Artificial Intelligence from Google is Now Multiplying
In May 2017, researchers from Google Brain introduced AutoML, an artificial intelligence capable of generating its own AI.
Recently AutoML created a new artificial intelligence without human intervention. It is called NASNet and the developers note that the new neural network surpasses the counterparts created by humans.
The AutoML program acts as a controller, which develops a neural network for a specialized task. Its goal is recognizing objects – people, cars, traffic lights, bags, backpacks – on real-time video. At the same time AutoML evaluates the work of NASNet and uses this information to improve its ‘child’; this process is repeated thousands of times.
According to researchers from Google, NASNet successfully provided images in the ImageNet control sample in 82.7% of cases. This is 1.2% better than the previous record. In addition, the less expensive version of NASNet has surpassed the best similar models for mobile platforms by 3.1%.
There are many options for using AutoML and its ‘subsidiary’ network, NASNet. Similar computer vision algorithms can be used to create complex robots or, for example, to help people with eye disorders. Such algorithms can also assist to improve the technologies of autonomous driving.
Apparently, the main disadvantage of creating such neural networks is the inability to control the processes of AutoML and the creation of other artificial intelligences. Thus, the system can create a huge amount of AI without even notifying a human about it.
However, many big companies are trying to take into account the security problems of AI. For example, Amazon, Facebook, Apple and some other corporations are members of the Partnership on AI to Benefit People and Society. The Institute of Engineering and Electrical Engineering (IEE) also proposed ethical standards for AI, and DeepMind, for example, announced the creation of a group that will deal with moral and ethical issues related to the use of artificial intelligence.