Already now one can observe examples of how artificial intelligence technologies are able to exhibit some features that seem at first glance and are characteristic only of man. We create humanoid robots, at least very similar to us, some are engaged in creating algorithms that can perform what people are usually capable of – writing music, painting, or teaching.
With the development of this sphere, companies and developers are beginning to look for an opportunity to change the very basis on which artificial intelligence algorithms are based now, and are accepted for researching real intelligence, and also how to effectively simulate it in engineering and creating software of a new generation. One such company is IBM, which has set itself the ambitious task of teaching AI to behave more correctly (like a human brain) rather than as a set of programmed algorithms.
Most existing machine learning systems are built around the need to use a vast array of different data. Whether it's a computer designed to look for ways to win a logic game, or a system built to detect signs of skin cancer based on digital images – this rule always works. But such a basis for work looks very limited and concise, and of course this is what essentially distinguishes such systems from how the human brain works.
IBM wants to change this. The research team from DeepMind has created a synthetic neural network, based on rational decision-making when working on a particular task.
"By giving artificial intelligence a lot of objects and a specific task, we force the network to detect existing matches," says Timothy Lillicrap, a computer expert at the DeepMind team, on the pages of Science Magazine.
In the network tests conducted in June, the system, given a variety of factors, was given various tasks related to the digital image. For example, this: "Before the blue thing in the image is an object. It has the same shape as that tiny blue thing that is to the right of the gray metal ball? "
In this test, an artificial neural network was able to determine the desired object in 96 percent of cases, while the conventional machine learning models were able to cope with the task in 42-77 percent of cases.
Recently, artificial neutron networks continue to improve in the understanding of human language. Researchers also want, in addition to making reasonable decisions, such systems could demonstrate and maintain attention, and also store memories.
According to Irina Rish, an IBM researcher, the development of artificial intelligence could be significantly accelerated and expanded through the use of similar tactics.
"Perfection of neural networks remains a subject of engineering, usually requiring a huge amount of time to come to the desired architecture that works best. In fact – this is a method of human trial and error. It would be great if these networks could create and improve themselves. "
Some, of course, may be intimidated by the idea of AI networks that are capable of creating and improving themselves, but if we find a competent way to monitor, control and manage this process, it will allow us to go beyond the limitations that exist at the moment. Despite the growing fear of the revolution of robots that will enslave us all, the development of the sphere of AI is prophesied by thousands of saved lives in medicine, the opening for us of the opportunity to visit and even settle on Mars and much more.