On YouTube, AI helps fight deepfakes
Published on 03/11/2025
Thematics :
On YouTube, AI helps fight deepfakes
Published on 03/11/2025
You are now Managing Director of YouTube France. How do you view the changes in the tech world over the last decade?
My background is in television production. Before YouTube, a creative had to pitch to a producer, who in turn had to find a broadcaster, before they had any hope of reaching an audience. YouTube revolutionised video and music by bringing the creator closer to their audience. Generally speaking, social networks and video hosting sites such as YouTube have removed barriers, giving an extraordinary boost to amazing talent that might not necessarily have had the right connections to break through. The ecosystem has grown in maturity over the last fifteen years. YouTube creatives have become entrepreneurs, with a sustainable business model. YouTube will be celebrating its twentieth anniversary next year – that’s very old in the tech world! We are proud that we return more than half of the advertising income to the channel owners, who have earned acclaim and admiration. Amazon Prime even brought out a documentary series on Squeezie (editor’s note: the first YouTuber in France). And a new generation is emerging, with creators like Océane and Nixen. As well as that, the smartphone has turned people’s habits upside down, it’s become the main means of using YouTube.
What about tech more generally?
Digital has transformed the timespan of information. We receive huge amounts of data almost instantly. Today’s students grew up with Google. They have amazing tools at their disposal to bring people and borders closer together. Of course we need safeguards to moderate content. Artificial intelligence reinforces the need to fight fake news. Tech has been rolled out very rapidly. It is essential for regulations to be set up to manage it all.
Would you say that tech has evolved mainly for good or for bad purposes?
I think there is a real need for regulation, as well as for clear, value-sharing, economic models. YouTube has a very robust infrastructure to neutralise fake news and content that contravenes its rules, for example, because it is too violent. As far as intellectual property is concerned, we have had agreements with bodies such as Sacem and SACD for more than ten years now. And the model has become permanent because of the income we redistribute. In addition, for the last three years, YouTube has published a report on its economic and societal impact.
YouTube’s number one priority is “to put AI at the service of human creativity”. What do you mean by that?
It’s all about using artificial intelligence, especially generative AI, to extend human creativity. First, there’s the fight against deepfakes. Half of everyone on earth is voting in an election this year. There is greater risk of disinformation via this digitally manipulated false content. These creations not only disrupt public opinion but also spread conspiracy theories.
In the face of this threat, YouTube has set up a system of dedicated labelling. A label warns the user when realistic content has been assisted by artificial intelligence. It is either generated by self-declaration from the creator or by being detected by our infrastructure. The objective is to protect the user, to provide some assurance about what they are seeing and hearing.
Alongside these defensive measures, YouTube encourages AI-assisted creativity. In particular, we offer an automatic dubbing tool, Aloud, capable of translating audio tracks into 80 languages by adapting the voice to the characteristics of the original creator. Another innovative tool, Dream Screen, allows users to generate backgrounds for their short videos from a simple textual description. These tools are aimed at reducing the time between the initial idea and its realisation, thereby facilitating the creative process.
Finally, YouTube Studio, the back-office management platform, offers creators valuable information about the preferences of their community, thereby enabling them to target their content more precisely. They can check whether a subject, for example yoga, will interest their community, and refine their editorial approach accordingly.
You mentioned the problem of deepfakes, videos of false information or which spread conspiracies. What exactly is YouTube, with its two billion users, doing to counteract this phenomenon?
The platform has put rules in place which forbid the promotion of violence, racism, or anything stirring up hatred as well as very strict policies concerning misleading content. If these directives are not followed, the content may be removed, or the whole channel may be blocked. We use automatic learning to identify and neutralise videos that break our rules of use. Up to now, AI has succeeded in automatically filtering 90% of inappropriate content, of which two thirds is dealt with even before these videos have been viewed ten times.
What are your memories of NEOMA?
I was in the 1999 year and was part of what was called Sup de Co Reims. I joined the School after giving up studying Medicine after a year and deciding at the last minute to switch to a BTS course. I have a wonderful memory of the induction weekend at Port Barcarès (editor’s note: near Perpignan). I was very impressed by the wide variety of profiles and by the inspiring team spirit of the second-year students. In the third year, I did a six-month exchange with the University of Belgrano in Argentina. I had my first job at Competeer, an e-learning company, which became a leader in the sector in Latin America.
Interview by Thomas Lestavel for the Alumni magazine of June 2024. To read the whole interview see all editions HERE.