An Adaptive 'Rock, Scissors and Paper' Player Based on a Tapped Delay Neural Network

This paper presents an adaptive 'rock, scissors and paper' artificial player. The artificial player is based on an adaptive neural network algorithm. The hypothesis is that human players do not adopt the optimal playing strategy, i.e. to use random moves, and that the artificial player could exploit this and adopt a winning strategy. To test this hypothesis a WAP-based and a web-based version of the artificial player was made available to the general public. A total of about 3000 human players have played against the artificial player, to date. Several different training strategies are evaluated, and the results show that efficient algorithms can be constructed. The best result being 72% won games for the artificial player and 28% won by human players. The paper also identifies future interesting issues for both game developers as well as researchers within Human Computer Interaction.