Editorial: Gaming / Facebook / Twitter / YouTube / Instagram / News / Discord

Today is the anniversary of one of the longest-running franchises in the video game industry: Pac-Man. Bandai Namco prepared various celebrations, including an alliance with Minecraft and a project by Amazon Game Studios that will allow players to create 2D levels. The Japanese company also collaborated with NVIDIA, which today showed artificial intelligence that can replicate titles without the need for a game engine, and the first experiment was Pac-Man.

NVIDIA announced today that thanks to its research division it managed to emulate Pac-Man, but now without the intervention of a human and without using a game engine; that is, by means of pure artificial intelligence (AI). This was possible thanks to NVIDIA GameGAN technology, an artificial intelligence model consisting of “a powerful neural network model” that was able to replicate the classic after training with “50,000 game episodes.”

This means that artificial intelligence had to replicate the gameplay (the way the character moves, the appearance of the ghosts and Pac-Man) and learn when to run away from the ghosts and when to go for them (and other mechanics, such as the teleportation of the portals, or the impossibility of crossing the walls) only after training with 50,000 episodes of the game, which is equivalent to “a few million frames”.

NVIDIA technology replicates game content in real time

According to the information, GameGAN is the first model in its class that can emulate the operation of a game engine with the use of antagonistic generative networks (GAN, for its acronym in English), this means that artificial intelligence was able to do it without knowing the basics of the game. These types of models work with 2 networks, one generator and one discriminator, which together can create new content that is so compelling that it can pose as the original game. Aside from the fact that intelligence can play the simulation, which is a fully functional version of the title, a human can also play it.

As a result of training thanks to artificial agents, GameGAN reacted to their actions and generated game content in real time to form multiple levels or versions.

“This is the first research to emulate a game engine using GAN-based neural networks. We wanted to see if AI could learn the rules of an environment just by looking at an agent’s game screen moving through the game. And he did it, ”said Seung-Wook Kim, NVIDIA researcher and project leader.

NVIDIA GameGAN evidences that AI will have a great impact on video games

GameGAN shows that it is possible that artificial intelligence can understand the rules of any game and that new content is generated from that knowledge, opening the possibility that “creating simulators for tasks like these may one day be replaced with neural network training “

“We were shocked when we saw the results, incredulous that an AI could recreate the Pac-Man experience without some game engine. This research presents exciting possibilities to help game developers accelerate the creative process of developing new level designs, characters, and even games, ”said Koichiro Tsutsumi, member of Bandai Namco Research.

We leave you with some videos in which you can see the NVIDIA technology running.

What do you think of this result? Are you interested in the potential of neural networks in video games? Tell us in the comments.

Bandai Namco also prepared other ways to commemorate Pac-Man’s 40th anniversary. A few days ago, the company invited its followers to share their fond memories with the iconic pet over the years. You can find more news related to Pac-Man by checking this page. You can’t miss the recent retro Friday, focused on this character.

Stay informed with us, at LEVEL UP.