Making Decisions Like Human, the Application of Deep Learning Algorithms in Game Testing

Abstract

Artificial intelligence (A.I.) is currently used widely in testing of video games. A.I. had surprisingly performance in board games such as chess and go. How does it perform in modern digital games? Currently, A.I. is at trials of checking the operation of the game and figured out bugs by simulations. However, the algorithms that were applied to the first generation of A.I cannot perform well in current digital games due to completely different constraints. Hence, Deep Learning algorithms are implemented on artificial intelligences or agents that are created for the digital game testing. Deep learning algorithms not only let A.I. can simulate how a human player is thinking when they are playing a digital game, but they also let A.I. can gain experience from billions of simulations based on the extreme speed of calculating. This study would give analysis on performance of A.I. or agents with deep learning algorithms, and show the limitations and barriers of automation of video game testing. Eventually, we explained why automated video game testing is still far away from us.