In the past, artificial intelligence programs looked at how we played video games, and then tried to repeat after the person and also went through the games. So AI learned how to play go and coped well with different video games for the Atari console. Therefore, researchers at the American Institute of Technology of Georgia decided to go ahead and teach artificial intelligence to recreate the video games they saw on their own.
The team published the results of their work in the article “Game Engine Learning from Video”, where they described in detail the process of recreating the game with artificial intelligence, which was shown to him earlier. Interestingly, the AI could only watch the gameplay, but did not have access to the game code. So far, the program does not do a very good job: the games are buggy, but nevertheless it is still possible to complete them.
To teach the program to recreate the games seen earlier, the developers provided AI with a visual dictionary that contains all the game sprites, and also provided the program with a set of basic concepts about the position of all objects and their speed. So AI was able to navigate through the video, analyzing what he "sees." Watching the game from the side, artificial intelligence breaks down the gameplay frame by frame, marks what he sees with markers, and comparing the pictures with its documentation, recreates the game on its own.
While AI can work with 2D-platform. On his account several successfully recreated games, among which there are Super Mario Bros. and Megaman, but in the future, developers are planning to expand the capabilities of the program, having trained it to make 3D games on video.
The article is based on materials .
Comments
Post a Comment