Perception of Realism of AI in VR Games

My Bachelor project was mainly about artificial intelligence (AI) in games. Specifically, I investigated how people perceive characters in a VR game, from the point of view of realism of their behaviour. A kind of limited Turing test.

In truth, I wanted to play around with game AI, and make stuff. I spent most of my time learning different AI methods, from simple state machines, behaviour trees, expert systems, to utility AI, neural networks and machine learning, and other algorithms.

One of my favorite things I learned was to question what is intelligence. Based on an interesting paper by Steven Omohundro called "The Nature of Self Improving AI", I learned that intelligence is defined through ignorance, meaning that when we don't know how something works, and it does what we want or what makes sense to us, we call it intelligent, otherwise it means we know exactly how it works at which point it's just a reactive system. A lot of philosophical debates later, I reached a point where I questioned my own intelligence and revealed to myself my own ignorance. But that's besides the point.

Long story short (for now), I made a game to try and put into practice what I had learned.

Viking Insanity was the end result, where the player is a viking with an axe and he is fighting other vikings that looked like in the picture. The VR version places the camera at the eye level of the character and uses the animations for locomotion.

There were many issues with the game and with the experiment, but I explored how it feels to play with a bunch of AIs and a real player, and try to distinguish the person from the AI-controlled character.

I earned a 12 for my project, which in Denmark means the top grade.

The game itself is not so fun to play outside the experiment, but it was fun making it and I still use lessons I learned from that project in my current work.