I just had a thought, what if some sort of genetic algorithm was implementet into a Internet-game... for example if there are bots in online games or even better in gameplay models like "overrun" (team of players against attacking CPU controlled enemys). While genetic algorithms are to "slow" to improve much on a single computer while playing against bots, wouldn''t it work if after every internet game the data was submitted to a main-server? This main server would choose the best AI according to the game-results and resubmit if a new internet game is started. A popular game like UT2003 or EnemyTerritory (of course it wouldn''t work with those games, but as an example) would create millions of "generations" in quite a short time!

One problem with that is there is too diverse a player base. What I mean is that GAs will tend to fine tune themselves to specific sets of criteria such as playing styles. With enough people in the mix, there would be a very broad range of play styles and the GA would tend to flatten out to become mediocre against all of them. This, of course, is highly game-rule dependant.

A possible solution, however, is to develop an algorithm that fine tunes different result GAs for varying play styles. That is, you could end up with multiple GA results that are tuned to specific subsets of styles of play. Then, through some observation and diagnostic criteria, the computer player could determine and choose, or even continously change from one "GA AI style" to another based on the type of player they are facing during that session.

Dave Mark - President and Lead Designer
Intrinsic Algorithm - "Reducing the world to mathematical equations!"

Evolutionary techniques have already been applied in this area. See:

http://dendrite.cs.brandeis.edu/papers/tronsab98.html

"With enough people in the mix, there would be a very broad range of play styles and the GA would tend to flatten out to become mediocre against all of them"

That isn''t necessarily true. If there is a playing style in the solution space capable of beating all other styles of play, then in theory the GA will find it (if correctly designed)

My Website: ai-junkie.com | My Book: AI Techniques for Game Programming

quote:

---

That isn''t necessarily true. If there is a playing style in the solution space capable of beating all other styles of play, then in theory the GA will find it (if correctly designed)

---

if the GA finds a unique solution to beat all the others, you''re better of redesigning the whole game :D