Brad Wardell's views about technology, politics, religion, world affairs, and all sorts of politically incorrect topics.
New philosophies
Published on December 26, 2005 By Draginol In GalCiv Journals

"So what did you do over your Christmas break?"

"Rewrote several key modules of the artificial intelligence engine in Galactic Civilizations..."

Galactic Civilizations has always been known for its strong artificial intelligence.  Since the game is designed to be a single player game, it bloody better have good computer players. 

Most people reading this don't know me very well but I'm actually a multiplayer fanatic.  I was one of the top ranked Starcraft players on Battle.net during its beta and was the top player on the "Free For All" part of Cavedog's Boneyards service (Total Annihilation multiplayer).  In short, I like to play multiplayer.

But I only play multiplayer because most games do not have very good single player experiences for strategy.  You quickly find their weaknesses and predictability and end up having to go on-line to get a challenge.  Turn-based games are too eradically paced for most people (certainly me) to get into playing them multiplayer. 

So when it comes to playing Galactic Civilizations II, I want the computer players to be able to play intelligently.  That means that the difficulty level shouldn't just be how much they cheat or how much free gold they get or whatever.  The difficulty level should determine HOW they play. 

In Galactic Civilizations I, the AI and the human played by the same rules up until Genius level at which point we cranked up the abilities of the computer players (gave them free money and such).  Playing by the same rules is much more CPU intensive and putting together a good strategy is also time intensive as well both in terms of CPU power and coding time. 

One of the most CPU intensive things in GalCiv AI used to be scanning through planets and exploring things.  So to avoid forcing the AI to do that, we made it so that the computer players "knew" where the good planets were. We made it part of the plot -- the aliens had long since explored the galaxy and the humans could not.

But in GalCiv II, you don't necessarily play as the humans.  So to be true to the story, the humans, when controlled by the AI, have to actually explore out the galaxy with scouts and what not.  In the process of doing this, we discovered that it doesn't make as big of a difference as we thought to the effectiveness of the computer player.  Moreover, on a modern machine (i.e. Pentium IV processor) it's not a major factor either.

Because in GalCiv I the AI got to know where the good planets were, some people would say "Well sure the AI in GalCiv is lethal -- it gets to know where the good planets are!".  So we've decided to eliminate this.  The AI colony ships no longer know where the good planets are. They have to go out and explore the galaxy just like a human player would. 

The ultimate goal is to make the AI in GalCiv II play virtually undistinguishable from a human player. 


Comments
on Dec 27, 2005
Ahhh... I look forward to facing off repeatedly against the final product.
on Dec 27, 2005
So, the AI should just be idling the CPU on an Athlon64, right?
on Dec 27, 2005
So, on an Athlon64, the AI should really take no juice at all.... Since of course it's more powerful than a PIV... of course, for the PIV folks, it would be nice if the game was multithreaded efficiently (it may be, I dunno)...
on Dec 27, 2005
It probably depends on whether or not Stardock has 64-bit optimized code sections for the AI... I'm guessing not;)
on Dec 28, 2005
Sweet! Thanks for keeping us in the loop, Draginol. I'm not ashamed to admit I'm rather glad to hear the AI players will now have to go out and find the good planets just like the human player. I was prepared to accept the original proclamation that they already knew where they were, but I really hate it when the AI has artificial bonuses--not that that's unusual for a TBS game, I know. In any case, I'm glad the exploration AI (for lack of a better term) is competent enough that you're comfortable in getting rid of its "colonizing omniscience".