From the recommendation of contents to the automatic writing of texts, to the conversational, vocal or written agents, the perspectives of the AI are not all equally realistic and promising. The economic models are not yet clear, and its applications to the sector are mainly experimental. "It's a billiard table with two or three bands," says Erwann Gaucher (France Bleu.). Artificial intelligence still brings little money to the media, and risk-taking is great: investments are scarce, although the potential is large and monitored closely.
This market, like the Internet of objects that is slow to keep its promises, is part of the great technological bets of today. For Jean-Dominique Séval, Deputy Director IDATE DigiWorld, the expectations are immense, the major upheavals announced but the path to a mature market and the return on investment associated, still uncertain. Between the current world market, estimated at just over 200 million Euros and the promise of a turnover generated of more than 10 billion Euros by 2025, there are still real uncertainties in the fields of Applications whose media are a perfect example.
Recognition and Recommendation
Since Amazon launched Alexa in 2016, the general public has discovered artificial intelligence at the heart of the digital home. For the time being, this personal assistant sold as a speaker is mainly used to manage very basic daily tasks (giving time, weather, traffic conditions on the way to work), but its development and commercial success leave The possibilities of speech recognition for the media. Beyond the digital home, voice-activated universal wizards, already studied for smartphones, will find their way into the connected cars of tomorrow. "It's both potentially fantastic and monstrous," judge Erwann Gaucher. "We know that the voice commands will be present in the car, which represents more than 30% of the time of listening to the radio. We will have to learn how to use this AI to make the car a privileged listener for the radio, especially since this AI of voice-activated assistants opens the door to the recommendation, "he explains. For television, it seems that the challenges of voice recognition are measured in the longer term. For François Fourrier (Canal +) "the battle of access is launched, but users need depth in the recommendation. And for that, the voice media is not yet adapted. We must think about the complementarity between speech and visual recognition. "
Because the main issue of the recommendation lies in the association of content content (content to content), and not solely on the analysis of past uses and consumption habits of users' profiles (affinity recommendation). A level of precision that technology can not yet achieve alone, as Claude de Loupy (Syllabs.) Recalls. "Speech recognition works very well when it comes to a JT presenter with a clear diction. But in a noisy environment with three people, it's more complicated, "he explains. While voice recognition is an enormous potential for the media, it is also the most versatile technology today compared to text and image. "In artificial intelligence, performance depends very much on the context. Depending on the environment, speech recognition performance can fall well below 80%. We must also take into account the semantics and the treatment of languages. Now, language carries within it all the complexity of human intelligence. We do not have the capacity to have a level of analysis equivalent to a human, "explains Claude de Loupy, recalling that, when it is well configured, the automatic generation of texts almost never makes errors .
Artificial intelligences capable of writing texts do not always pass the door of redactions, but they have an undisputed effectiveness. Erwann Gaucher (France Bleu) said that no mistakes had been made in the 36,000 articles written to communicate the results of the regional elections in December 2015. He now thinks of using the same system for weekly sports results. "We are in the soft belly of the information that everyone has, but that we must nevertheless communicate. I prefer to entrust this to an editing algorithm, and that sports journalists add value to the field rather than write the results every Sunday night, "he explains.
Olivier Delteil (Les Echos) answers journalists who are afraid of being replaced by robots rather than a time saving and qualitative for their work. "It allows them not to give the value of an action five times a day, but to analyze its evolution beyond the stock market. We will wait for mentalities to evolve and make pedagogy, "he says. In addition to the automatic generation of texts, Olivier Delteil notes "quite impressive" progressions in the automatic translation of texts, even if the human must always intervene at the end.
At the base of Artificial Intelligence: data management
As for the use of data, it now appears to be confined to commercial purposes, as the information media are not structured to use reliable data in relevant journalistic use. "The current structure of the media is not adapted to data collection. Today, we are not able to process, store and identify data. It must be taken into account in AI investments, and costs become astronomical, "says Erwann Gaucher (France Bleu). Claude de Loupy (Syllabs) regrets that the news media does not consider artificial intelligence to be a priority. "Innovators tell us that a robot would allow journalists to do their jobs better and develop business opportunities. But it is not in their priorities. As a startup manager, I'm not moving forward. From technology vendor, I turn into a content seller. The media is the most innovative industry that exists, but a revolution in the production of content is at work, and they are being overtaken by companies on these subjects, "he deplores. The phenomenon of brand media, that is to say brands which produce their own content, goes in this direction. "Red Bull now sells a lot of content. When a brand of lemonade sells content to the media, we see an evolution, "continues Claude de Loupy.
Man and machine: AI low vs. AI strong
The replacement of man by the machine in the media is still largely a matter of fantasy, at least in the near future. According to Claude de Loupy (Syllabs), we are in the presence of a weak intelligence which confines itself to very precise tasks. "I do not see anything at the moment in the laboratories that suggests that in less than a hundred years there will be a strong intelligence, but perhaps a new method will come out tomorrow that will make me say that we will be in 5 years, "he said.
For François Fourrier (Canal +), the work of man will always be indispensable in the media, because "as efficient as artificial intelligences, they will not be creative, they will not be able to judge value, Manage the emotional, "he thinks. A point of view that is not unanimous among other specialists in the sector. Julien Breitfield of the agency Fabernovel recalls that during an experiment two personal assistants Google fell in love. "Artificial intelligence means making choices. The machine gives answers. The machine is not asked to be intelligent, it is asked to give meaning. When two machines fall in love, it is not because they are intelligent, but because they understand and say things that make sense, "he argues.
A completely conversational media, where the user would ask questions about the actuality to a machine, is rather science fiction for now. The chat model on the news introduced by Quartz does not allow questions directly to the machine, the user must choose between several questions written before. "At this point, it's a new way of doing storytelling, but not conversational. News is not a given, and is therefore not easy to exploit. What robot could understand and analyze the hesitant waltz of François Fillon's supporters during the campaign today? In the short term, it is unthinkable, "slice Erwann Gaucher (France Blue).
We are already confronted with a weak artificial intelligence, with functions limited to hyperspecialized tasks. On the other hand, the delay of evolution towards a strong artificial intelligence is still nebulous, and its consequences on the media world unreadable.