AutoML was born in May by Google and its goal was to create other artificial intelligences on its own
The result, an artificial intelligence that recognizes information better than any other created by humans
There are many technology experts, such as Elon Musk or Bill Gates or Stephen Hawking, who fear the scope and dangers linked to the development of artificial intelligences too clever.
Now that one of the AIs of Google has created an artificial intelligence that is better than none made by a human being, is it time to join them? Of course, it is better than other virtual minds in a function: recognizing images within a video on the move, so it’s not like there’s a higher mind than homo sapiens on Google servers.
All this experiment has its origin in May, when Google announced the creation of AutoML, an artificial intelligence that generates other partners specialized in specific functions.
The technical aspects behind this feat are very complex, but, in summary, what AutoML has achieved is to automate the behavioral models normally created by human programmers and engineers. After several attempts, one of the ‘daughters’ of this virtual mind has been a success.
NASNet is the artificial intelligence in question and is able to recognize people and objects such as traffic lights, cars, bags or traffic lights with a precision that surpasses other AIs programmed by traditional methods: with a person behind.
An example of how it identifies NASNet images (Google Research) The relationship between mother and daughter is very intricate, and every time NASNet carries out an operation, behind it is AutoML correcting and improving its behavior.
Repeating this thousands and thousands of times, the second artificial intelligence is getting better and better in its sole task. According to the researchers after this project, NASNet has evaluated the images of the ImageNET test with 82.7% accuracy.
Compared to previously published results, it is an improvement of 1.2% over other intelligences programmed by man.
In turn, it is also 4% more efficient, so it not only acts better, but also consumes less resources.
With these results, NASNet could be a perfect candidate to create automatic surveillance systems that recognize people and categorize them in public spaces. As always when an artificial intelligence emerges in some task, there arises once again the fundamental question about ethics in its creation and about the problems that can arise from a machine that can create another machine.
This is not a matter of science fiction and post-apocalyptic panoramas, however. Many worry that artificial intelligences that are ‘reproduced’ can forge the prejudices and problems that are already present in some automated systems.