Evaluation of Using Neural Networks on Variety of Agents and Playability of Games

Non-player characters which benefit from artificial intelligence play a significant role in video games. Due to the vast state spaces of some video games these agents must have access to the hidden information of the game's environment. As a result, these agents become invincible. To solve this problem, programmers of commercial games have to manipulate the data available to these agents and deteriorate their performance by using factors such as field of view, distance, time, and so on in order to make the game playable. But often this procedure greatly affects the playing experience and tends to create non-player characters which have non-human behavior which will cause the discontentment of users.Another problem with conventional methods is the lack of diversity in the behavior of agents. This limitation makes the agents predictable. To overcome this problem, different programming approaches are required. Creating different agent functions can be time consuming and therefore not feasible when a wide range of behavior is required.In this paper a new method is introduced to create agents by using neural network. These agents are trained via faultless agents and are aimed to create different levels of difficulty and in the meantime provide diversity in their behavior. By using neural network, the process of learning can be simulated, and therefore agents can present different player archetypes and provide new challenges for the user every time.