For the first time ever, a machine has beaten top human players in an arcade video game.

Artificial intelligence now has the ability to study a problem and adapt certain behaviors to perform at a higher level. In this case, the system studies games and how they're played proficiently without even observing how humans perform the task.

Like Us on Facebook

Researchers from Google have developed a new artificial intelligence (AI) system that can beat the best and most skilled human players at classic arcade games. This AI capability can someday be applied to driving cars or even perform medical surgery.

This ingenious ability called unsupervised learning means that machines can improve and adapt their performance without any human guidance or interference.

AI can now recognize handwritten zip codes and even recognize song samples. Researchers now believe the same process used to assist machines to learn skills can also be used to learn new things they were once incapable of doing.

The company responsible for this new breed of AI is called DeepMind, which was purchased by Google for US$500 million. Formed by computer scientist Demis Hassabis and his team of 50 employees, they developed new AI technologies utilized by computers to perform a wide variety of tasks.

The system used for this AI is called the Deep-Q-Network that allows the artificial mind to absorb and learn new tasks. The processes consist of two parts. One involves a deep neural network that observes the pixels in the game and learns to make moves based on present conditions.

The other process is called Q-learning that enables the computer to pinpoint which moves are the most effective by recording and analyzing how points can be gathered after a number of different sequences of moves, and then repeating those that are most effective.

Researchers compare this to reward structuring found in animal and human brains.

To test this new AI and its problem solving skills, the machines played a classic video game called Breakout where a ball is bounced against a wall. Without any human guidance, the artificial mind quickly learned several tricks used by human players like targeting the ball at the same spot on the wall to accumulate more points.

According to Tomaso Poggio from the Massachusetts Institute of Technology, this ability suggests computers can now learn similar, realistic tasks like driving a car. It can eventually learn abstract reasoning and even social perception.

Some 49 classic Atari arcade games from the 1980s were used for the AI experiment. These games were chosen since they're simple enough for the machines to learn but also complex enough that winning isn't by chance.

This study was published in the journal, Nature.

TagsGoogle AI Can Now Play and Master Arcade Games Without Instructions, AI, google AI, DeepMind, atari arcade games, space invaders, breakout game, AI artificial intelligence computers machine video arcade games