Facebook is listening to us. Or rather, we listened. An investigation by the US media Bloomberg reveals, Tuesday, August 13, that Mark Zuckerberg's company paid subcontractors to retranscribe sound clips of conversations of some users.

After long denied doing so to better calibrate its ads, the social network has declared an end to this practice, which is nevertheless a very valuable source of information for digital giants, aka Gafam (Google, Amazon , Facebook, Apple and Microsoft). This confession comes a few weeks after similar revelations concerning Amazon, Apple or Google.

In April, Bloomberg was already alerting Amazon employees to listen to users' queries with their Echo connected speakers. At the end of July, the Guardian revealed that Apple was doing the same with the queries sent to his voice assistant Siri. Microsoft and Google have also been affected by similar scandals.

Why this rush to sound recordings? For Nicolas Mourier, independent consultant and specialist of connected objects, "it's a gold mine, much more than textual conversations," says the expert. "The human being is worried about the cameras we see ... [...] On the other hand, listening to a person makes it possible to obtain a thousand times more data and this, in an easier way: one retrieves one's opinions but also information on his social class, his interests ... What Facebook is interested in is not surprising, he had until then a train late for the Gafam who produce voice assistants. "

A listening to refine the algorithm?

Facebook, like other digital giants, has justified - after the surveys - conduct these tapping to improve its services. Facebook's subcontractors checked whether the artificial intelligence of the network correctly interpreted the messages that had been made anonymous.

A justification that sounds right for Nicolas Mourier. "There are two interests to this kind of transcription: compare what the human understands with what the recognition engine includes and see if the latter does his job properly and then feed it, feed it, refine it."

"It's the amount of data to compare that makes the difference in performance.It's not enough to have the best voice recognition engine in the world technically.Google is better because it has the ability to compare what you are saying with the mass of accumulated knowledge, "he explains.

And the expert to give an example with the French nursery rhyme "Once upon a time, in the town of Foix, a liver vendor, who sold liver ... She said to herself: my faith is the first time and the last time, that I sell liver, in the town of Foix "to illustrate his point. "In the event of an error during the declination of the nursery rhyme, the best voice recognition engine in the world will retranscribe this fault as it stands, and Google will be able to correct it by comparing it with its database."

Lack of transparency

Facebook has long denied using audio recordings. Its founder, Mark Zuckerberg, even rejected the idea during a congressional hearing in April 2018. "We do not do that," he said firmly. "You're talking about a conspiracy theory that says we're listening to what's going on in your mic and we're using it for advertising," the young billionaire told a question from Senator Gary Peters. Later, Facebook had specified that it only accesses the microphone in case of consent. However, Nicolas Mourier believes that targeted advertising is the goal of the digital giants.

"We are collecting voice information and then advertising, it's the economic model of Facebook and the model of Facebook and Gafam.There is no doubt about it," says -t it.

For Numerama, a media specializing in new technologies, Gafam's lack of transparency regarding these practices is "problematic".

"Companies [should] find other solutions, for example, they could better ensure users' consent, or even offer compensation to those who accept ... After all, they provide them with very useful data that will undoubtedly end in commercial products ", explains the French media.

"You have to do pedagogy on the subject and explain to people to pay attention." Explain that if the speaker reacts to "OK Google" or "Alexa" is that we are listened to constantly, "warns Nicolas Mourier . "And especially explain that bringing an ear into his house is much worse than an eye."