Developing AI Technology for Europeans with Transparency and Responsibility
Update on AI Development at Meta
Latest Development as of June 14, 2024, 7:30 AM PT:
Recently, we received an unfortunate request from the Irish Data Protection Commission (DPC), which is our main regulatory body in Europe, to postpone the training of our large language models (LLMs) using public content shared by adults on Facebook and Instagram. This is concerning because we have incorporated feedback from regulators, and the European Data Protection Authorities (DPAs) have been aware of our plans since March. We believe this decision is a setback for innovation in Europe, especially concerning AI development, and it delays the benefits that AI can provide to the European people.
We remain confident that our training methods are compliant with European laws and regulations. It’s worth noting that training AI is a common practice among many tech companies, and we pride ourselves on being more transparent than some of our competitors.
Our goal is to introduce Meta AI and its supporting models to people across the globe, including Europe. However, without incorporating local data, we would risk offering a subpar experience. Unfortunately, this means we are unable to launch Meta AI in Europe at this time.
We are dedicated to collaborating closely with the DPC so that individuals in Europe can experience the same level of AI innovation as individuals in other regions.
This delay will also give us the opportunity to resolve specific requests made by the UK Information Commissioner’s Office (ICO) prior to beginning the training phase.
Original Announcement on AI Expansion
Originally Posted on June 10, 2024, 6:00 AM PT:
For several years, we have been focused on developing the next generation of AI functionalities throughout our family of applications and devices. This year, we aim to expand AI capabilities at Meta, including our generative AI features, to individuals in Europe.
Currently, AI at Meta is operational in numerous parts of the world. Our cutting-edge open-source large language model, called Llama, along with the Meta AI assistant—arguably the most advanced free AI assistant available—are already available for use. To effectively serve our European audience, our AI models need to be trained on relevant data reflecting the unique languages, cultures, and other regional characteristics of European users.
If we do not train our models using public content shared by Europeans—like public posts or comments—our AI features will struggle to accurately comprehend significant regional languages, cultures, or trends emerging on social media. We believe that without this localized training, Europeans may be underserved by AI that overlooks their rich cultural background.
We are following the footsteps of other companies like Google and OpenAI, both of which have already utilized data from European users for AI training. Our methods are more transparent and provide better user control than many similar companies training their models with publicly available information.
Commitment to Responsible and Transparent AI Development
When developing AI technology, we recognize the responsibility to establish best practices and policies that align with local laws. We consistently consult with our main privacy supervisor in the EU, the Irish Data Protection Commission, and have applied their feedback to ensure our AI training complies with EU privacy regulations. We are also collaborating with experts, such as academics and consumer advocates, to develop best practices.
Our aim is to maintain transparency so users are aware of their rights and the options available to them. Since May 22, we have issued over two billion in-app notifications and emails to individuals in Europe, explaining our plans and offering an objection form for those who wish to refuse the use of their data for AI model training.
We have simplified our objection form to ensure it is easy to access and complete. Designed for better comprehension, it is user-friendly and more accessible to individuals with lower reading abilities, despite the fact that we do not use content from accounts belonging to users under 18 for training our Llama models.
We respect all objections from European users. If an individual submits an objection form before the training of Llama begins, their data will not be used in this or any future training sessions.
Our primary objective is to create useful AI features based on publicly shared content from users over 18 in Europe. This includes public posts, comments, or photos shared on Meta’s platforms. It is important to note that our models do not compile personal data about individuals nor are they designed to identify any specific person. Instead, they analyze shared information to uncover patterns, such as understanding local phrases or references.
We do not utilize private messages between friends and family for AI training. In the future, we may include other types of content, such as user interactions with AI features and chats with businesses via Meta AI.
To ensure we comply with the EU’s General Data Protection Regulation (GDPR), we plan to rely on a legal basis of “Legitimate Interests” for processing public data on a large scale for AI training while respecting users’ rights.
We are committed to developing AI that is not just imposed on Europeans but rather created for them. We believe it’s essential to inform users of our plans and give them the choice to opt out if they prefer. The best way to achieve this balance is through transparency about the information utilized by our AIs and providing users with options to control their data. This approach is precisely what we aim to deliver.
The Future of AI in Europe
As Europe stands on the brink of significant technological advancement, some voices advocate for stricter data regulations regarding AI. It’s important to clarify that these extreme positions do not align with established European law and would ultimately hinder access to AI that other regions enjoy. We oppose this outcome and believe it is crucial for Europeans to be able to share in the technological advancements that AI offers.
Europe, as a leading global region, has the potential to spearhead AI innovation. However, important questions remain: Will European populations have fair access to groundbreaking AI technology? Will the experiences provided by AI reflect our diverse culture and history? Is Europe willing to watch as other parts of the world benefit from innovative technologies that foster community and spur growth?
AI represents the next major frontier in innovation. We are witnessing an exhilarating moment in technology where breakthroughs are rapidly occurring, and the potential is limitless. At Meta, we are eager for Europeans to be actively involved in this journey.
