Neural network DeepMind start to teach other AI abstract thought

Obvious for man to artificial intelligence can become an impossible task in the company, so DeepMind decided to teach the neural network to think abstractly and to connect to other neural networks in order to help and work together to solve complex problems.

To teach the AI to recognize pictures is easy, but to determine which of the objects is larger than another, if they are at different distances — it’s more difficult. For this to DeepMind have developed a neural network which specializiruetsya on such tasks, and then connect it to the AI to improve its capabilities.

During the training, the AI showed him images of different three-dimensional objects, and then began to ask the new program of difficult questions, forcing you to think rationally. To the question, what is the size of the cylinder to the left of the brown metal thing that is left of a large ball, connected to a neural network artificial intelligence answered quite quickly and correctly. Such problems for the AI, the developers have prepared quite a lot, but in 95 percent of cases, the program answered correctly, demonstrating little separation from the people of the rivals.

Neural network Google began to teach other AI in January of this year, but now she can teach them a lot more interesting lessons. Practical application of the design not yet invented, but the fact that artificial intelligence has become smarter and smarter, definitely happy.

Neural network DeepMind start to teach other AI abstract thought
Vyacheslav Larionov


Date:

by