The second type of tasks AI can manage is data analysis. Arficial intelligence laboratories insides the GAFAM companies work for most of their time on the analysis of titanesque quantities of data that they collect from billions of users. Those technologies are called Machine Learning or Deep Learning according the level of complexity of the analytics methods that are used.

 

Experts driving those artificial intelligence platforms try to identify profiles, behaviors, interests of each and every user and adapt their algorithms who select search engine results or push te publications on their newsfeed or images on their photo sharing application, that are the most similar to the ones that generated some interest in the past from a given user. First goal : increase the time spent on applications as much as possible. Second goal : display the most relevant and engaging ads. Third goal : trying to predict consumer behavior, for example, to show them contents or offers to bring them to buy or to go on using the app.

GAFAMS companies are not the only ones in this areas : some ecommerce start-ups, sometimes very efficient and successful, develop a proprietary AI technology to optimizer their revenue, select products with the highest potential, whereas some major corporations try to get a better understanding their customers habits. Success of these research and development works is closedly linked to the relevancy of the technical bets, the accuracy of analyis models and the agility of participating organizations. Understand : all of AI machine learning or deep learning experiments are not successful.

Machine Learning and Deep learning to teach machines the way humans perceive informations : through the analysis of hundreds of millions of faces, computers end up understanding technical characteristic of a picture showing a human face as well as the technical differences between the faces of two people or the shared characteristic between the face a women with or without make up. These technologies o face.

These facial recognition technologies allow, for example, a smartphone to automatically classify your selfies in a special folder or even to use the face of the owner to unlock access to the smartphone. The same type of technology, used in speech recognition, translates a person's speech into a sequence of words and used in character recognition, converts handwritten text into a Word document. On the other hand, it does not allow the computer to understand the meaning of the sequence of words and even less the quality of the content of the handwritten text. At best, it would be able if coupled with a search engine technology, to understand the main subject of the text and sometimes to identify the sub-topics.

Artificial intelligence can also be interested in modeling the reasoning of a human when analyzing data or evaluating a situation. How does an accountant understand and analyze a financial statement? Balance sheet publishing softwares have integrated the variations in a value (for example, production) from one year to another, but one can now imagine that by decomposing all the stages of the reasoning of an accountant , it is possible to create software that launches alerts when the cash flow of a company evolves abnormally, detects the most profitable products of a company, or makes recommendations on wage policy based on decomposition and the evolution of the wage bill. On paper, as is often the case, the potential is there, in practice, models of analysis of this type, are often insufficiently elaborated to reproduce the subtilities of human reasoning. But the models are evolving.