Meta AI Trained Using Public Facebook and Instagram Posts
Meta, in a recent statement, revealed its utilization of public Facebook and Instagram content to train its artificial intelligence systems. The company’s latest virtual assistant combines both publicly available text and images, according to a spokesperson cited by Reuters.
The process involved training Meta’s new AI assistant using a wide array of public posts, encompassing text and photos. This revelation was made by Nick Clegg, Meta’s President of Global Affairs, during an interview with Reuters.
It’s important to note that the AI system has been designed to respect user privacy. Private posts, shared exclusively with friends and family, remain untouched by the AI’s training data, as confirmed by Clegg in the Reuters interview. While specific details regarding privacy measures were not disclosed in the interview, Meta assured that efforts have been made to safeguard private information when using public datasets.
Additionally, Meta clarified that private messages exchanged on its platforms were not utilized to train the new AI model. Clegg emphasized the company’s commitment to excluding datasets heavily laden with personal information from the AI’s training data. As a result of privacy concerns, Meta refrained from using platforms like LinkedIn for AI training.
Meta introduced its virtual assistant, Meta AI, to the public in a Beta release on Wednesday. The primary functions of this assistant include helping users generate digital stickers based on text prompts, providing photo editing capabilities through text instructions, and engaging in conversations with AI personalities, such as rapper Snoop Dogg playing the role of a “Dungeons & Dragons” Dungeon Master.
Meta AI was developed using Meta’s AI language model, Llama 2, and a text-to-image model named Emu. Both of these models were trained using publicly available content from Instagram and Facebook, as disclosed by an anonymous Meta spokesperson.
Clegg anticipates potential legal challenges related to the use of copyrighted content for training AI. While Meta believes its approach falls under fair use, he acknowledged that litigation might be necessary to settle the matter.
The use of copyrighted content for AI training has raised concerns among authors, artists, and developers. Some have taken legal action against AI companies like Meta, fearing that their work is being employed without their consent to train technology that could potentially undermine their careers.
Meta’s policies stipulate that content posted by Facebook and Instagram users remains their property, provided it does not infringe on the intellectual property rights of others.
As of the time of this report, Meta’s press team had not responded to requests for comment outside regular business hours.
