Playing to Lose: AI and "Civilization" (Soren Johnson)

[theskibum]

I was reading the thread and thinking to myself " how much posts will take until someone compares civ to chess?"

Said that, read above.... and add that chess terrain is isotopical and that the number of units does not increase with time ever. I remember doing a small back of envelope calc about this ... found it . The complexity of Civ IV is atleast a dozen times of order of magnitude higher that chess in t0 ( things get worse as turns go by in general )...

@Ignorant Teacher

The AI never had and does not have a strategy to win by space ( besides Better AI mod ). It simply techs normally and picks the SS parts to build more or less by luck. Worse, it is actually coded to not launch a SS if there is one in the air that can get there faster ( like if it was impossible that certain earth-bound event make the SS derail ) ...

@Psyringe

I've seen that line of argument somewhere else not long ago by the mouth of Sid What is wrong with the people that make games nowadays? Do they think they are the only enlightenend ones , a special breed that can understand alone and by themselfes odds and fairness while all the others don't ? Bah ... In the end this will only create a circular reasoning: if you reward stupid behaviour by making a unfair system, expect people behaving stupidly ( it is rational to be stupid when the system rewards stupidity ). That reinforces the designers perception that people are stupid, so they make ... ad aeternum

The kickassest post in this whole thread. Great link in there by the way, another entertaining read

 

[For Nall]

Well at least the Civ5 AI plays to win and judging by Gregs 2 hour gameplay it looks like it does a good job of it.

 

[r_rolo1]

Civ IV AI also looked good when it got out ... and it was ,atleast compared with Civ III one But give time and you will start to see the inevitable issues it surely has

 

[AutomatedTeller]

The problem with ANY AI is that it's a small group of people who program it, and there is an ENORMOUS group of people playing the game, analyzing, finding weaknesses.

The AI cannot learn from game to game, while humans can.

If you played against the game just yourself, without looking at the code, the AI would be a lot better....

 

[Musketeer935]

The AI cannot learn from game to game, while humans can..

That's debatable. In theory, you could make an AI that learns from game to game -- when it is successful at something, it analyzes why it did well, and saves those parameters that made it do something good in a file in the games files. It does the same thing when it does something bad.

Say, when it fails to take a city, it looks and realizes that the Swordsman it used to attack the Hill City w/ Longbows was not a good idea, seeing how much higher the units total strength goes. The game may then raise the parameter of how close the computer would like to be to the enemy unit's strength with its attacker before attacking again. It may also realize that it did slightly damage it, and therefore if it used many Swordsmen, it may win eventually.

Also, if it were me, I would make each AI learn independently of the others. Shaka will not use the knowledge that Washington has learned, or at least, not all of it. Perhaps some of the things would 'diffuse' to other AI, but not everything.

This would be extremely difficult to make it recognize WHY it did well or poorly, but, in theory, it would be possible.

 

[Dr.Null]

Also, if it were me, I would make each AI learn independently of the others. Shaka will not use the knowledge that Washington has learned, or at least, not all of it. Perhaps some of the things would 'diffuse' to other AI, but not everything.

Please get to work on this immediately!

 

[Musketeer935]

Please get to work on this immediately!

Sadly, I said in theory, and at the moment am not really sure how to do this with civilization. Perhaps some of the much more advanced modders would be able to do it; I have little experience with CivIV's game files.

I also lack the actual gaming skill required to code it, either -- you have to know how to play pretty well to be able to make it see WHY it does well/poorly.

Perhaps I may look into it though, since you've sparked my interest.