Gaming your way through a PhD: 'Studying games is not just cool, it's important'
Studying video games for your PhD research. It might sound unlikely, yet it is exactly what PhD candidates Giulio Barbero and Matthias Müller-Brockhausen did. Barbero has obtained his PhD already and Müller-Brockhausen will defend on November 5.
Did you know that there is a Game Research Lab at the Leiden Institute of Advanced Computer Science (LIACS)? 'Something we might look at are algorithms that build something in a game. Like a house in Minecraft,' illustrates Müller-Brockhausen the work they do in the group. 'That algorithm could also be useful in other applications.'
Improving games and AI at the same time
For one of his projects, Müller-Brockhausen wanted to understand how AI can play the game Tetris Link. ‘I compared different algorithms to find out what the best way is to approach the game,’ he explains. ‘This helps us evaluate and improve AI algorithms.’
Procedural content generation (PCG) is a computing method where algorithms create something in a game. Like the Minecraft house mentioned by Müller-Brockhausen. This can be done manually by a person, but these algorithms are faster and provide slightly different results each time, making it versatile.
Reinforcement learning (RL) is a type of machine learning. Such an algorithm learns to perform a task autonomously by trial and error, solely guided by a reward function that’s defined by a human.
He then continued by combining two techniques (see box). ‘RL has the promise to give us novel, often unexpected outcomes that PCG wouldn't have come up with on its own,' he says. 'You can compare it to ChatGPT. That can only remix whatever input it has learned from. But RL could invent entirely new things. I use this for example to generate chatter for a character in a game.'
AI inspired by human behaviour
Barbero’s work is complementary. ‘I look at how people play games and then use that to improve AI. We actually have an office Playstation that contributed to a scientific paper! We had people play a game and talk out loud what they were thinking and doing. We analysed their approach to figure out which steps are difficult for the AI.’
He also looks at AI in education. 'Students use AI all the time. It is important that we learn how to use it well and understand potential risks,’ Barbero explains. ‘For example, I look at restricting the info an AI gives a student, so they don't get the entire solution handed to them on a plate.'
‘If you engage students with a topic in an exciting way, they are so much more motivated.’
Juggling research and teaching
Both PhD candidates teach their own course next to their research. ‘It is a lot of work to manage everything in the first years, but now I find it very fulfilling,’ says Müller-Brockhausen. ‘Students enjoy the topic and really go above and beyond.’
Besides teaching, he enjoyed the independence they got in their PhD: ‘We had a lot of freedom. Our supervisor, Mike Preuss, gave us a topic and then we could follow our own path.’
A research topic that is cool ánd worthwhile
Studying games is very worthwhile according to Barbero. ‘I always loved gaming, but my parents didn't,’ he laughs. ‘Their famous last words were that I would never achieve anything in life with games. But I found a way.’ He adds on a more serious note: ‘Everyone thinks our work is cool. But I realised it's not just cool, it's important. If you engage students with a topic in an exciting way, they are so much more motivated.’