«As a European, I don’t want my children to use artificial intelligence trained on data from California or Iowa, but on data relevant to my culture». Nick Clegg, president of Meta’s Global Affairs, starts from this assumption by explaining the giant’s strategy on AI.
British, former deputy prime minister of the United Kingdom (from 2010 to 2015) and still a supporter of the Liberal Democrats with constant donations in view of the 4 July vote, Clegg is among the most powerful executives of Meta, after Mark Zuckerberg, since Sheryl’s exit Sandberg, who was considered the company’s number two and whose role she inherited. And he is – in his words – “one of the highest ranking Europeans in Silicon Valley”.
L’Europe is at the center of his speech for two reasons: on the one hand, Meta dedicates a new acceleration program to European startups that gives them access to open source models and other resources in the field of AI. On the other hand, it recently decided to suspend the launch of its generative artificial intelligence Meta AI in Europe at the request of the Irish privacy protection authority, which acts on behalf of its counterparts in the Old Continent.
What is the goal of the AI Startup Program?
«A bit of context: Europe has a huge problem. All the big AI companies are Chinese and American. Almost all of the top hundred companies in Europe were founded over 40 years ago: we are marketing on the past, not the future. What we will do, at the Station F incubator in Paris, is make our technologies available. In Italy, for example, the Turin-based Aitem already uses our Llama3 linguistic model for the software that helps veterinarians make diagnoses.”
What return are you expecting?
«Nothing direct. Of course, we don’t do it just out of altruism: we want more and more people to use Llama, experiment and innovate. How they manage activities and products will then depend on them: it’s like giving wood, for free, to a carpenter. I think it’s a way to create a real partnership between Silicon Valley and European entrepreneurs.”
Europe also imposes caution on you: for now you cannot launch your AI and train it with data published on Facebook and Instagram.
«We spoke with regulators in February, March and April. We’ve shared our plans, fulfilled requests, and sent more than two billion notifications, giving people the ability to opt out of their data from training (the correct term is opt-out: to object you must fill out a form, otherwise the data will be used, ed ). Data that is already public (texts, photos and videos. Private exchanges and profiles are excluded, ed ). Our competitors, such as OpenAI and Google, are already training their models on European public data (but they don’t have social platforms, ed ). And it is important to do so, otherwise Europeans will use American AI and not culturally, linguistically and geographically specific ones for Europe.”
Will you introduce the request for explicit consent?
“We will see what the regulators ask of us and we will act pragmatically.”
Have you already used social media data to train AI?
«Yes, but it was only public data and granted under a regular license».
How to deal with copyrighted content?
«We must start from the distinction between private and public: once we have crossed the threshold in which you have made your data public, be it an article or a post on social media, we must choose as a society. And Europe must make a choice as a continent, taking into account the fact that the more data you make available, the more results you get in terms of productivity.”
The point is also how the value of the contents is recognized.
«I think that for publishers, in the long term, it would be a disadvantage to exclude content from AI training: it would not be digested by the systems and would not be reflected in the results that people get. There will be an increasingly greater need for journalism and news care, I don’t think the best and most productive path is that of a legal battle.”
