouldnt ignore that .999 average and start with a new test.

 

[pob]

I don't know what more I can offer. You say you think the AI cheats; I say it doesn't. Until you do something to fairly prove otherwise, we go with the null hypothesis of "the AI doesn't cheat roll". Do a t-test on a save where this happened and then we can talk