Liam Walsh
Liam is a Co-Director at Intelligency and heads up the agency's Digital Intelligence & Paid Social activity. Over the last decade, he has worked with brands from the world of sports such as Premier League clubs to entertainment such as Channel 4 and Disney.
Meta’s Use of Public User Data to Train AI Systems
Meta has this week admitted to using user data to train their AI software. This includes public posts, photos, and comments from both Facebook and Instagram are being leveraged to train its AI systems. While the company had previously revealed plans to utilize user content for AI training, it was not known that this practice had been ongoing for nearly two decades, since 2007. Mark Zuckerberg, the company’s CEO, introduced the concept of scraping posts as a new direction and initiative for the social media giant rather than something that had already been happening.
This revelation raises serious privacy concerns, especially as Meta claims to have access to more user data than systems like ChatGPT. The scope of data collection includes text, images, and interactions from millions of users across the platform, making the company’s actions a topic of global scrutiny.
Privacy and Legal Concerns in Data Scraping Practices
Meta’s approach to data scraping has been controversial due to the lack of transparency for users and their lack of control over their data. Furthermore, Meta was legally required to offer opt-out options for users in the EU and UK, this has not been the case in the US, Australia, or other regions.
Advocates of privacy have claimed that this vast data project has been done without any explicit consent and consequently reflects poorly on Meta. It also creates an uncomfortable comparison with Apple which has committed to a more privacy-focused approach, further deepening concerns about Meta’s privacy practices.
Admission During Australian Inquiry into Meta’s Practices
During an Australian inquiry, Meta’s global privacy director, Melinda Claybaugh, initially denied that the company had been using public posts from as early as 2007 to train its AI systems. However, under further questioning, Claybaugh admitted that the practice was indeed taking place for all public posts, except for content from users under 18.
