We can already see examples of how artificial intelligence technology is able to show some that seem at first glance, characteristic only of human traits. We create humanoid robots, at least very similar to us, some are engaged in that create algorithms able to perform something that usually only people are able to write music, paintings or teaching.
With the development of this sector of the company and developers are starting to look for an opportunity to change the very basis on which we are creating artificial intelligence algorithms and are accepted for the study of real intelligence, and of the way how to simulate it in the engineering and the software development of a new generation. One such company is IBM, who set an ambitious goal to teach the AI to behave (more correct to say to work) more like the human brain, not as a set of programmed algorithms.
Most of the existing systems of machine learning are built around the need to use huge sets of data. Be it a computer, designed to find ways of winning the game, or the system, built to detect signs of skin cancer on the basis of digital images – this rule always works. But such a basis for the work looks very limited and concise, and of course that is what distinguishes such systems from how the human brain operates.
IBM wants to change that. The research team from DeepMind has created a synthetic neural network, which is based on rational decision-making when working on a particular task.
Rational machine
“Giving artificial intelligence a set of objects and a specific task, we force the network to discover existing compliance,” comments on the pages of Science Magazine Timothy Lillicrap, computer specialist DeepMind team.
In tests of the network conducted in June, the system, in the presence of a variety of factors were given various tasks associated with a digital image. For example, such: “the blue thing in the picture is the object. It has the same shape as that tiny blue thing is to the right of the grey metal ball?”
In this test, artificial neural network was able to identify a desired target in 96% of cases, while conventional machine learning models able to cope with the task in 42-77% of cases.
Recently artificial neutron networks continue to improve in the understanding of human language. The researchers also want in addition to sound decision-making, such systems could demonstrate and maintain attention, and to store memories.
The words of Irina Rish, IBM researcher, the development of artificial intelligence could significantly accelerate and expand the use of such tactics.
“Improving neural networks are still a matter of engineering usually requires a tremendous amount of time to come to the right architecture that work best. In fact it’s the human method of trial and error. It would be great if the networks could themselves create and improve”.
Some, of course can scare a thought about the AI networks are able to create and improve, but if you find a competent way to monitor, control and manage this process, it will allow us to go beyond today’s limitations. Despite the growing fear about the revolution of robots who enslave us all, the development of AI, predicting thousands of saved lives in medicine, opening for us opportunities to visit and even to live on Mars and much more.
IBM sees artificial intelligence as a set of common algorithms
Nikolai Khizhnyak