AI has learned how to create video games just by watching people play
In the research document titled 'Game Engine Learning from Video
Do not know how the field of AI will develop in the future.
Usually when we teach an artificial intelligence (AI) about video games , the purpose is usually to let AI know how to play the game. That's how computers can defeat people in Go Go or Atari games. But a research team from the Georgia Institute of Technology is trying to develop another AI, which is to want artificial intelligence to learn how video games work.
In the research paper entitled 'Game Engine Learning from Video' , the team describes an AI system that recreates game engines (software written to design and develop video games) of games like Super Mario Bros. just by watching people play the game, this AI does not need to access the game's code, just watch the pixels and learn from it.The AI game engine created is quite a lot of errors, but still acceptable.
Classic Mario game.
This is the first time an AI system has learned how to make games , but there are still many shortcomings and limitations in this project. The AI system does not learn everything about games from scratch but is provided by two researchers with important information.
The first is a visual image library containing all sprite (a two-dimensional image or animation integrated into a larger scene) of the game. The second is the basic concepts, such as the position of objects and their speed, AI will rely on this to note what it sees. With the above information, the AI analyzes the gameplay in each frame, arranging what it recognizes and analyzes the game rules.
The AI system does not learn everything about games from scratch but is provided by researchers with important information.
'For each frame we will use analytical tools to review and collect events that take place. For example, when Mario is in a state, or the movement speed of the object in the frame, " Matthew Guzdial, the lead author of the study document shared with The Verge. On top of a Goomba mushroom and the next frame Goomba disappears, from which, the AI determines that the game rule is when Mario is on Goomba's head and his speed decreases and Goomba will disappear. "
Over a period of observation, the system will create basic rules and then record them as a logical sequence of 'if, then' commands and similar commands, combined, the AI will build the game engine. These rules can be exported and converted into programming languages to create games.
On the left is a normal game screen, on the right is an AI-generated game.
Currently, the system only works with 2D games because it requires human assistance to determine what's happening in the game. For 3D games with much more "terrible" data, this is time consuming as well as more advanced observation technology.
However, in the future, the research team thinks that technology like this can be used not only to research games but also to apply them to life. But from now until then, we have to wait for breakthroughs in AI so they can understand the world as people. It's the future, now, after Mario, the team is heading to the Mega Man game.
- The brain is destroyed by this popular video game
- Video: The reason many people like to feel scared when playing adventure games
- Action video games can increase vision by 20%
- Playing games for 30 minutes a day enhances memory
- TOP 10 worst games of 2007
- The more you play the game, the easier it is to get fat
- Research shows that video games are very useful for surgeons
- Foldit - play video games and contribute to medicine
- Play creative game
- Top most addictive game on Iphone
How can AI destroy humanity? Switzerland uses artificial intelligence to forecast lightning Network neurons diagnosed 216 rare genetic diseases Birth of artificial intelligence can read human thoughts The difference is all about everything between that day - now Genetic research through video games Is the computer just as smart as a 4 year old? Microsoft successfully integrated a micro-AI system into Rasberry Pi 3