GalCiv III AI: The challenges

Published on Wednesday, June 24, 2015 By Brad Wardell In GalCiv III AI Talk

Unlike GalCiv II, the GalCiv III AI and overall system is extremely moddable.  I'm glad they did that but I'm having a hard time making the AI play the game as well as I'd like.

Whether people like it or not, skill at a game is the result of knowing effective, specific, strategies. You can't procedurally generate a strategy. Some of it needs to be hard-coded and that's where we're struggling because GalCiv III is all about data mining and procedural design.  It is much more like a Chess AI than what I've done in the past.

For instance, in Starcraft, I'm a mid-level diamond league player. I have very specific build orders, very specific map control strategies that rely on timing and a very tight control of resources.  In GalCiv III, presently, I can't even make a ship focus on a particular type of defense.  Heck, getting the AI to not decide to build endless constructors is a lot harder than you'd think.

Anyway, you're not here to read me complain about our own game.  So let's roll up our sleeves, we've got years to do this.  What I can say is that the AI system in GalCiv III is on a much better foundation than GalCiv II.  There's much more we can do. But it's going to take time and frankly, a lot of community help.  

What I need form you guys are saved games with an explanation of a strategic error you see the AI making.  I don't need help in coding what the AI should do because I can assure you, all those functions are there. It's simply calculating that doing something else is a better idea.

GalCiv II way:

If (transportisnearmylanet)then (do specific bad things to that opponent)

GalCiv III way:

Go through and weight zillions of possible things you might do, sort those weights and execute various empire-wide build/movement decisions.

Much of the work on GalCiv III will be making decisions more locally.