Artificial intelligence can dream
This has significantly improved system performance.
Google’s DeepMind developers have improved the algorithm performance by teaching AI to dream. As a result, the new Unsupervised Reinforcement and Auxiliary Learning agent or Unrea system passes the Labyrinth game 10 times faster than any other artificial intelligence, reports Bloomberg.
According to experts, the ability to dream is one of the reasons for this result.
Bloomberg reports that Unreal “mentally” reproduces past attempts at games, focusing special attention on those moments when the system earned points.
The developers claim that this is comparable to a person’s ability to dream, and at the same time analyze his negative experiences.
It is also known that Unreal has mastered 57 new games, including Breakout. At the same time, the system coped with this task much faster than other algorithms created by developers.
Experts say the system can play games 880 percent better than a human.
DeepMind is a Google project dedicated to the development of computerized learning algorithms as well as artificial intelligence. In the field of vision of the company’s developers are such tasks as reducing energy consumption, as well as improving the quality of medicine. For this purpose, the algorithm is already being implemented in the British healthcare system, as well as the power supply system.