Meta, on Monday, declared its intention to train AI models using public material, including posts and comments from Facebook and Instagram, in the EU, having earlier suspended these plans owing to legislative pressures related to data privacy issues.
The business said it would begin training its AI on users’ material in the EU this week. Users’ engagements with Meta AI will furthermore be used to refine its models.
The news follows the launch of a restricted version of Meta AI in the EU last month, subsequent to its introduction in the U.S. and other international markets.
Meta has been training its AI on user-generated content in the U.S. for several years; however, it has encountered opposition in the EU due to the bloc’s stringent privacy regulations, notably the General Data Protection Regulation (GDPR), which mandates a definitive legal foundation for processing personal data to train AI models.
In June 2024, Meta announced a suspension of its intentions to start training of its AI systems using user data in the EU and U.K. subsequent to objections raised by the Irish Data Protection Commission (DPC).
The DPC oversees Meta inside the EU and was representing many data protection bodies within the region.
In September 2024, Meta announced the resumption of its initiatives to teach AI systems using public postings from the United Kingdom user demographic.
Currently, Meta has declared its intention to do this with public postings from its EU user base as well.
“Last year, we postponed the training of our large language models utilizing public content while regulators elucidated legal obligations,” said Meta in its blog post.
“We appreciate the opinion issued by the EDPB in December, which confirmed that our initial approach complied with our legal obligations.”
Since that time, we have worked productively with the IDPC and anticipate further delivering the comprehensive advantages of generative AI to individuals in Europe.
Beginning this week, customers in the EU will get in-app and email alerts informing them that Meta will start using public data and interactions with Meta AI to train its models.
These alerts will include a link to a form enabling consumers to decline the usage of their data. Meta asserts it will acknowledge any objection forms already received, in addition to that filed hereafter.
Meta asserts that it does not use private communications or public data from users under the age of 18 in the EU for training its algorithms.
“We assert our obligation to develop AI that is not only accessible to Europeans but specifically designed for them,” states Meta.
“This underscores the necessity for our generative AI models to be trained on a diverse array of data to comprehend the remarkable nuances and complexities inherent in European communities.” This encompasses accents, colloquialisms, hyper-local knowledge, and the unique manners in which other nations use humor and sarcasm about our goods.
Meta asserts that it is emulating the practices established by businesses such as Google and OpenAI, both of which have previously used data from European consumers to build their AI models.
Simultaneously, the DPC continues to examine the methodologies used by designers of Large Language Models in training their AI services.
Last Monday, the regulatory authority revealed it carried out an investigation into xAI’s training of Grok.
