Whenever people say that video games are a waste of time, I inevitably find myself thinking about stuff like the Mario AI Competition. The goal of the competition is simple: design an AI using one of many different methods that can successfully play through segments of Marks Persson's open source Java-based Infinite Mario Bros, a clone that generates infinite randomized Mario levels that grow progressively more difficult over time. Since there's no way for the program to anticipate level designs, it'll have to make decisions in an intelligent fashion.
The video you see above displays the winning program from the Mario AI Competition in action, playing through a randomly-generated Mario sequence that would probably kill most human players instantly. The program was written by Robin Baumgarten using the A* algorithm to help the program make decisions about which of many possible decisions it should make regarding how Mario should behave. The result is an AI that could probably complete any extant Mario game with ease, possibly better than many human players.
Add Us On