Aritificial Intelligence (Photo : www.wired.com)
A new technology recently released by the Google-owned research company DeepMind features a software system that had taught itself to play 49 different video games. The company added that the new software named "deep Q-network" has the capability of defeating human professionals.
According to the company, the new software starts from scratch and through a series of trial and error it can develop its own algorithm of beating a certain game.
DeepMind co-developer Demis Hassabis explained that, "The only information they (the system) get is the pixels (on the screen) and the game score and the goal they've been told is to maximize the score."
Following this philosophy, the algorithm-based software starts off with zero experience and then builds experience from all the possible moves. The software randomly chooses any key until it reaches an algorithm that will enable it to start scoring.
The software was tested on many games including the 1970's classic video-game "Breakout," and "Ms Pac-mac." It was also tested on various simulation games like pinball, tennis, boxing and a racing game called "Enduro."
The company said that the system successfully beat professional players in some games but performed poorly in some including "Ms Pacman."
The interesting part however is that the software was able to discover some strategy that the game's creators hadn't even thought of. One given example is in the game "Breakout." The software managed to create a new strategy that entails creating a tunnel through the side of the wall and then send the ball behind it. This will enable the ball to break the wall from the back without ricocheting outside of the game area.
