A new program, which can find its way to a virtual maze, was presented by Deep Mind, Google’s subsidiary in the field of artificial intelligence.
The originality lies in the fact that the deep learning algorithm of the new artificial intelligence system can “spontaneously” mimic the activity of human brain neurons (so-called “mesh cells”) that allow a person to know his place in the space, something vital to navigate a labyrinth.
The great surprise was that the new program – a deep neural network using multiple layers of artificial neurons to process information – reproduced the complex nerve codes that, as has been demonstrated since 2005 in laboratory animal and human experiments, the brain uses for to orientate.
In other words, artificial intelligence has used itself the same mental intelligence strategy in space, which is also used by biological intelligence. This achievement is expected to help neuroscientists in their research into brain function.
Deep Mind researchers and University College London (UCL), headed by Darsan Koumaran, who published the publication in Nature magazine, used the new system to train a digital rat to finds the exit to a digital labyrinth more quickly and by a person skilled in this game.
Experts spoke of achievement – landmark for artificial intelligence. To date, such programs have achieved impressive things and beat people in various board games (chess, poker, Go, etc.), but navigating a labyrinth is a special challenge in the field.