An AI that “solves” Super Mario Bros.

Picture: Nintendo, Flickr user labnol (CC)

Super Mario Bros. has been around for 28 years, and is an important part of not only gaming history but international popular culture, and has spawned untold bundles of merchandise, fan films, street art, and even ghost stories. The ways that people are engaging with the iconic sprites from our childhood are seemingly unending; they engineer numerous mods, remixes, and even path-finding algorithms that allow bots to play for our amusement (appropriately called ‘Infinite Mario‘).

Moving this last concept towards its ultimate end, computer scientist Tom Murphy has now designed a program that can “solve” NES games like other mathematical problems.

via  Nobel Intent (WIRED UK):

At SigBovik 2013, [Tom Murphy] presented a program that “solves” how to play Super Mario Bros., or any other NES game, like it’s just another kind of mathematical problem. And for those who know that SigBovik is an annual computer science conference dedicated to spoof research, hosted on April 1 every year, Murphy stresses that this is “100 percent real.”

He outlines his method in a paper, “The First Level of Super Mario Bros. is Easy with Lexicographic Orderings and Time Travel… after that it gets a little tricky,” but he also presented the results in this video:

It’s interesting to see where the AI fails and succeeds differently than humans. For Mario, the AI seems pretty efficient overall but dumb in places; getting stuck at a particularly long jump in World 1-3. Humans still have the advantage at certain problem-solving skills, as his programs fail miserably at Tetris.

Others contend each year at the Mario AI Championship, but this latest develepment goes far from simple pathfinding and automatically generated responses to computational problem-solving.

Read more from the article at Ars Technica.

, ,

21