“Mario Lives!”, winner of the AI Videos Competition People’s Choice Award, has been causing a bit of controversy since it was posted to YouTube on January 15th. The offhanded manner the video, and many of its users, describe Mario as “alive” or “self aware” have rubbed a lot of people the wrong way, but that might not be the most important thing about the video. It’s biggest impact may be demonstrating the way a game world can make a wonderful platform for developing and demonstrating new ideas in artificial intelligence.
It occurs to me that the most compelling thing about the video isn’t what the Mario agent does, how it was accomplished, or how the researchers explain the system. The true impact comes from how easy it is to understand the rules they’ve set up and relate to the way the character is made to act.
The “Mario AI” project was created by researchers at the University of Tübingen, Germany. I’ve not found any detail on why they chose Mario as their subject, but one of the main results is obvious – it made the project more approachable. Pretty much everyone recognizes Mario and the virtual world he inhabits. Many of the people commenting on the video have played one of the dozens of Super Mario games. After all, Wikipedia reports the Mario franchise as the best selling games of all time with more than 400 million copies sold. (Wikipedia contributors 2015)
The motivations and mechanics chosen for Mario’s mind, things like craving coins to assuage hunger and jumping on objects in the game world to discover how they work, made intuitive sense to these viewers. They were likely simpler for the researchers to choose in the first place. By leveraging a familiar game world as the platform for these creations, AI developers don’t have to create their own rules before getting into the important part of their work and observers don’t have to take the time to learn these rules before determining if the AI agent is behaving intelligently.
This ease in understanding both goals and results displayed by the video may have contributed to its victory. The video didn’t have the beauty of swarming robots and wasn’t the most advanced display of modern AI techniques, but it had a charm all its own. The highly visible “reasoning” process undertaken by the Mario agent was refreshingly different than the generally hidden enemy AI and more approachable than the environment and rules created for robot demonstrations. Of course, the general popularity of Mario Brothers games and a the good bit of media coverage it generated must have been significant factors as well!
As for the question of AI Mario’s self awareness, io9.com has a brief, and fairly critical article on the subject. (Dvorsky 2015) Readers should be wary of finding any certainty from such criticisms, as they are hollow without reference to some definition of consciousness that takes into account both the internal states described as required
Dvorsky, George. 2015. “No, This Artificially Intelligent Super Mario Is Not Self-Aware.” io9. Accessed January 25. http://io9.com/no-this-artificially-intelligent-super-mario-is-not-se-1680630827.
Wikipedia contributors. 2015. “Mario (franchise).” Wikipedia, The Free Encyclopedia. http://en.wikipedia.org/w/index.php?title=Mario_(franchise)&oldid=644587931.