Google's AI program can teach itself video games from scratch

Google scientists have designed software that could do better than humans on dozens of Atari video games from the 1980s. Computers have already humbled human champions in Jeopardy! and chess. The exercise was aimed at making computers that could teach themselves to succeed at tasks, learning from scratch, trial and error, just like humans. The computer program, Deep Q-network, (DQN) did not start off with too many instructions but over time it did better than humans in 29 out of 49 games, and in certain cases, like video pinball, it did 26 times better, a new study released yesterday by the journal Nature revealed. It was the first instance of an artificial intelligence programme bridging different types of learning systems, according to study author Demis Hassabis of Google DeepMind in London. According to Hassabis, Deep Q could learn and adapt to unexpected things. Video game-playing AI is nothing new as anyone who had played against the computer knew, but in the absence of a real human opponent, most games allowed players to challenge the ''computer.'' In those games, however, the AI came with a series of specific rules that guided its behaviour. Deep Q, on the other hand, was given only one objective – maximise the score and from there, it ''watched'' the gameplay to learn new strategies in real time. Much like the human brain, it learned from experience. It looked trivial in the sense that these were games from the '80s and one could write solutions to these games quite easily, according to Demis Hassabis, who co-founded AI company DeepMind Technologies. Hassabis told BBC, what was not trivial was to have one single system that could learn from the pixels, as perceptual inputs, what to do. The same system could play 49 different games from the box without any pre-programming. He added one literally gave it a new game, a new screen and it figured out after a few hours of gameplay what to do. What was more impressive was DQN could take these strategies and apply them to games it had not played before. In other words, when DQN got better at one video game, it was actually getting better at a whole host of games. Source: ArticleReference-Image: