Tweet
Beating Civ 'by the book'
Using a "Monte Carlo" search framework, researchers have used a programmed algorithm runs a series of simulated FreeCiv games, in order to evaluate the possible utility of various moves. Through multiple iterations, the algorithm develops a sense for identifying the best move. Simulating evolution, the algorithm will sporadically insert a random move in order to sample alternative, possibly better possibilities.
In order to improve the game AI algorithm, the researchers did a full-text analysis on the game manual using a neural network, in order to establish if the algorithm could learn from general guidelines. By doing this, they were able to boost the original 17% winrate to an astounding 54%.
One could speculate if this technique might have some potential applications to further develop the AI in Civ5.
Read more at http://arstechnica.com/science/news/...on-by-rtfm.ars
In order to improve the game AI algorithm, the researchers did a full-text analysis on the game manual using a neural network, in order to establish if the algorithm could learn from general guidelines. By doing this, they were able to boost the original 17% winrate to an astounding 54%.
One could speculate if this technique might have some potential applications to further develop the AI in Civ5.
Read more at http://arstechnica.com/science/news/...on-by-rtfm.ars