This blog’s focus is on the presentation of my topic. The text below somewhat resembles what I plan to say.
Welcome to my presentation about artificial intelligence. When you hear about AI, you might think about destructive robots from certain movies. Or you might think of self-driving cars or other innovative projects.
However, AI is nothing more than a tool – a powerful tool that can be applied to many different fields. Every modern device supports some form of voice recognition. Chatbots are getting more and more popular.
The topic of my project covers a small field of AI – I focus on bots that learn to play games. I found it very interesting when the first AI mastered chess on a level even the best cannot compete with. It is even better when it comes to video games. I wondered what is about and more so: how it works.
There are several approaches that could be implemented here. I chose reinforcement learning. This method is all about learning by trial and error. The top left image about flappy bird demonstrates it quite nicely. The AI gives a random input – and slowly figures what works and what does not. Let’s take this concept and apply it to a classic – Super Mario World.
SethBling created MarI/O, which puts reinforcement learning to a test. The AI is programmed to press random buttons and analyze the results. The goal is that it figures out which buttons to press in any given scenario – which really just is playing Mario.
I decided to download and install the project on my own PC and gave it a shot. On the left side you can see some background configurations and supervising consoles while, on the right side, there is the actual game. I can definitely say that the AI was learning. However, most of the time it would decide to run into the same wall over and over.
Back to the roots. I started the project by doing research on the topic. The book “Reinforcement Learning” was a great start since it covers the most essential details. It is a bit outdated, however. Therefore, I also went through some MIT press journals and YouTube videos.
The best find of my research is the Lua scripting language. MarI/O was developed in this language. It is open source. It runs on every operating system. It is fast. Most importantly – it was ranked as the best scripting language for games by experts.
In the future I want to take closer on another language – Python. Furthermore, I want to look on TensorFlow, which is the go-to AI plugin for Python. With my acquired knowledge I want to develop my own AI that will learn to play another classic: Metroid.
I mentioned self-driving cars in the beginning of this presentation. Well, they already exist: in GTA 5. Hobby programmers tried to implement self-driving cars into the game like they would work in real life – and were successful. This is a good example because it shows what can be done with AI in this field. I hope this was interesting to you – thanks for listening.
