In may this year we wrote about the project AutoML technology artificial intelligence (AI) of Google, designed specifically to create the other AI. Now Google has announced that it AutoML was able to exceed the developers of the AI and is able to create their own software for machine learning, which is more effective and powerful in comparison with the best examples of similar systems developed by man.
AutoML recently set a record in efficiency and speed of cataloguing images on the specified conditions, showing the result in 82% efficiency. Although this task in itself proved to be relatively simple, AutoML were also able to outperform automated systems and special augmented reality system in a more complex task of determining the location of multiple objects in the image. In this test AutoML handled in 43% of cases, while human-made systems showed efficiency level of 39 percent.
The results are impressive, because even in such a huge company like Google, there are only few people with enough experience to lead the development of AI systems. To automate this sector requires a very broad skill set, but as soon as the result will be achieved, it may completely change the industry, as noted in Google.
“Today, to create such software can only a few thousand specialists in machine learning in the world. But we want to make hundreds of thousands of other developers, too, could take part in it”, — quotes the words of the CEO of Google Sundar Pichai Wired magazine.
A significant part of metalocene associated with simulated neural networks of the human brain, as well as the need to run huge volumes of various data through the networks. Of course, the most difficult task is precisely to mimic the structure of the brain and force it to solve more complex problems.
Existing neural networks are still easier to upgrade or customize for specific tasks, rather than developing a new system from scratch. However, a study like the one we’re talking about, suggests that it is only a temporary phenomenon.
Since the new AI will be easier to create ever more complex systems, which is used to perform the tasks to fulfill that right now people are simply unable to, it is very important that a person remained as a key element, without which the system simply cannot function. Truly a full-fledged AI will be able to use a biased interpretation of certain matters, for example, stereotyped appreciated the parallel between ethical and gender features. However, if engineers will devote more time to solving this potential problem now rather than leaving everything for later, in the future it will have less chance at a real appearance.
In General, Google tries to hone AutoML to such a state that the developers could use it in a real decision task. If they do, the effect of using AutoML can have an impact far beyond the walls of the company.
“We want to democratize”, — quotes the words of Pichai Wired magazine.
Software machine learning Google learned of reproducing itself
Nikolai Khizhnyak