Image: Jens Buttner

Meta puts AI plans in Europe on hold

GDPR concerns over training data access stall rollout
Image: Jens Buttner

3 July 2024

Facebook and Instagram parent company Meta has postponed the launch of its AI software in Europe for the time being, but the social networking giant is still of the opinion that its approach can comply with the General Data Protection Regulation (GDPR). Data protection activists see things differently.

At issue is how the company wants to use posts on its social networks Facebook and Instagram to train its artificial intelligence. At issue is the familiar problem users face of having to opt out from a feature – or in this case handing over their data – instead of being explicitly asked for their consent.

The issue remains important for users because it also sheds new light on the issue of data protection.

What does Meta need the data for?

Similar to AI software such as ChatGPT Meta AI creates texts and images and can answer its users’ questions. To ensure that this works as well as possible, the AI is trained with data. Meta wanted to use user posts from Facebook and Instagram for this purpose.

For AI expert Gregor Schmalzried, this would be a competitive advantage over other AI providers due to the amount of usable data alone.

Another advantage that Meta would have with the posts from Facebook and Instagram is that they are local data. “Many of the large language models have a certain bias towards the American world, simply because most of the data used to train these models comes from there,” said Schmalzried.

One example of this is ChatGPT. “Sometimes you read a sentence where it feels as if it really has been translated word for word from English,” Schmalzried continued.

It’s a problem Meta has accepted and cited translation issues as one reasing for postponing the launch of Meta AI in Europe.

Meta AI is already available the USA, Canada, New Zealand and Australia. It is also integrated into other Meta applications there, such as WhatsApp and Facebook Messenger. Users can ask Meta AI questions, for example, which the AI will then answer. Unlike ChatGPT, the AI software also uses search engines such as Google and Bing.

Why things went differently in Europe is due to GDPR. The company may have anticipated that it could be more difficult here. According to Schmalzried, European users would have been the only ones given the option to object to the use of their posts. However, even the option to object was not sufficient for the EU, according to data protection experts. The Hamburg data protection authority stated: “In the case of genuinely private posts that were only shared in small groups, such use for a different purpose would only be possible with the appropriate consent.”

In other words, Meta would have needed the expressed consent of users instead of just giving them the opportunity to object. The German Federation of Consumer Organizations (vzbv) also criticised the fact that the objection was too complicated and that users had to justify it.

In the case of public posts, European data protection experts are still examining whether Meta also requires the active consent of users.

Meta, on the other hand, is convinced that it has a “legitimate interest” in training the AI models and is therefore allowed to use the data accordingly, putting it on the right side of GDPR.

Is Meta AI still coming to Europe?

As long as the data protection concerns have not been resolved, Meta AI is still unlikely to be launched in Europe. According to expert Schmalzried, this is not an isolated case: “Claude, for example, in my opinion the best chatbot in the world, was not available in Europe for a long time. The same with Google Gemini.”

The problem is that good AI has not yet been built in Europe, he says. For Schmalzried, this raises the question of whether international companies will actually continue to accept the regulations in Europe and adapt their AI software to them with delays. Or whether they will no longer offer their tools in Europe because it is no longer worth it. Schmalzried says: “I wouldn’t worry about this at the moment, but in the long term we should at least think about this scenario.”

Read More:

Back to Top ↑