Genetic AI Project

[Quaetam]

About the genetic AI: If it got smart enough, wouldn't it realise there are better things to do than video games

 

[GoodGame]

I think option #2 is a cool idea, but it's not worth your time if you're not getting paid (unless you're working on a Ph'D in this topic). It's probably not an easy mod, plus the AI project would have to start over from scratch honestly. Perhaps you could try to offer your services to Firaxis to get some sponsorship on this?

Okies. Big issue - cheating.

The AI cheats. It can see units and cities that it shouldn't be able to. I've confirmed this with tests.

Two options:
1) Leave it as it is. While it may cheat it doesn't cheat much, or in really obvious ways.

2) Try and fix it. This would make the AI worse, but would give us a more pure, and probably fun, game. It would require quite a bit of work and a load of new code to deal with unknown factors.

Personally, while I'd like to say 2... it's alot of work, and it'd probably be me doing it, so I'd like to hear other people's opinion before I dive in. How crucial is a non-cheating AI? IMO Firaxis made a decision to have it cheating for the sake of competance, and perhaps for the sake of simplicity, however I believe it could be fixed without that much of a loss of competance.

 

[GoodGame]

EDIT: Anyone having trouble running 0.33? I had this running beautifully for a day then it died. Did a total uninstall of CIV4 including RegCleaner, and did a Reinstall with 1.61 patch and I can't get it to load the AI storage menu (the one that replaces Dawn of Man screen). I've placed the AIDATA##.txt everywhere and it doesn't work.



Question---you mean quite simply--don't add a single variable within a set, if you run duel sized map?

Does that mean run the same map type, or is Shuffle preferred. I've been doing a set with Duel, Normal speed, and Terra maps, with the default of 2 player slots. Will that skew even worse than normal for Duel-size, than if it was Shuffle? I notice that Terra tends to be short on real estate early in the game, and my one personal test game with an AI trounced it quite badly, although it did a good job of expanding in its corner.

EDIT 2: Hmm. A closer look shows that it scored 28544 on one duel map. I would suggest running larger map sizes for more representitive results - duel tends to skew quite violently. If you do want to play duel I suggest you keep the same settings for the whole set, and it'll then get normalized when going through the filter.

 

[Quitch]

So this bit the dust? That's a real shame

 

[Bringa]

Is this project dead?

 

[Samuelson]

The worst of my fears of this project have come true.

 

[Quitch]

On the upside there is a BetterAI project underway for Civ4.

 

[aegisx]

Too bad he never posted his scripts to do the analysis and generations. Someone else could have continued on while he was away.

 

[civ_king]

I wish someone continued this it would be awesome

 

[Ajidica]

Yay! the two year thread necromacy! I doubt this project would show much, they would be better off re-writing the whole AI, or going to work for Firaxis.

 

[Tholish]

What might be good is if the AI were highly accessible to amateur modding. It could be a game in itself where you script an AI and see how it does or how you do against it. There could be a whole forum for players to share AIs like there are for Unit Art and Civilizations and such. There could even be a play mode where you don't play yourself you can just run an AI, but you can mod the AI on the fly.

 

[Craterus22]

Yay! the two year thread necromacy! I doubt this project would show much, they would be better off re-writing the whole AI, or going to work for Firaxis.

but this project was making progress... its a shame the guy stopped.

 

[Tholish]

Another idea is if you wanted to do an experiment like Great Apple was doing, instead of having one AI and testing that, you could have muliple AIs per dll, one per leader (basically copies of every block concerned with AI behavior, but running only if they test positive for a particular leaderhead), and test them in big groups on autoplay. Maybe even have a separate program for generating source code for them by randomly varying certain things, whatever lends itself to random variability. Or even integrate the AI code generator into the game program and truly have random personalities generated at game start (randomness factor adjustable from 0 to some high number, with zero meaning no chance of non standard ai). Then if you wanted to you could have successfull random personalities automatically recorded in every game and set to recur occassionally (depending on how random the game was). And maybe even send themselves in to a database..

 

[krasny]

This is how Skynet got started.

Expect a visit from Sarah Connor.

 

[TheLadiesOgre]

If we got this project back on its feet again, couldn't we improve the BetterAI by using TGA's idea? Basically take the work from the BetterAI project, try to have it improve itself and winnow any weakness found therein?

 

[Ajidica]

Someone did make an AI mod that allowed multiple civs using different AI's to go against each other. Might be useful here.

 

[Tholish]

Do you have a link to it?