Developed by NVIDIA, GameGAN manages to recreate the classic Pac-Man without using a game engine
When an old video game celebrates its birthday, an increasingly common type of homage paid by programmers to teach an AI how to play it. But, for the celebration of 40 years of PAC MAN, a NVIDIA took this kind of homage to another level, and not only taught a IA playing the game, but developed one that was able to create its own version of the game without the aid of any game engine (game engine).
Launched in May 1980, the first PAC MAN it was a huge success for arcades and one of the main responsible for the popularization of video games. Throughout its 40 years of history, PAC MAN it ceased to be just a game and became one of the most famous and successful franchises in the history of video games, selling about 48 million copies distributed across different platforms and raising revenue of about $ 14.10 billion.
And to celebrate the 40 years of success of the game that NVIDIA developed artificial intelligence GameGAN, a model of IA that manages to develop a fully playable version of PAC MAN just watching videos of people playing the game.
What PAC-MAN's GameGAN
THE GameGAN an IA created by NVIDIA which was developed from adversarial generative networks (generative adversarial networks, in English). Also known by the acronym GANThis is a machine learning method that uses two competing neural networks, being a generator and a discriminator, to analyze a content and create a new one that is practically identical to the original.
Created in 2014 by Ian Goodfellow (a researcher in the field of AI who currently works for the Apple), at IAs GAN-type devices were developed to show that computer programs can learn to copy something even if they are unaware of the gears behind its operation, more or less like the human brain, which can make us imitate something even without understanding all of it. basis behind what, for example, copying a dance choreography by a pop artist even though we have no knowledge of music theory or even have taken dance lessons.
Thus, a team of researchers from NVIDIA led by Seung-Wook Kim caused the AI to observe the millions of possible frames (that is, each movement that the player could make in the game) of the phases of the PAC MAN original, in addition to the controls used by another IA who had been trained to play the game, and then checked whether the GameGAN I would be able to create a game just like the original just from that experience.
The result was better than expected: the GameGAN not only managed to recreate exactly the entire game environment (the maze full of balls with the PAC MAN, four larger balls in the corners and four ghosts), but also understand practically all of his logic, from the simplest ones (like the fact that the character does not go through walls and needs to eat all the balls on the map to pass the level) until more complex (like the fact that ghosts must always chase the player, but if the protagonist eats one of the larger balls, these ghosts should all turn blue and start running for a certain period of time).
The future of the gaming industry
Although it was not the first IA GAN type to be developed, the GameGAN gives NVIDIA it was the first created for the specific use in video games, and the result obtained can be used as a basis for a change throughout the industry.
This is because, if new experiments with GameGAN achieve results as promising as those obtained in the game PAC MAN when using the IA in more complex games, it can become a powerful tool for game developers, helping them to create new scenarios and characters much faster, decreasing the time needed to develop a video game.
And the same technology can be used for other functions as well. According Sanja Fidler, director of the research lab at NVIDIA in Toronto, the GameGAN can be the first step towards the development of a IA who could learn how to drive a car just by watching videos of people doing this activity and how they behave in different traffic situations, which could be a new approach to the development of autonomous vehicles.
For now, the NVIDIA is only publishing the results obtained by the research, but has already promised that by the end of the year it will release a fully playable version of PAC MAN developed by GameGAN at the IA Playground, the platform where company researchers make free of charge some programs developed in their laboratories that use artificial intelligence.