Minimax AI within Non-Deterministic Environments

True artificial intelligence implementations in modern video games are a rarely seen occurrence. Most reliable sources say that the reason for this has to do with the unreliability in non-deterministic domains and poor performance. In this research, we used a well-known decision algorithm, minimax, in an attempt to focus on both of these issues while still providing emergent intelligence to the system. Minimax has existed since 1928, and has been proven sound in games of perfect information such as chess, or checkers. We refer to these games as deterministic domains in which every action has a known resulting outcome. We will look to extend minimax to non-deterministic domains through the use of expectiminimax. The result of this research was that the minimax and variant expectiminimax algorithm were both applicable in a commercial game domain, providing both acceptable performance and emergent intelligence in both deterministic and non-deterministic environments.