
Google has taught artificial intelligence to create a more effective software for machine learning, which a person is able to develop, writes the publication of Wired .
We are talking about the project "Automated machine training" (Automl), launched in May. Within its framework, artificial intelligence creates neural networks.
At the moment, he was already able to set a record: the algorithm developed by him has achieved 82% of success in solving the problem to determine the shown images. The comparison of the results achieved by software from artificial intelligence and developer-man is impressive: in the first case, the algorithm was able to recognize up to 43% of objects in one image, in the second-only 39% of objects (this is the best result among the programs created by a person).
Google CEO Sundar Pichai claims that the task of the Automl project is not to replace developers with artificial intelligence over time, but only to help them “at the initial stage of setting up self -learning systems”. So far, such systems are created by machine learning specialists, who in the world have only a few thousand, their work is so complicated. When artificial intelligence will be able to conduct machine training itself, hundreds of thousands of developers will be able to come to the industry, Pichai claims.
However, independent experts do not share his optimism: in their opinion, Google pursues precisely the purpose that he denies - tries to displace from the market for the development of simple artificial intelligence algorithms due to the automation of this process. Analysts especially note the suspicion of the fact that the company does not allow third -party specialists to work on the Automl project.