The AI Complaint - Page 2

Parasite · #11 May 13th, 2005, 07:07 PM

My AI complaint is in mainly in the tactical combat. Things happen that about anyone would call "not smart". Just last turn in a game of mine I had a missile ship that could not attack sats. It just wandered around the board in a random way. Sure enough, it wandered in range of the sats and was shot at. There are many cases where you could give damage and not take any, yet you get destroyed. these could be changed (I hope) with a few simple rules. Add a strategy called "Suicide" to allow them to be ignored at need and there you go. Also there is a simulater. It would take some processing time, but a battle could be fought several turns ahead in it, then the best movements to make based on these be chosen. A novice fleet or ship could not look ahead this way, then each level of experience could look one or more turns ahead in the simulator to enable better planning.

narf poit chez BOOM · #12 May 13th, 2005, 07:54 PM

What you want is called a 'Learning AI'. Just commenting.

I made a pretty pathetic one that plays tic-tac-toe against itself once.

Kamog · #13 May 14th, 2005, 10:10 PM

A learning A.I. would be wonderful. If the A.I. could change its ship designs to counter the design of enemy ships, decide what facilities to build based on a long-term plan and make changes in its plan as the situation evolves, and learn from past mistakes that would be so absolutely fantastically excellent.

Kana · #14 May 15th, 2005, 03:22 AM

Quote:

narf poit chez BOOM said:
What you want is called a 'Learning AI'. Just commenting.

I'm not sure...but I thought that Galactic Civilizations had as some form of Learning AI. It would take into consideration your strategy, and research and diplomacy and such, and do x to your y. Gal Civ was ok...it was hurt by single player only, and the ship construction was weak and boring. And research had a very complex tree....

Kana

Renegade 13 · #15 May 15th, 2005, 03:25 PM

GalCiv AI didn't seem to "learn" when I played against it. It never seemed to learn that my strategies always worked, and it's always failed.

Fyron · #16 May 16th, 2005, 02:49 AM

The learning is primarily when people submit games to the "metaverse" and the game developers analyze what good players did, then update the AI to copy good strategies.

Parasite · #17 May 16th, 2005, 11:36 AM

I don't really want a learning AI (i.e. one that will learn on the fly and get better as it plays more) I just think some form of decision tree would be good. For instance, if there is a stack of 100 DUC 1 sats that can shoot at range one(typical for an AI empire) and you can kill them at range two, don't close to range 1 despite your orders to close. The ship experience could just let more of the tree be used. A novice ship could close and say... "Those are what sats look like!" as they die. A ship with just a little ex could just not go there and continue to bombard the planet from the other side.

I believe a lot of range juggling could be hard coded in. Close to his movement speed plus his weapon range. If he closes, then close with him and get the first best shot. Simple things like that. Especially for single ships of small numbers of ships. In large battles these effects become less powerful.

Sivran · #18 May 18th, 2005, 04:58 AM

A good first improvement would be to not use all available combat movement points, if they are not all needed to get within range.