We applied neural networks to a basic fighting game to see what we could learn about it. By having the AI play multiple games against itself the learning process can be accelerated.
Neural Networks used in the realm of artificial intelligence (AI). Vaguely inspired by how a brain works, they allow software to make decisions in a way that has been figured out by the software itself. While the conception of neural networks dates back to the early 50s, they are currently quite the hype as they’re becoming more and more useful as a consequence of the evolving processing power. After all, large neural networks run a lot of calculations when executing normally, and even more when being trained….
While neural networks are often mentioned together with AI, not all applications of AI are powered by neural networks, especially in games. Instead, most computer controlled adversaries in games will react in ways that have been entirely defined by the developer of the game. However, when using a neural network, instead of telling the computer controlled adversary or non-player character (NPC) what it has to do, we tell it what it can do and we reward it when it makes correct decisions.
As part of this case, we applied neural networks to a basic fighting game to see what we can learn about it. More on this will be shared an upcoming blog post in which I will discuss the process of creating the prototype. The process of training my NPCs and seeing them come to life was really fun, so be sure to look out for that one!