A Neural Network Learns To Play Mario Kart

Seth Bling, a programmer who loves to play around with doing weird things to games, has trained a neural network to play Mario Kart and win gold trophies after watching fifteen hours of gameplay.

The MariFLOW system, which is what Bling calls the network, watches footage and tries to copy the player behavior it observes. In Bling’s case, he tried to get the system to match the way his father plays Mario Kart. The video explains that the program essentially tries to predict what buttons the player would press and employ those.

The program has so far won gold in the Mushroom Cup and silver and Flower cups. Despite this, it’s probably safe to say this is the least cheating an AI has been in Super Mario Kart.

Check out the video below and if you find yourself picking up strategies you might need to check if you’re a replicant.