Computer science researcher Tom Murphy built an AI program that will play Super Mario and other Nintendo Entertainment System games on its own, reports Wired UK.
Murphy presented the program at SigBovik 2013, an annual computer science conference showcasing "spoof research." Despite the nature of the conference, Murphy stressed that his AI was "100 percent real," presenting a research paper titled "The First Level of Super Mario Bros. is Easy with Lexicographic Orderings and Time Travel . . . after that it gets a little tricky" alongside the AI's demo.
Murphy's program uses lexicographic ordering, a mathematical system used to order values. Murphy built two programs, LearnFun and Playfun. Murphy recorded himself playing the first level of Super Mario Bros. and fed all information stored in the NES memory, including the location of enemies and the buttons Murphy pressed, into LearnFun. PlayFun then took the information stored in LearnFun and played the game, its goal being to increase the value of Mario's score.
"[PlayFun is] trying to find the sequence of inputs to make those values go up in the RAM," Murphy explains in a demonstration video, posted above.
Murphy's AI works on other games including Bubble Bobble, Karate Kid, Hudson's Adventure Island and Pac-Man, but fails at games like Tetris because it attempts to achieve the highest score in the easiest way, which is laying blocks down randomly.
Read more about Murphy's AI in his paper, available through the Carnegie Mellon University website.
http://www.cs.cmu.edu/~tom7/mario/mario.pdf
: