Google’s AutoML neural net has created its own artificial intelligence “child” called NASNet, and NASNet outperforms any human-created machine learning tool at computer vision tasks. More specifically, NASNet is 82.7 percent accurate at identifying the content of images. This is 1.2 percent better than any other AI tool has performed at the same task.
The experiment shows that AI tools can be very successful at developing AI tools on their own. And because AutoML is open source technology, it also opens up the possibility that other developers will soon create their own “AI children” that can perform even better than NASNet.