Breaking out of mazes and solving puzzles with artificial intelligence
The developers of our association are constantly developing the flexibility of learning and final thinking of artificial intelligence to achieve greater versatility in its application. The result is a whole family of programmes and algorithms for self and each other. In this way, our systems have become capable of solving non-trivial tasks and even applying “ingenuity” to these solutions.
The expediency of such an approach in the training of intelligent systems has eventually come to the experts of various companies engaged in the development of artificial intelligence. In particular, DeepMind, one of Google’s divisions, has focused on modelling the neural network of the copying lattice neuron, which in turn is responsible for navigation and coordination in mammals.
To train this recursive neural network, the developers built a network of virtual labyrinths from which programmes must find the shortest way out. A feature of these mazes is that the exit is blocked by a ‘locked door’. The essence of this approach was to configure the algorithm in such a way that, having studied the maze system, he reacted to the ‘door opening’ factor and found the shortest way out, rather than repeating the ‘environment’ as the standard programmes did.
The results of these experiments allowed a deeper understanding of the principles of neural networks in the brains of all mammals, and humans in particular, which gave a new impetus to the development of neurobiology in general and the diagnosis of brain diseases in particular. And in the development of artificial intelligence, these developments make it possible to build more complex and organised systems that more accurately copy the work of the brain.