DeepMind – program learns through artificial intelligence

The British company DeepMind, which was acquired by Google last year for 400 million pounds, developed for the first time a software program with artificial intelligence, which learns on its own from scratch to play well‑known video games – and often does it better than humans. The program named Deep‑Q‑Network, inspired by the functioning of the human brain, has so far learned to play 49 classic Atari games. In more than half, it is equally good, if not better, than an experienced human player. The researchers, led by the British‑Cypriot founder and vice‑president of Deep Mind, former neuroscientist Demis Hassabis (and with the participation of Greek‑origin Γιάννη Αντώνογλου), who published the related article in the journal «Nature», according to the BBC, «Science» and «New Scientist», stated that for the first time a self‑learning system (software program) managed to learn something as difficult as a video game.

The supercomputers Deep Blue and Watson from IBM have achieved impressive feats, but these large artificial intelligence systems were largely pre-programmed with the game's instructions and appropriate strategies. In contrast, the «light» system of Google Deep Mind started with basic information, doing the rest on its own as it progressed, thanks to its advanced algorithms.

It took about two weeks of training on the new system for it to learn, through continuous testing and errors, the “secrets” of each of the 49 games. Moreover, in some cases, it discovered new “tricks” that players had never encountered before.

Thus, he learned on his own to play games such as Space Invaders, Pong, Enduro, Video Pinball, Boxing, Breakout, Pac-Man, Private-Eye, etc. Although these video games are old, from the ’80s decade, it is considered an achievement – a milestone that the system managed to learn on its own to successfully play so many games, without special programming beforehand.

This achievement in the field of the so‑called «deep» machine learning opens the way for the creation of truly more «intelligent» computers and machines in general. For example, autonomous vehicles without a driver, robots that will understand their environment or devices that will fully comprehend human speech, so that they translate correctly in real time.

There are certainly those who fear that artificial intelligence will ultimately become dangerous for humanity. Among other things, cosmologist Stephen Hawking has warned that the development of a fully developed artificial intelligence “could mean the end of the human race». Demis Hassabis, however, appeared reassuring, saying that “we are still decades away from any technology that should worry us». But then?

Source: Newsbeast.gr