Google AI Made Its Own Child AI That Outruns Itself

Artificial Intelligence is progressing at a rapid pace. Owing to its importance, Google has also been working on making AI tools to keep up with the pace. It has unveiled an AI system which has created  its own Child AI which is better than its human-made systems.

In essence, it can replicate itself and has the potential to compete with other man-made AI tools . This offspring AI has also been successful to outperform its parent AI (AutoML project by Google).

In May 2017, Google made an announcement that AutoML project had what it takes to create its own AIs. AutoML is an AI tool that acts as a controller neural network. NaSNet is an AI that was created by its parent AutoML earlier this year. Now, they have proved that it is better than its parent AI. AutoML of Google uses reinforced learning to make NaSNet, better in doing task.  

NaSNet can identify and recognise objects such as people and cars, backpacks, handbags, traffic lights etc while watching real time video with a significantly higher level of accuracy than a neural network designed by a human.

And, researchers have found that NASNet was 82.7 percent exact at identifying/detecting pictures on ImageNet's approval set. NASNet's outcomes are 1.2 percent superior to any beforehand distributed outcomes. It tells that NASNet is better as it has outperformed the majority of existing computer vision applications.

You may also like : 5 Best Artificial Intelligence Tools

NASNet  is a visual machine learning AI. It could be used to help visually impaired people to regain sight. Such types of AIs could also be useful for designers to enhance self-driving car technologies. Thus, if it is fast in recognizing objects coming into its path, it will be equally faster to how to respond to them. Therefore, it could be in use of increasing car safety.

While such AIs are of great help to the society, we cannot deny the fact that it also raises serious concerns. What if, it becomes so smart that it could create hundreds of such AIs and what it can do with it. For instance, imagine a scenario in which AutoML makes AIs so quick that human culture can't keep up with. But NASNet could securely be applied in automated surveillance systems in the near future.

Google researchers have said that we trust that the larger machine learning community will have the capacity to build on these models to address large number of computer vision issues we have not yet envisioned.