Evolutionary artificial intelligence methods for games

Abstract

This work presents an approach on a research on the construction of neural networks with an AI (Artificial Intelligence), applied in games. The aim is to explain how a character can behave with an AI implemented in the game system. There are methods that can be used to evolve an AI in games, among them we can mention, HyperNEAT, there are also test environments. Previous research done through old games, showed that there were limited space actions, with the elapsed time, modern games have vast and diverse action choices, this is due to the player's ability to select several actions among a combinatorial space of hundreds of possibilities, thus granting rich sets of challenges. These challenges have not yet been overcome, so past research suggests that you are more likely to take a specific approach when trying to build a general-purpose AI model. The imitation algorithm with clustering seems to have a good solution for games that are not complex. HyperNEAT, however, appears to be better than the others, due to the probability of solving complex problems.