Neural Networks and Video Games: Developing an Interactive AI
Lynn Ziegler, Computer Science
For years, video games have been a field of innovation and progress, but mainly in the realms of graphics and control. Game logic is a relatively overlooked field of game design. My research focuses on creating an adaptive type of game logic using neural networks. Using unique strategies, neural networks were trained to control games as a human player might. The purpose of this experiment was to verify that it is possible to control games using a neural network and to measure how well it could control them. This offers new possibilities for programming adaptive adversaries for human players to face off against. The ultimate purpose of such adversaries would be to keep the game fresh and exciting, eliminating predictability. In my final experiment, the neural network's task was to control the flow of the game instead of playing the game. A human player plays the game while the neural network directs which types of enemies the player will fight. It's able to learn which enemy types are performing poorly against the player as well as which are performing well. My thesis explores this as a possible way to achieve an adaptive gaming experience.
Tice, William, "Neural Networks and Video Games: Developing an Interactive AI" (2011). Honors Theses. 112.
This document is currently not available here.