Software for artificial intelligence, Google has set itself the Space Invaders learning to play at a higher level than that of a human. In a game of Pac-Man was the artificial intelligence , however, lousy bad.
Google’s deep Q-network-software for artificial intelligence knew in 22 out of a total of 49 classic Atari 2600 games themselves better, to learn to play a human game-expert. The scientific journal Nature publishes about the research. When 20 games was the man better, and in the other 7 games was the battle undecided.
The software comes from DeepMind, a company that Google, in 2014, took over. The software looks at the last four frames of the gameplay to the success of actions to analyze, and new moves to determine, writes MIT’s Technology Review. That short attention limited the possibilities of more complex games to get the hang of. To the ghosts in the maze of Pac-Man to avoid would DeepMind more in advance should be able to see. At BreakOut were 600 pots of need to the optimal strategy and the game successful play.
Google is the software now so that the attention span and memory improve. In addition, deep Q-network, the possibility of more games to analyze in place of random actions for trial & error. The group is already working to make the software games of a Super Nintendo console and early pc titles to learn.