A Better AI.

[kettyo]

A good deal that brings you "traded with our worst enemy" demerits from half the world isn't necessarily so good.

I like and would miss this aspect of the game.
Alliance politics is a good part of Civ4 diplomacy.

I actually like you just can't be in good terms with everyone. It's almost certain you'll have some conflicts in every game (with enough civs).

Overall my favourite strategy game is Master Of Orion but this is the only major flaw there. In MOO you can do easily to be peaceful with everyone while they kill each other and you can win a diplomatic victory very easily.

 

[Tiberias]

The goal of improving the AI should be to gradually reduce the level of outright cheating it needs to do to maintain a given difficulty.

Agreed; I think everyone agrees that the ideal AI would be a "Deep Blue" equivalent that could beat even the best players without any cheating necessary. Of course, that will be difficult if not impossible, so some small amount of cheating will be necessary to maintain competitiveness against the best of players. I'd prefer more "invisible" cheating, however, such as production and research bonuses, than more outright cheating, like loads of starting units or nearly-free upgrades.

 

[Uncle_Joe]

Maybe certain "tricks" like chop-rushing wonders should only exist starting at Prince or Monarch. Also, maybe we could limit the dagger opening strategy to Prince, minimum. This would allow Noble players to basically play like Noble players, while giving more veteran players (Prince/Monarch) a challenge, as they discover advanced concepts (chop rushing, slavery abuse, etc).

This is something I dont agree with at all. I think the AI should play its 'best game' at nearly all levels (ie, at least Noble+). Having to use different levels to get different tactics is a not something I'd like to see.

 

[Elhoim]

Yeah, I think that the AI should always play as intelligent as it can. Then fine tune the difficulty levels with handicaps and bonuses.

 

[kettyo]

Yeah, I think that the AI should always play as intelligent as it can. Then fine tune the difficulty levels with handicaps and bonuses.

Absolutely agree.
Beginner players can also learn the most from watching the AI so it's silly to turn down their playing skills.
Handicaps is the better way.

 

[kettyo]

I think the main problem with the "computer player" is that it is expected to do two or three sometimes mutually exclusive things all at once: (1) play according to his "character" and "circumstances" like religion, etc (2) try to win (3) prevent the human player from winning easily. Suppose you are playing an MP game of civ but you cannot attack unless your opponent has a different religion and you are expected to attack the leading player no matter what. How frustrating would that be?
The AI is expected to do so many silly things like refusing an advantageus offer just because the other player "has the wrong religion" and it is also expected to play "smart". You just can't be stupid and smart at the same time.
Perhaps, it is best to have 2 different builds of this AI, one historic AI and another human-like AI that just ignores religion, gifts, borders, etc.

Very good points.

First i think there should be only one build. The best possible. Optimized for singleplayer.
It's already a huge job to improve, bugfix, and fine-tune the current one for Blake and Iustus.

The idea of throwing away the relations aspect of the SP game i would not like.
AI players could play like that in MP games (where everyone is enemy after all) but in SP i don't think so.

AI leaders should follow their personality though in the same time they should actively try to win the game through spacerace, culture or domination (according to their personality and abilities).
This is surely not impossible. Just try to win by different approaches and playing different habits.

Actively trying to prevent others from winning is an interesting question.
I think only enemies (annoyed and furious) should attack if this feature would ever got in the game. This way it would be fun. And applied to leader AI's too of course.

 

[Uncle_Joe]

Actively trying to prevent others from winning is an interesting question.
I think only enemies (annoyed and furious) should attack if this feature would ever got in the game. This way it would be fun. And applied to leader AI's too of course.

'Trying to prevent others from winning' does not necessarily mean that these civs need to 'attack' a winning Civ. Simply denying them resources or luxuries could suffice. But even that is a slippery slope to run on for an 'AI only' mod. I imagine many players would be upset if their 'allies' suddenly started cutting off their resources when they were building the Space Ship or somesuch....

 

[Elhoim]

If we want the AI to behave like MP, we should give them the ability to quit when they are losing.

 

[kettyo]

Following the new AI doctrine of weighting threat on cities and defending them according to this i think a new negative diplomacy modifier would be very important to heavily penalize backstabbing your friends.
Mainly because of the human player who can declare war on anyone without restriction.

 

[Arlborn]

If the AI ever try to prevent a player to win in any future version, it only should happen for Furios or Annoyed AIs and REALLY depending of their personality.

And I cmplete agree in reducing some stupid bonuses of the AI, but dont make it too flat.

 

[kettyo]

I imagine many players would be upset if their 'allies' suddenly started cutting off their resources when they were building the Space Ship or somesuch....

In ideal case allies should help each other win the game (sharing the victory a bit somehow like in SMAC) to prevent an enemy to be the winner.

 

[Roland Johansen]

Building so many AI units as it does now and letting it still have free upgrades and negligible WW isn't making the AI smarter, just better at exploiting its cheats. Both WW and unit upgrades were balanced based on a lot fewer units used in wars. Right now fighting wars at deity feels a lot more like fighting against the AI cheats than anything else. Making 50+ units just to capture a few cities only to find that you're crippled by WW while the AI is still running at full research/production and just upgraded everything to the next tier of military tech with its free, innate, permanent Leonardo's Workshop anyway is 1) a completely artificial situation that has nothing to do with what Civ is supposed to be, and 2) more importantly, not fun in any way.

The goal of improving the AI should be to gradually reduce the level of outright cheating it needs to do to maintain a given difficulty.

While I understand that this must be frustrating, I would like to point out that the changes proposed by Blake for his next build are not going to change anything about the war weariness aspect of what you're describing.

War weariness is related to winning and losing battles in areas where another culture is stronger than yours (so normally inside the territory of an AI). So whether you're playing at deity level or at settler level, the destruction of massive amounts of foreign units and capture of their cities in their territory will only result in war weariness for you and not for the AI.

If you want to avoid war weariness, then you have to destroy the AI units inside your own cultural borders. The AI will then suffer war weariness (and the modifiers on AI war weariness will then be in effect).

Simply said, war weariness is only hitting the aggressor and not the defender. This is partly for balance reasons (making war not the single best option to win a civ game) and partly because of the fairly realistic notion that someone who is desperately defending his home is less likely to complain about the war then someone who is fighting somewhere in a foreign country for a cause he doesn't understand. It's not completely realistic, but then again it is a game.

If you want to know exactly how war weariness functions, then you can read this War Academy article.

 

[cf_nz]

Your changes including things which modify the save file format, so I cannot investigate this crash.

If you can post a zip with your sourcefiles (leave out all the python and boost stuf, you can also leave out any files you did not modify, if you have that handy), then I can look into the issue. Without your sources though, it is impossible to even run these saves.

I still hope to look into this, but the plan is to post a build in the next 12 hours, and most of those I will be sleeping, so if this is a real bug, the fix will likely not make it in the next build.

-Iustus

Ok, I'll get those to you (Edit: I've had a 10 min look through the code and have found a potential issue not related to BetterAI, I make those changes first and try the save again). I've got to go to work now so won't be for about another 10 hours, sorry. Damn time zones.

 

[solanacea]

Agreed; I think everyone agrees that the ideal AI would be a "Deep Blue" equivalent that could beat even the best players without any cheating necessary. Of course, that will be difficult if not impossible, ...

It seems to me that not everyone wants a brutal, no-nonsense, Deep Blue-like AI. Most want an AI with character: a chess player who will take your unprotected knight only if he is angry enough with you or one who will try to checkmate you with two bishops all the time (cultural victory anyone?) because it's his character. Could such a computer defeat Kasparov in chess? Possible if it started with 7 queens and 8 bishops.
Civilization is a more difficult game than chess for computers because it is not a zero-luck, perfectly calculable game. And it took IBM almost a decade to build a supercomputer that could challenge the best of mankind in chess. No wonder they are not in the Better AIgoodjob: ) business.
Honestly, I can see only two ways to go here: 1. A historic AI that cheats heavily to compensate for all the historic mistakes it is expected to make, 2. A brutal, no-nonsense AI that is designed to challenge humans with minimal cheating. I am no programmer but little bit of both may be too much to ask.

 

[MishGolden]

Take a look at my post here in this thread about installation.

The pic from that post:

From another post, mod install:

You can tell if you have it installed correctly if you see "=== Better AI ===" in the score breakdown (this is an old picture, it now shows the build date as well).

You see the score breakdown by mousing over your player name in the 'scoreboard' with the alt key down:

I will see if Blake can get those pictures posted in the first post of this thread.

-Iustus

(Apologies, in advance, if this question has already been answered in the over 70 pages of this thread.)
I can't seem to verify that I installed "BetterAI" correctly. I followed the first procedure in the thread quoted, but when I mouse over score, the words "Better AI" do not show up.
Any suggestions on what I might be doing wrong?
Thanks so much.

(Installing it as a mod does seem to work, though.)

 

[Blake]

Here are the justifications for FLAT bonuses across the board.

AI barb/animal bonuses: The AI is tactically weak which means it suffers more at the hands of barbs, this tactical weakness is applicable at every level.

AI upgrade bonuses: The AI makes relatively poor use of it's upgrade cash relative to what even the dumbest human would - even a not so smart human will instinctively upgrade units near hostile borders. The AI, even improved, does not have those instincts - this is true on every difficulty level. The AI also relies more on "Quantity over quality" when it comes to winning wars and making them pay full price would harm them more than it does a human who targets upgrades much more smartly.

AI supply bonuses: Again "Quantity over Quantity". If the AI is to have a prayer of succeeding, it needs to send in more units than a human would, and/or the units spend more time milling around rather than getting to the point of the invasion. Hence a discount is justified.

AI war weariness bonuses: Again Q/Q is a factor, but also the AI is intrinsically worse at manipulating this game mechanic, also at higher difficulties the AI relies all the more on quantity while humans are all the more better at manipulation hence a scaling on difficulty is justified.

AI inflation bonus: Ok I don't have a good reason for this one, other than as an extra free pass on top of other commerce free passes. Regardless, since inflation is already reduced by other difficulty-sensitive bonuses there's no need to scale inflation by difficulty too (in particular there are the per-era bonuses which means AI expenses go down over time.

 

[Zombie69]

The issue here is not whether to keep the original levels of bonuses for the AI. I think we all pretty much agree that's not the best thing to do. But the bonuses should remain scaled by difficulty level, otherwise you run into other problems like Roland mentioned, plus you lose the challenge for the human player at high difficulty. Like i said, just keep the sliding scale but modify the range accordingly.

 

[Zombie69]

Sorry, i was replying to a post on the previous page which i thought was the last.

Regarding Blake's latest comments :

- Quantity over quality
As has already been mentioned, Deity AIs have more quantity, and therefore need a bigger bonus to be able to upgrade / supply those units.

"at higher difficulties the AI relies all the more on quantity while humans are all the more better at manipulation hence a scaling on difficulty is justified."

Isn't that the case for all the bonuses that were previously scaled and that you made flat? The way i see it, it is.

 

[JimMac99]

CustomAssets "should not" be loaded be loaded in MP, but CustomAssets ARE loaded in MP so for now it is sufficient for all players in a MP game to have the DLL in their CustomAssets folder. The Correct way is probably using it as a mod for all players but I've never actually tried doing that.

PS. Hopefully the next build will be tomorrow... had quite some traveling to do today. Sorry for not getting a lagfix build out...

A quick thank you for your prompt reply to my query. (I hope Firaxis dont rush to fix this 'bug' ,as I quite like the way it works at present.)

And now... continue your arguments over the bonuses...

 

[Uncle_Joe]

Honestly do the AI bonuses really have to have an impact on this mod? Anyone who doesnt like the bonuses can change them at will.

IMO, concentrate on making the AI play as well as it possibly can and let individual players worry about how much handicap they want the AI to have.