Meta will start training its AI models using the social media posts of all its users, but users in the European Union will be able to opt out, a luxury the rest of the world does not enjoy.
The move, announced by Facebook's parent company today, is ostensibly aimed at bringing machine learning systems to Europe.
Meta previously didn't include its European user base in its AI training data, presumably to avoid legal conflicts with the continent's privacy regulations, and now, despite the complaints, the company is moving forward with its plans.
“To properly serve European communities, the models that power Meta's AI need to be trained on relevant information that reflects the diverse linguistic, geographical and cultural backgrounds of the Europeans who use it,” the social media giant said.
“To achieve this, we want to use the content that EU people choose to make public in Meta products and services to train large-scale language models that power our AI capabilities.”
Training an AI from user data is sure to be controversial in Europe, so Meta is trying to cover itself in two ways: First, by “public content,” Meta refers to posts, comments, photos, and other content posted on social media platforms by users over the age of 18. Private messages appear to be strictly banned from training data.
Meta also said it has sent billions of notifications since May 22 to give European users the opportunity to opt out before the AI training rules come into force worldwide on June 26. The Instagram giant said any user can opt out, no questions asked, and that their posts will not now or ever be used to train AI models.
This is quite different from other parts of the world, where opting out is not an option. Indeed, it is already too late to opt out of having your training data used for Meta's LLaMa 3, but even training future models is mandatory for Facebook and Instagram users outside the EU. Presumably non-European users will be able to choose to opt out in the future, but for now it is an EU-only feature.
Meta seems to think it's in a good position to start using European user data, but it's unlikely to meet any resistance: The social media giant signaled its intentions with an update to its privacy policy last week before making a public announcement, sparking complaints across Europe from consumer privacy advocacy group noyb.
Noyb argues that collection of user data should be opt-in by default, not opt-out. The fact that data cannot actually be deleted from LLM and other AI models could also raise issues due to the European Union's right to be forgotten.
Moreover, Meta does not have a good relationship with the EU. Earlier this year, the EU opened an investigation into Meta, including one into child safety and another into misinformation surrounding the now-ended EU parliamentary elections. It's unclear whether Meta will ultimately get its way, but it's hard to imagine that challenges won't be filed against the social network at some point.®
