AI getting smarter at high difficulty levels?!

[Psyringe]

At what difficulty?

I would assume the AI bias does come into play until prince or monach like in the old civ games, and that below noble there would be a player bias.

Since the original claim was that the AI cheated on Noble difficulty (as opposed to what is written in the manual), I chose that difficulty level.

You can easily check yourself whether there is a RNG bias for higher levels. Just load my save game, enter the world builder, save it there, leave the game, and then load your world builder save as a scenario. You can now specify a difficulty level.

Btw, I've never heard of a combat RNG bias for the AI in old Civ games. Do you have a link to some analysis on that topic?

 

[Thormodr]

The AI may certainly "cheat" on higher levels with production bonuses but I have never seen any solid evidence that the AI cheats with the RNG.
I have seen improbable wins both ways. I have won battles that I had no business winning too. Zorlond, it really does seem like sour grapes to me.
I think Firaxis did a pretty good job with the AI. Like other posters have mentioned if we really got a much better AI, the game would be delayed 4 years and nobody's computer would be able to run it.
I accept the game for what it is. If you don't like it, form your own game company and do better. Don't want to do that? Play another game then. I daresay you won't find any truly intelligent AIs in any other game. Why demand it from Firaxis?
Thanks Psyringe for contributing something worthwhile to this thread.

 

[Zorlond]

Stop with the recitation of failure! "Can't do it! Can't do it! Can't do it!" Making me sick. Can we fly? "Can't do it! Can't do it! Can't do it!" Humans recited that for centuries until someone smacked us over the head and pointed out that yes, we CAN fly.

The longer we assume we cannot possibly make a decent AI to play a game, the longer we hold ourselves back from developing an AI that could actually do something useful. So stop the recitation of failure!

To Psyringe: That's why I compared total surviving units as well as stack vs stack. If things truly were equal, then what type of units were involved shouldn't matter, it'd all balance out in the end. Which is why you set up sixteen units in each stack in the first place, to increase the likelihood of notable balance (or imbalance). If it was just six units, then I'd see your point. But you set it up with 96 units on each side, and the final tally of survivors should make a clear indication of any favoritism.

 

[Thormodr]

Man, if we all think positive thoughts the earth will heal itself. Jeez, the whole problem all along was that we weren't thinking positive. Those drugged up hippies with bellbottoms the size of circus tents were right!
Everyone join Zorlond for a group hug and miracle of miracles, we'll all have a genius level AI.
Back in the real world, this can't be done at the moment. I am sure in time there will be a more than competent AI that doesn't need any cheats. I sincerely believe that. I also think the AI is much better than it used to be. A step in the right direction.

 

[Sorceresss]

Man, if we all think positive thoughts the earth will heal itself. Jeez, the whole problem all along was that we weren't thinking positive. Those drugged up hippies with bellbottoms the size of circus tents were right! Everyone join Zorlond for a group hug and miracle of miracles, we'll all have a genius level AI.

 

[martin5959]

Just cheat. After about 4 cities and saving & reloading a couple of times, activate world builder. After some tries U get a green box and U can add anything - units, buildings, tech. Because U cant beat AI at the higher levels!

 

[Zorlond]

Self-fulfilling prophecy.

Look it up.

 

[Psyringe]

To Psyringe: That's why I compared total surviving units as well as stack vs stack. If things truly were equal, then what type of units were involved shouldn't matter, it'd all balance out in the end. Which is why you set up sixteen units in each stack in the first place, to increase the likelihood of notable balance (or imbalance). If it was just six units, then I'd see your point. But you set it up with 96 units on each side, and the final tally of survivors should make a clear indication of any favoritism.

So, you first claimed that the AI cheats by getting favoured by the RNG in battles. You couldn't be bothered to provide a proof, so I devised a test and attached it to my post so that anyone can check it. We couldn't find any evidence for your claim, despite your telling us that the cheating is so plain obvious that it just requires a little attention to spot it. We also couldn't find any evidence for any of your other claims, like the "magical unit creation", for which I made a test too, since you didn't provide one.

Now, since there's absolutely no evidence on which you could base your claim, you turn around by 180 degrees and now claim that the result of our test (which very soundly proves your hypothesis to be wrong) in fact proved that "something else" is wrong because the results are "too imbalanced". You neglect the fact that a certain degree of imbalance is to be *expected*, no matter how many trials you perform. This is not a proof that "something" about the RNG is wrong, it's called random variation. If you think that the results are too far away from the middle to be the result of random variation, we can easily increase the number of trials, which will further balance the random variation.

Since your previous performance in this thread leads me to believe that you probably can't be bothered to do that yourself, I'll prepare a test with more trials and upload it shortly.

 

[Aeson]

Combat uses the same formula for Human vs Human, AI vs AI, and Human vs AI. Regardless of difficulty level. The only variation is with Human vs Barbs or AI vs Barbs, as defined in the XML.

There is one unit production cheat (ie. not a bonus defined in the XML) the AI gets. The first military unit built gets some free production towards it. Not sure if it's on all difficulties, but definitely on Noble. (This really is only a factor in Warrior rushes.)

 

[Psyringe]

@Zorlond: So, here's a test with even larger numbers. We have 64 battles now per island, which means 6*64=384 trials (ignoring the tanks). If your hypothesis is true, and "something" about the RNG is wrong, then it should show up again. If my hypothesis (the null hypothesis, "there is no bias") is true, then about 50% of the battles should be won (and the numbers should be even more balanced now).

Perform the test as before with the following save:

View attachment CombatRngAiBiasTest64.zip

Then press F9 and click on "statistics". You can now see exactly how many units of each type you've lost. In my test, I've lost 33 warriors, 33 swordsmen, 28 spearmen, 32 riflemen, 30 SAMs, and 38 knights. So I lost 194 units. meaning I won 190 battles and lost 194.

As expected, the larger number of trials resulted in an outcome even closer to 50:50.

If you still think that "something" is wrong with the combat RNG, I'd really like to know what and why.

 

[Armorydave]

Great analysis Psyringe. I play exclusively at Monarch level and haven't noticed any clear example of combat bias (or any other "hidden" bias). I have even seen barbarians win battles against the AI Civs something that I *never* saw in well over a 1,000 of hours of Civ 3 play.

 

[Efexeye]

There is no AI combat bias- just strings of bad luck that people point to as "proof" that the AI "cheats".

 

[Sorceresss]

There is no AI combat bias- just strings of bad luck that people point to as "proof" that the AI "cheats".

Dogmatic...

 

[Thormodr]

Exactly. Humans have selective memory. If you play a game, lose a few key battles you are inclined to think that the computer cheats. Of course you forget all the times you won when you shouldn't have. It evens out over time because *gasp*, it's random. Wow, who would have thunk it.

 

[EL_OSO]

There is no AI combat bias- just strings of bad luck that people point to as "proof" that the AI "cheats".

I was guilty of thinking that way until I figured out how to display the odds and now I choose when to fight and when to fortify or retreat. Now my games are more manuever orientated and I find that the AI is foolhardy enough to attack me consistently at low odds. I still lose some battles but not nearly as many as before.

The AI in Civ IV is better than previous versions but the one design theme that really bugs me is the level of intelligence that the player is provided with. In previous versions, I had no idea where the AI's weakpoint was without spending gold to investigate a city or planting a spy. Now it's all displayed including their combat promotions. I still feel as though warmongering is a mechanical excersise due to this. The only thing that changed is the SoD is replaced with a god's eye view of the battlefield.

Instead of giving the AI bonuses on higher levels they should have implemented more "fog of war" IMHO.

 

[Efexeye]

Dogmatic...

What? I don't understand...

dogmatic-characterized by assertion of unproved or unprovable principles
of or pertaining to or characteristic of a doctrine or code of beliefs accepted as authoritative
relating to or involving dogma; "dogmatic writings"
wordnet.princeton.edu/perl/webwn

I think it's been pretty well proven that it is a RANDOM NUMBER GENERATOR that determines combat. Ever flip a coin 10 times and have it come up heads 9 of the times? Well, damn, the universe must be biased against tails, then, right? *eyeroll*

RANDOM means that the RNG has just as good a chance of rolling a "7" (or whatever number) on every single turn, regardles of what else happened prior. Sometimes you're gonna have "runs"- it's why people get on streaks shooting craps. I still have yet to see ANY proof of where the combat system somehow favors ths AI.

 

[Zorlond]

Well, on another note, I just stumbled on evidence of the AIs in cahoots with each other, with no declaration. Imagine a U-shaped continent, large world (this continent is the only one on the map, aside from one large island, so it's freakin' big). I'm sitting pretty on one end. Xander and Ghengis are sitting at the FAR end. Between us are a minimum of two other countries.

Not only do they both declare war on me, not only do they ignore each other (no alliance), and not only did they both snag Open Borders from every other country in the game, but they then had the nerve to send 6+ units each at me, at the same time. Land assault. No ships.

At the very least the others shoulda chomped down on the both of them while their troops were away.

(yes, save can be provided, slap an e-mail up and I'll send it. note, save is at the point of the first wave of troops, mixed unity, second wave comes in about 2-3 turns behind those)

 

[Psyringe]

Erm, what exactly is your complaint? The AI is programmed to go into dogpile war mode, this is one of the three war modes. It does so against you, and it does so against other AIs. The AI may even be able to coordinate its attacks, which is only fair, since you can do the same.

So when two players get together against an AI, then this is okay, but when two AIs get together against a player, then it's cheating?

What exactly is your claim, and in which way do you think your save proves it?

Just upload your save here. In case you can't, I'll send you my mail address in a PM.

 

[Zorlond]

The AI is programmed to go into dogpile war mode

Well at least you admit it. Too bad you failed to see my point. Let's take it from an all-human-player perspective. Would you call it completely fair in a six-player game for five of the players to spontaneously gang up on one? I sure as hell wouldn't, and I'd never play with those five ever again, regardless of who got to be the 'one'.

Another point, lengthy, distant wars are supposed to be expensive, weakening the kingdoms in entirety. Again, I ask why the other AIs didn't capitalize on this. Another point, they were the aggressors (yeah, no surprise from these two), I was happy to stay where I was and leave everyone else out of it. Last point, Isabella directly to my south claimed not 10 turns previous that Ghengis was a mortal enemy. Now she and Gheng are best of pals?

Face it, this situation is ludicrous.

 

[henrycccc]

The game is completely fair, don't blame your inability to wage a war on it. THe AI cheats in only minor insignificant ways, so if your can't win stop looking for cheats and just go back to setter.