There are gamers who can complete Super Mario Bros in less than five minutes, and gamers who still struggle with the finer elements of mushrooms, pipes and coins. But how would Mario fare playing his own games?
Thanks to a team of cognitive modelling researchers at the University of Tübingen in Germany, we may soon find out. Their latest project: “An Adaptive Learning AI Approach for Generating a Living and Conversing Mario Agent.”
The team have entered the annual video competition run by the Association for the Advancement of Artificial Intelligence (AAAI), publishing a YouTube video explaining their work in the hopes of winning its People’s Choice award through public votes.
The project is aiming to create a Mario who “gets to know his own world” by becoming “aware of himself and his environment”, enabling him to make his own decisions based on his curiosity, hunger, happiness and fear, as well as through voice commands delivered by a researcher.
The video shows Mario being taught that if he jumps on a Goomba then it “certainly dies”, with his knowledge converted into tags, and then natural speech. It also shows how his inner emotional state can be used to govern his behaviour.
“Mario will collect coins if he is hungry, whereas when he is curious, he will explore his environment and autonomously gather knowledge about items he does not know much about yet,” explains the narrator.
The Mario agent may not be ready for a five-minute Super Mario Bros run just yet, and it’s quite jarring hearing him speak with computer-speech tones rather than the familiar sound of Charles Martinet, who voices him in games.
Even so, Mario’s popularity should make him a strong contender for the award, even though rivals include self-organised robot swarms, a robot that children can teach to write, and technology to help self-driving cars negotiate crowds of pedestrians.
