Can Your Computer Beat Your High Score?

You might have thought it was embarrassing when your little sister bested your high score at the classic vertical arcade games down at the local arcade, but did you know in today’s world, you have another competitive threat? Researchers have been perfecting artificial intelligence in computers for decades and computer scientist and former video game developer Demis Hassabis’ recent presentation shows how far computers have come.

Computers Can “Think”

While computers do not have social skills and you’d never mistake a program for a person, Hassabis’ development company, DeepMind, have perfected software which allows computers to “think” analytically. They can research a puzzle on their own and come up with solutions and data. To accomplish this, DeepMind designed a computer learning program called the “Deep-Q-Network.” There are two elements in this program – the deep neural network and Q-learning.

The deep neural network is learning method based on perception. With this skill, the computer can view the game just as a human player would. Q-learning allows the computer to keep track of when it makes correct decision by keeping track of points and outcomes, storing information allowing it to remember the data when presented with the problem again the next time. Combining both skills gives computers a definite edge over the human brain when it comes to classic arcade games, as DeepMind found out through testing.

Classic Arcade Game Scores

Over the course of two weeks, DeepMind allowed the program to work from one desktop computer on 49 different classic Atari games. In past tests, computers have not been able to overcome basic video game levels because of their inability to apply the principles to more complicated, extended problems. This time, the results were astounding, as the computer outscored humans at a rate of 20-30 percent.

DeepMind and Google

Google acquired London-based DeepMind for an estimated few hundred million. Who knows, maybe sometime in the next few decades you will be relaxing in the front seat of a self-driving DeepMind developed car. DeepMind’s initial startup funds were supplied by Tesla Motors. Either way, the goal is to teach computers to apply learned skills to complex issues, minimizing the need for human efforts in the not so distant future.