What is the basis for the fantastic performance of artificial intelligence in the digital industry?
The evolution of algorithms is based on Big Data. This data is collected via the Internet and its diverse and varied platforms that involve millions of people around the world. But it is still necessary to be able to use it and to teach algorithms to use it!
There are 2 methods:
Supervised learning and unsupervised learning.
Unsupervised learning
Machine learning is a category of algorithm that allows software applications to predict results more accurately without being explicitly programmed. The basic principle of machine learning is to create algorithms that can receive input data and use statistical and mathematical analysis to predict a result. In practice, some systems can continue learning and refine their statistical analysis.
Supervised learning
It relies on human labour. In France alone, RN education employs 260,000 people.
It is armies of behind-the-scenes workers around the world who ensure the learning of algorithms. Indeed, for a system to be able to recognize a cat in an image, it must be confronted with millions of images of cats, animals or objects. Only humans are able to describe a cat and differentiate it from a dog, for example.
Once this preliminary work has been done, we will ask humans to teach AI to differentiate breeds, then colours and their shades, to locate contours to the nearest pixel?
In short, a titanic task!
But here, we’re only talking about the cat, this work is even more essential when we’re talking about autonomous driving where data quality is a matter of life or death. The analysis and tagging of a single traffic scene alone takes an hour and a half.
Thousands are needed because as the analysis progresses, the error gets smaller and smaller.
In addition, thousands of variants have to be integrated without forgetting any because, to use a simple example, you can train an artificial intelligence to recognize the expression “Open Sesame”. It will then be able to recognize it even with a voice it has never heard or an unlikely accent. On the other hand, it will not be able to recognize “Alibaba” if it has not been trained to do so.
It is not only the training of the algorithms that is concerned. Indeed, companies offering virtual assistants, such as Amazon, google or Apple need to know that the AIs respond correctly to requests and are not deceived by background noise, accent and that the system is capable of adapting properly.
This area of quality control is a growing field, even though it has been controversial recently.
Our data is currently being analyzed, our conversations peeled by armies of small hands in order to make AIs learn all the subtleties of our existences and our lives and to make our world evolve towards more and more digital for the best… or the worse.
M.D