Robots are better at Video Games than you

Robots have apparently become better at video games than Humans. A group of researchers at Google's Deepmind Artificial Intelligence Research group have developed self-learning computers that is able to teach itself to play video games as well learn advanced strategies to destroy all humans in computer games. Deepmind uses classic arcade games such as Breakout in which you have to break down a wall by bouncing a ball against the blocks to break them. The AI figured out by itself that it can break a hole in the wall and then bounce the ball behind it to score more points.

Deepmind uses a system they developed known as "DeepQ Network" or "DeepQN" which combines neural networking and acute learning. The Deep Neural Network is a perception tool that has been created with animal vision in mind which helps the DQN see and interact more like humans do. While the Q learning is a reinforcement learning system that is based on math.

In the case of the AI the points are the reward and the directive is that the AI wants to maximize the reward. Other Researchers have tried directing AI to get max points but what the AI fails at is making short-term sacrifices for long-term gains, a fail in logic.

Breakout and Space Invaders are two games that are easy to learn but hard to master; a complex strategy is needed to really maximize your points. The AI is currently only able to play classic games like those listed above, so it hasn't quite gotten to current-gen game level. So far it has taken on 49 different Atari 2600 games. It's successes have not been mind boggling by average; scoring an average of 20% - 30% more than Humans have in Space Invaders. However, in Breakout the AI scored x10 more points than the best Human has. The next step is to transfer strategic knowledge gained from playing games, to other games.