Meta Gains EU Approval for Training AI Using Social Media Data

Meta's AI Training Plans with European User Data

Meta Gains EU Approval for Training AI Using Social Media Data

Meta intends to leverage publicly shared content from adults within the European Union to fine-tune its AI models. This content includes:

• Adult users' posts and comments on platforms like Facebook, Instagram, WhatsApp, and Messenger.
• Interactions with Meta's AI assistant, such as questions and queries submitted via chat systems.
• Chats with Meta's AI chatbots.

The objective is to enhance the AI's ability to comprehend European cultures, dialects, colloquialisms, and local knowledge, thereby making it more efficient for European users [1][2][3].

Exclusions

Meta will not use data from users under 18, private messages, or data from users who have chosen to opt out [2][3].

Opting Out Made Simple

Users in the EU can elect to opt out of their public data being utilized for AI training. Here's a step-by-step guide:

1. Notification: Upon receiving a notification through their apps and emails, users can learn how their data will be utilized.
2. Opt-Out Procedure: The notifications will contain a link to an objection form that users can complete to impede the collection of their public data for AI training [3][5].
3. Recognition of Pre-existing Objections: Meta has declared that it will honor both existing and new objections submitted by users [3].

The move was preceded by Meta's AI assistant's launch in the European Union last month across multiple platforms [6].

Reasons Behind the Move

Meta asserts that diverse training data is crucial for developing AI that caters to European users effectively. The company aims to ensure its AI understands regional dialects, colloquialisms, local knowledge, and cultural nuances, such as handling humor and sarcasm [4].

Meta points out that its advanced AI models will need to process text, voice, video, and images, making it essential to equip them with a wide range of data [4].

Regulatory Journey

Previous to this, Meta faced opposition from privacy advocacy groups in July 2021, leading to a temporary halt in its AI training plans [5]. The Irish Data Protection Commission requested Meta to postpone its rollout until a review was completed.

The complaints alleged that Meta's updates to its privacy policy would allow the company to utilize years of personal posts, private images, and online tracking data to train its AI products [5].

Meta has now received confirmation from the European Data Protection Board that its approach aligns with legal requirements [5]. The company claims it has actively engaged with the Irish Data Protection Commission throughout the process [5].

Industry Parallels

Meta follows in the footsteps of companies such as Google and OpenAI, who have already employed user data from Europe to train their AI models [6]. Meta notes that it has been training its generative AI models for other regions using similar methods since inception [6].

The approval comes amidst the EU's implementation of the AI Act, a legal framework for artificial intelligence that encompasses provisions for data quality, security, and privacy [6].

Other tech companies, like Twitter (X), have also faced scrutiny in Europe and have agreed to cease using personal data from EU users [5]. Google is under investigation by Irish regulators regarding its AI development practices [5].

[1] Meta to use European user data for AI training: https://news.ycombinator.com/item?id=27987745

[2] Meta's AI Training plans and exclusions: https://www.theverge.com/2022/4/14/23026118/meta-facebook-eu-ai-personally-identifiable-data

[3] Meta's AI training plans and user control: https://edri.org/facebook-decides-to-start-large-scale-eu-user-profiling-for-ai-models/

[4] Meta's rationale for using European user data: https://www.reuters.com/technology/us-meta-gets-eu-approval-use-user-data-train-ai-models-2022-04-15/

[5] Meta's regulatory journey: https://www.ft.com/content/fb540dab-38ec-4e4e-bbf3-bc6942f1d2e9

[6] Meta's AI assistant in the European Union: https://techcrunch.com/2022/03/24/meta-launch-ai-assistant-eu/

1. Meta's recent initiative involves using adult users' posts, comments, interactions, and chats with AI assistants and chatbots from within the European Union to fine-tune its AI models, aiming to improvethe understanding of European cultures and dialects [1].
2. Meta will not use data from users under 18, private messages, or data from users who have chosen to opt out, ensuring privacy and adherence to user preferences [2].
3. Users in the EU can opt out of their public data being utilized for AI training by following a step-by-step procedure provided through notifications within their apps and emails [3].
4. Meta's AI training plans were initially met with opposition from privacy advocacy groups, but the company has now received confirmation from the European Data Protection Board that its approach aligns with legal requirements, allowing them to proceed [5].
5. Companies like Google and OpenAI have already employed user data from Europe to train their AI models, and Meta's approach aligns with the EU's implementation of the AI Act, a legal framework for artificial intelligence that encompasses provisions for data quality, security, and privacy [6].

Read also: