An AI has recreated Pac-Man without ever actually playing Pac-Man, a feat akin to understanding, say, a mouse simply by watching it scamper through a maze.
The game is human playable and essentially recreates the entire Pac-Man experience, from the dots to the ghosts to the power pills. The ghosts move in the same patterns they moved in in the original game and – sometimes – little Pac-Man opens and closes his mouth.
The AI, called GameGAN, learned to play Pac-Man by watching 50,000 hours of recorded gameplay. From the release:
The GameGAN edition relies on neural networks, instead of a traditional game engine, to generate PAC-MAN?s environment. The AI keeps track of the virtual world, remembering what?s already been generated to maintain visual consistency from frame to frame.
No matter the game, the GAN can learn its rules simply by ingesting screen recordings and agent keystrokes from past gameplay. Game developers could use such a tool to automatically design new level layouts for existing games, using screenplay from the original levels as training data.
With data from BANDAI NAMCO Research, Kim and his collaborators at the NVIDIA AI Research Lab in Toronto used NVIDIA DGX systems to train the neural networks on the PAC-MAN episodes (a few million frames, in total) paired with data on the keystrokes of an AI agent playing the game.
Want to know how hard this is? As we noted before, this is akin to turning a camera on a simple system – a cockroach swarm or mouse – and being able to predict its next move based on previous experience. While Pac-Man isn’t a living system, it’s only a matter of time before this becomes commonplace.
To quote Neo: