Meta’s New AI Training Initiative in the EU

Meta’s Plan for AI Training

Meta is gearing up to train its artificial intelligence models using data from users of its applications, such as Facebook and Instagram, in the European Union (EU). The tech giant has announced that users will be informed about this AI training process and will have the option to opt-out if they prefer not to have their data utilized for these purposes.

Data Usage and Limitations

According to Meta’s announcement, the training data will consist of users’ public posts, comments, and interactions with Meta AI. Importantly, private messages exchanged with family and friends will not be included in this data collection. Additionally, only data from users aged 18 and above will be considered for AI model training.

Notification and User Consent

Starting this week, users will receive notifications via in-app alerts and emails. These notifications will contain a link to a refusal form that users can complete if they choose to opt-out of data processing. Meta’s current privacy policy indicates that the company is proceeding cautiously, taking into account feedback from EU regulators.

Regulatory History and Goals

Last year, Meta paused its AI training plans in Europe following a request from Ireland’s data protection authority. The company aims to use data from EU users to develop AI models that are more localized and sensitive to cultural nuances. Factors such as dialects, local idioms, regional humor, and typical speech patterns are critical for creating sophisticated multimodal AI systems. Meta believes that this data will enhance the accuracy and user-centricity of AI-generated text, audio, video, and images.

Comparative Context

This initiative follows Meta’s decision last year to start AI training with data from UK users. It’s worth noting that EU and UK users benefit from more stringent legal protections regarding personal data processing compared to US users. Yet, the volume of data Meta plans to acquire in the near term is significantly less than what it has historically amassed. Last year, Meta acknowledged that it had been using all publicly available text and images from adult Facebook users for AI training since 2007.