LONDON — Meta wants to use data from users in privacy-conscious Europe to train its artificial intelligence models, the social media giant said Monday, citing data protection concerns as it struggles to keep up with rivals such as OpenAI and Google.
The company, which owns Facebook, Instagram and WhatsApp, said that to better represent the “languages, geography and cultural references” of its users in Europe, it must use public data from those users to build its Llama AI large language model to learn .
Meta's AI training efforts are hampered by strict European data privacy laws, which give people control over how their personal information is used. The Vienna-based group NOYB, led by activist Max Schrems, last week complained to 11 national privacy watchdogs about Meta's AI training plans, urging them to shut down the company before it starts training the next generation of Llama.
AI language models are trained on vast amounts of data that help them predict the most plausible next word in a sentence, with newer versions typically smarter and more capable than their predecessors. Meta's AI assistant feature has been baked in it Facebook, Instagram and WhatsApp for users in the US and 13 other countries, but especially not Europe.
“If we don't train our models on the public content that Europeans share on our services and others, such as public posts or comments, models and the AI features they power will not accurately understand important regional languages, cultures or trending topics. on social media,” said Stefano, Global Engagement Director of Meta's privacy policy blog post.
“We believe Europeans will be ill-served by AI models that are not based on Europe's rich cultural, social and historical contributions.”
Fratta said other companies, including Google and OpenAI, have already trained on European data. Meta will not use private messages to friends and family, nor content from European users who are under 18, he said.
Since May 22, the company has sent 2 billion notifications and emails to European users explaining its plans and linking to an online form to unsubscribe, Fratta said.
The latest version of Meta's privacy policy will go into effect on June 26, indicating that training for the next model will begin shortly thereafter.
