How the AI knows the progress of your Wonder building
- Thread starter Mylene
- Start date
The test in the mountains is interesting. For clarity, I suggest you SIP every player and remove all units but a single warrior and a worker, so the human only needs to fortify the warrior, sleep the worker, select future tech, build wealth, and set the research slider to 0% (in that order, the order of your clicks may make a difference in the RNG). Give everyone only Mining, Masonry, and Currency, which shouldn't really make much difference but will reduce the number of things happening so it's easier to notice when the AI changes its moves.
Presumably this will result in history repeating exactly, over and over. Try it a few times to make sure.
Next, try changing things at certain times. Perhaps at turn 10 you change your science build, or increase the slider, or tell the worker to move, or switch tiles worked, etc. Note the changes you make and see if the AI changes its actions in any way when you play differently.
Next, try building the pyramids instead of wealth. See if the AI builds the same as when you build wealth, or if it changes. If the AI builds the same, it means the AI is most likely NOT cheating and the RNG is most likely NOT affected by your build choice. This is not definitive proof BTW. If the AI changes its build, it means either a)the AI is cheating, b)the RNG has been affected, which in turn changes the AI build, or c)something unexpected.
Repeat the 'build Pyramids' test a few times to make sure the AI plays the same each time when you're building the pyramids. Then do the same as you did in the 'build wealth' test, changing the same things you did in that test at the same times, and see how the AI reacts. I suspect that changes which affected the AI in the first test will affect the AI in the second, and changes that did not affect the AI in the first test, again will not affect the AI in the second. At this point we'll have a pretty good idea of what player actions affect the RNG/AI decision making.
Try going wealth -> pyramids at different times, and see how/if the AI reacts to your build changes. Then try going pyramids -> wealth at different times, again seeing how the AI reacts. If there is no reaction at all, no matter when you make the changes, then the 'cheating AI based on your build' hypothesis is disproved. If there is a reaction, AND it follows some logical pattern, then we have a little more evidence to support the 'cheating AI' hypothesis. This will not be definitive proof BTW, it's much easier to disprove things than it is to prove them. If the AI reacts in a seemingly random way, we have a little more evidence to support the 'it's all RNG' hypothesis. Again, not definitive proof.
Presumably this will result in history repeating exactly, over and over. Try it a few times to make sure.
Next, try changing things at certain times. Perhaps at turn 10 you change your science build, or increase the slider, or tell the worker to move, or switch tiles worked, etc. Note the changes you make and see if the AI changes its actions in any way when you play differently.
Next, try building the pyramids instead of wealth. See if the AI builds the same as when you build wealth, or if it changes. If the AI builds the same, it means the AI is most likely NOT cheating and the RNG is most likely NOT affected by your build choice. This is not definitive proof BTW. If the AI changes its build, it means either a)the AI is cheating, b)the RNG has been affected, which in turn changes the AI build, or c)something unexpected.
Repeat the 'build Pyramids' test a few times to make sure the AI plays the same each time when you're building the pyramids. Then do the same as you did in the 'build wealth' test, changing the same things you did in that test at the same times, and see how the AI reacts. I suspect that changes which affected the AI in the first test will affect the AI in the second, and changes that did not affect the AI in the first test, again will not affect the AI in the second. At this point we'll have a pretty good idea of what player actions affect the RNG/AI decision making.
Try going wealth -> pyramids at different times, and see how/if the AI reacts to your build changes. Then try going pyramids -> wealth at different times, again seeing how the AI reacts. If there is no reaction at all, no matter when you make the changes, then the 'cheating AI based on your build' hypothesis is disproved. If there is a reaction, AND it follows some logical pattern, then we have a little more evidence to support the 'cheating AI' hypothesis. This will not be definitive proof BTW, it's much easier to disprove things than it is to prove them. If the AI reacts in a seemingly random way, we have a little more evidence to support the 'it's all RNG' hypothesis. Again, not definitive proof.
so we have this giant thread, people who take the time to check the code, and similar but less thorough tests being done. Anyone willing to do what I suggested?
And no, I won't do it myself. What evidence we have suggests it's all RNG or it's an alternative explanation, such as an AI reaction to you getting the tech first. And tbh until someone presents contradictory evidence, I'm satisfied with what we've got. But it does seem that there's a lot of interest in this and I'll be surprised if someone (Mylene?) doesn't rise to the occasion.
And no, I won't do it myself. What evidence we have suggests it's all RNG or it's an alternative explanation, such as an AI reaction to you getting the tech first. And tbh until someone presents contradictory evidence, I'm satisfied with what we've got. But it does seem that there's a lot of interest in this and I'll be surprised if someone (Mylene?) doesn't rise to the occasion.
TheMeInTeam
If A implies B...
- Joined
- Jan 26, 2008
- Messages
- 27,995
What we need to find is what sets of human actions influence the RNG and which ones do not, as a separate puzzle from whether wonder commitment influences AI start times. Once we know which things push us further down the RNG string, we can avoid them.
The big problem would be if switching to a wonder influences the RNG, because at that point we'd truly need a crapload of tests...whereas if lots of human moves don't effect RNG this becomes quite easy.
Haven't any of you tried to save scum battle outcomes w/o "new random seed on reload"? I've done a few tests with that in the past and it is actually rather hard to do something same-turn that influences the game to the extent where you do not get an IDENTICAL combat result each time (even down to damage done), where something that tripped the RNG should theoretically cause different combat results (IE a shift down the RNG results line). I have serious doubts that tile swapping influences RNG, but maybe the governor on does so? We need to know this first such that random noise isn't influencing our evaluation of the AI wonder evaluation being cheating or not.
The big problem would be if switching to a wonder influences the RNG, because at that point we'd truly need a crapload of tests...whereas if lots of human moves don't effect RNG this becomes quite easy.
Haven't any of you tried to save scum battle outcomes w/o "new random seed on reload"? I've done a few tests with that in the past and it is actually rather hard to do something same-turn that influences the game to the extent where you do not get an IDENTICAL combat result each time (even down to damage done), where something that tripped the RNG should theoretically cause different combat results (IE a shift down the RNG results line). I have serious doubts that tile swapping influences RNG, but maybe the governor on does so? We need to know this first such that random noise isn't influencing our evaluation of the AI wonder evaluation being cheating or not.
Not right now, maybe in a few days Keilah.
It sounds like a lot of small pieces of work, for the slim chance to get a definite result.
I was aware of the combat results mechanics (years ago when they made me mad
i would attack in a different order with my units, sacrifice the weak ones on the battles that are locked in as losses, to get "passed" them), but i didn't think the whole game reacts so much to little things like changing city tile. For that reason i fear this cannot be proofed...or maybe that is good cos it keeps us thinking, and those little mysteries keep a game alive
It sounds like a lot of small pieces of work, for the slim chance to get a definite result.
I was aware of the combat results mechanics (years ago when they made me mad
i would attack in a different order with my units, sacrifice the weak ones on the battles that are locked in as losses, to get "passed" them), but i didn't think the whole game reacts so much to little things like changing city tile. For that reason i fear this cannot be proofed...or maybe that is good cos it keeps us thinking, and those little mysteries keep a game alive - Small changes made by the human can make large differences in the course of the game due to requiring a different sequence of calls to the random number generator.
- Working a mine means an extra call to the random number generator due to the resource-discovery chance.
- The AI does not spy on your wonder progress to determine whether to build wonders of their own.
I think every one of these points has been exhaustively proven in this thread, so before we keep rehashing them could everyone please take a moment to carefully re-read things like KidR's post on page-2 and Dirk1302's post on page-4?
Remaining open questions - does the AI "spy" on your wonder progress to determine whether to research a wonder tech or not? Is the AI less likely to start researching a wonder tech when you have it already? Will the AI ever stop researching a tech mid-way if you finish discovering it?
Me too.
i fear this cannot be proofed...maybe that is good cos it keeps us thinking, and those little mysteries keep a game alive
True, it'll be a nice opportunity for us naysayers to speak up whenever somebody mentions it in a walkthough game - "what a load of rubbish" we can say
I don't think so. In my test I put hammers into various wonders and world history played out exactly the same every time. That shows that the part-wonders did not change the teching of the AIs, because when anyone in the world techs differently it causes different RNG rolls to be required. I saw that happening when history changed about 15 turns after I built a library, even though I was 100% isolated. I assumed it was because even isolated civs affect tech discounts worldwide, I don't know though.
True, it'll be a nice opportunity for us naysayers to speak up whenever somebody mentions it in a walkthough game - "what a load of rubbish" we can say
lol up to you my friend, many of these somebodys who think that the AI is less likely to build a wonder if you started earlier are Deity players
While i went out on a limb here as well, and god knows how exactly it works, the many peoples who are sure that a Deity player is 100% wrong with this is baffling.
TheMeInTeam
If A implies B...
- Joined
- Jan 26, 2008
- Messages
- 27,995
Ah, good call. I had forgotten about the small % resource pop per mine worked per turn.
I'm not seeing a lot of evidence for "AI cheats to see if you're building wonders" then. Probably, you can still help yourself by researching a wonder tech ASAP as the AI de-values techs that have been researched, but wonders look pretty hard to game beyond that...
More than one deity player has claimed that having some military is a legit deterrent from early DoW over the years...even after it was proven wrong via people looking at the code. The naysayers have a legit point here: anecdotal evidence is about as reliable as a person's memory of RNG outcomes being fair, and there's no reason deity players are an exception. Without solid code-based evidence or a conclusive test w/o noise, *no* player can convincingly assert that the AI cheats to detect wonders, even though it absolutely *does* cheat to detect your power rating, trades with worst enemies when it hasn't met you, etc (thus showing precedent). If there's one thing playing Firaxian games have taught me over the years, it's not to rely on them for logical patterns, balance sense, decent controls, or other basic expected game mechanics
.Remember we're talking about the same game that will hide up to +5 diplo relations from the player, move units against your will, and have civs get angry because they settle near you. Wonder cheating absolutely is NOT out of the question in a model like that, but it isn't guaranteed, either...and evidence for it is looking pretty shaky.
Hehe, I doubt playing on deity makes a person less susceptible to superstition.
The people sure there's something in the code when programmers have looked through it and report there isn't, is also baffling
The people sure there's something in the code when programmers have looked through it and report there isn't, is also baffling
Well yip, iam just saying Deity cos there i feel enough pressure that the AI can compete for wonders.
I also look at the dates very closely, and when or when not i get a wonder.
Your post is really good TmiT, there is not much logic behind the diplo mechanics and all the things the AI knows in Civ4, i wouldn't be surprised if they get information on everything you do via the rng.
But i like this. A single player game without a massively cheating AI would be boring, we could see that in Homm3 where on a random map you had to ally up several AIs on 200% to have a (usually small) challenge.
I also look at the dates very closely, and when or when not i get a wonder.
Your post is really good TmiT, there is not much logic behind the diplo mechanics and all the things the AI knows in Civ4, i wouldn't be surprised if they get information on everything you do via the rng.
But i like this. A single player game without a massively cheating AI would be boring, we could see that in Homm3 where on a random map you had to ally up several AIs on 200% to have a (usually small) challenge.
azzaman333
meh
/thread
AutomatedTeller
Frequent poster
Move units against your will? I've never heard of that.
The worst enemy thing, say you know Joao and trade with him, then Catherine comes along and says "Ah, so you're this "TheMeInTeam" my enemy Joao has been talking about - you must be the one who gave him Alphabet, you bastard".
Seems reasonable that when they meet you they can put the name to the face and go from there. Is that cheating? No more than it is when we see AIs getting techs from nowhere and thinking "well there's only 3 of us on this continent and I sure know I didn't trade it to you". Or any other simple educated guesses we make from studying the various info screens.
Don't know what's come over me this weekend - I seem to be in a "disagree with everyone" mood Many apologies.
Seems reasonable that when they meet you they can put the name to the face and go from there. Is that cheating? No more than it is when we see AIs getting techs from nowhere and thinking "well there's only 3 of us on this continent and I sure know I didn't trade it to you". Or any other simple educated guesses we make from studying the various info screens.
Don't know what's come over me this weekend - I seem to be in a "disagree with everyone" mood
Many apologies.Don't know if this is what he means but if you share a tile with a 'friend' and war breaks out, your units will be removed from the shared tile, often very inconveniently. One game I had a small surprise stack all set to take out Sury's Holy City on the 2nd turn of war - Kublai Khan moved a scout onto the tile my units were on, then declared war, forcing them out of range of the city and giving Sury time to reinforce. I was displeased.
On the other hand, Kublai IS supposed to be creative... genius level move IMO!
Zechnophobe
Strategy Lich
My Skepticism alarm goes off in Civ just like it does in the real world.
Of course with Civ, there really is a secret Cabal of 'developers' who are controlling things behind the scenes to bring me down.
Of course with Civ, there really is a secret Cabal of 'developers' who are controlling things behind the scenes to bring me down.
Deity player or not, that's nonsense. The AI does not get "information on everything you do" via the rng, as has been demonstrated over and over in this thread.
What you do (whether seen by the AI or not) affectst the RNG and therefore the AI as well as you. Setting a worker to auto-explore might cause the AI not to beat you to a wonder because they build something else or/and it might cause a slave uprising in your capital on the next turn.
So far so good. The main thing is however, that a change in the RNG sequence is as likely (for practical purposes) to make an AI decision worse as it is to make it better. That means you could also have the same situation in your original savegame, where the RNG is in a different state and switching to the mine to increase production leads Mansa to start building the Oracle - even though it's certain he will fail.
To use an analogy if I go to a shop to buy a lottery ticket five minutes before you do I will affect your personal outcome of the lottery - but you don't get any information about me at all. And since you have limited information your expected outcome (and mine) is the same. Same in Civ. Only you have hindsight because of the reload game feature.
As for your other point, try playing chess against an AI. Good luck. As for Civ 4, one of the things I like about it (and many other players) is that the AI gets a bonus to production, science, starting units, etc. but plays more or less by the same rules as you do. The diplomacy mentioned above has some very minor exceptions, but they are few (hidden +5 is just a boni for me). There's also the increased range of visibility for naval units. If you remember Civ 1 AI "wonder building" you will know how much CIV AI has evolved. Okay, Civ 5 seems to be a step back in that regard.
Dirk1302
Deity
I agree, Civ4 AI has some flaws but it's way beyond any other Civ AI (including ciV) in the franchise.
Oh yip, how could i overlook you. Iam sure with 4 posts you already showed that you are qualified to spot nonsense in Deity games
Similar threads
