It’s not hard to see that Google’s got some big bucks simmering away under its AI research projects. And they certainly seem to be making good progress. Thanks to the creation of AutoML in May 2017 by the AI crunchers at Google Brain, we’ve got Artificial Intelligence that’s capable of creating its own AIs. So what’s new? Well, in a recent turn of events, Google AI has now progressed to creating a ‘child’ that surpasses every other human-made AI system like it.
Google AI creates its own ‘child AI’ : All You Need to Know
Super-impressive and highly sophisticated, this new child AI called NASNet has successfully left its human-designed competitors in the dust. Awesome as it is, it does raise concerns as to what other stuff AI could figure out without human assistance.
Google’s machine learning models were automated through a process known as reinforcement learning, where AutoML served primarily as a controller neural network that worked on training a child AI to achieve a set task. AutoML was introduced to ease the labor of designing machine learning models through automating the process.
NASNet was delegated the task of identifying objects (people, vehicles, pets, accessories… you name it) in real time. And through AutoML, it was possible to measure NASNet’s performance and use the information to develop its AI, all the while repeating this procedure several hundred times.
In order to test NASNet, researchers tested it on the ImageNet Image Classification and COCO object detection data sets, which Google identifies as two of the top academic data sets available in computer vision. NASNet scored 1.2 percent better when tested against ImageNet and 4 percent better when tested against COCO, with researchers also claiming better system efficiency (a higher mean average precision value) for NASNet.
Google AI has certainly kicked off to a promising start. So what does this mean for the future? AI researchers at Google seem to have their sights set on everything from self-driving cars and AI-powered robots down to precise medical treatments. The company has also taken steps to open-source its AI for image classification and object detection in the hope that ”the larger machine learning community will be able to build on these models to address multitudes of computer vision problems” not yet resolved.