Facebook and Instagram’s parent company, Meta, has disclosed that it has used public images and postings from these social media sites to train its new artificial intelligence (AI) assistant. Nick Clegg, the president of international relations at Meta, revealed it in a Reuters interview.
The AI being developed by Meta involves training it on a combination of text and images obtained from public posts. Notably, private posts accessible only to an individual’s friends and family have been excluded from this training process, according to Clegg. Additionally, private messages within Meta’s platforms are not being utilized to train the new AI.
Clegg emphasized that Meta has taken precautions to ensure that private details and personal information are excluded from the datasets used to train the AI. However, specific measures to achieve this were not outlined during the interview.
Meta recently launched its virtual assistant, Meta AI, to the public in a Beta release. This assistant is designed to assist users in creating digital stickers based on text prompts, editing photos using text instructions, and engaging in conversations with AI personalities like rapper Snoop Dogg playing the role of a “Dungeons & Dragons” Dungeon Master.
The AI models used to build Meta AI, Llama 2 (an AI language model) and Emu (a text-to-image model), were trained using content from Instagram and Facebook posts.
But there have been questions raised about using protected content to train AI. Regarding the fair use of copyrighted content in AI training, Clegg acknowledged the likelihood of legal challenges. Users of Facebook and Instagram maintain ownership of the content they upload, as long as it complies with Meta’s policies and does not violate the intellectual property rights of others.
As the use of copyrighted content for AI training comes under increasing criticism and legal scrutiny, Meta’s actions and their repercussions are expected to be the topic of ongoing arguments and potential legal action.