Updated neural network algorithm allows mechanisms to reason
The development of artificial intelligence happens currently due to the success of deep learning. This is a set of machine learning methods for creating ideas about objects and tasks in robotic systems. As you can see, this is not about ordinary algorithms for tasks (as is the case, for example, with chat bots, automatic cleaners and other similar machines), but about more complex neural network algorithms.
Deep learning works in a way that imitates human abilities in mechanical entities: to see, hear and reason. Similar algorithms have been used for a long time, for example, in the work of Google-assistant and even when performing a normal search query.
Neural Network Algorithm: From Commands to Computer Logic
American scientists with neural networks have analyzed more than 17 thousand scientific papers, published since 2000, devoted to research in the field of artificial intelligence. It turned out that in the early 2000s, most scientists were engaged in neural networks. In 2010, the votaries of science realized that systems needed to be learnt.
Initially it was thought that it was necessary to program robots with all the knowledge available to humanity. This branch was a dead end: after processing all the information, only a search engine or help desk can be obtained.
Instead of knowledge, robots tried to engender common sense. This is due to the fact that in order for the technology to work properly, too many rules and algorithms need to be captured in the form of program code. Instead, the system was learnt to see, hear and analyze information.
Manual coding of millions of rules has been replaced by automatically obtaining the necessary information from the data array. For example: if a human thinks that his personal assistant Siri has become smarter with time, this is not the delusion. As a result of interaction with an individual, the system learns itself, therefore, it becomes more intelligent.
There are several areas of machine learning, but the main ones among them differ in the degree of human participation in it. The controlled area is when the system consumes data arrays specially allocated for it. Reinforcement learning simulating school is also practiced: with good and bad grades and a variety of incentives. The result of this kind of researches was the victory of AlphaGo system in 2015 over a human (world champion) in the competition in the Chinese logic game Go.