Classification of Humans and Bots in Two Typical Two-player Computer Games

Abstract

The aim of this paper is to explore whether or not the artificial intelligence (AI) embedded in traditional and typical computer games can be easily identified by players. We pay particular attention on AI simulation in simple games. Therefore, a 2D shooting game and a car racing game are developed by using different game AIs (game bots). Meanwhile, a Turing Test for computer game bots is implemented to test the abilities of above two computer games playing agents (bots) to simulate human game playing behavior. For the 2D shooting game three types of game AIs, random moving, finite state machine (FSM) and neural networks (NN), are used as agents to test. The results demonstrate that more humanized AI shows better human game playing behavior then others. For the race game a waypoints AI are developed. The experimental results show that although the paths design have a variety of changes, only 15% players be cheated. Careful review of the player's controls shows that players are more often fine-tune the location of the car to center it in the track. The research results can be seen as the based line of simple game AI development.