Google has artificial intelligence, capable of creating other artificial intelligence systems, which are better at detection and recognition, than systems developed by people.

Large companies are investing a large amount of capital in the complete development of artificial intelligence systems. Well, it seems that we are at the beginning of a new world, in which an artificial intelligence system develops other artificial intelligence systems. This advance has been made through Google's AutoML project, which has been developed for computerization through a system that beats the most cutting-edge. Researchers at Google Brai, in May this year, created the initiative that allows a machine learning algorithm to learn how to build other machine learning algorithms.

This idea was developed in order to see how an artificial intelligence was capable of creating another artificial intelligence system without human intervention, with the sole objective of seeing how to achieve a greater deployment of this technology. There are very few people capable of developing these technologies and they are in high demand, so a project that is capable of creating artificial intelligence that developed others create other artificial intelligences for other fields and companies, more quickly. The slow development of the AI ​​itself puts this field at risk, according to experts such as Dave Heiner, a Microsoft advisor.

The CEO of Google, Sundar Pichai, has boasted during a presentation of the Pixel 2 and Pixel 2 XL, that AutoML and today could again boast about a promising initiative. His biggest challenge to date. They have succeeded in automating the design of machine learning models using a system called reinforcement learning. Researchers have made artificial intelligence act as a neural network of controllers, which is capable of creating another smaller artificial intelligence network, which is dedicated to a specific task. The creation, dubbed NASNet, has outperformed all human-created counterparts.

The function of this new system is capable of recognizing objects that are broadcast in videos in real time. You have to recognize people, cars, bags, backpacks and other items that are seen on the screen or image. AutoML has the potential to evaluate performance and using this data, autonomously refines the lesser artificial intelligence by having it repeat the process thousands of times. It is a complex function, normally performed by humans, but one that is fundamental.

Google researchers cross-checked NASNet’s identification data against “two of the most respected large-scale academic datasets in computer vision,” namely the ImageNet image classifier and the COCO object detection dataset, and found that it outperforms all other human-developed computer vision systems. According to the data, it was 82.7% more accurate at predicting the ImageNet validation set, which is 1.2% higher than anything previously seen. The system has 4% more evidence with an average accuracy of 43.1%. In terms of computational resources, it outperformed the best similarly sized models designed for mobile platforms by 3.1%.