Users may see labels on possibly artificial intelligence (AI)-generated content in the coming months, according to Meta.
Meta said on February 6th that it is developing universal technical standards for AI content identification, including audio & video, in collaboration with industry partners.
People are curious about the boundaries between human & synthetic content as they become more hazy. According to Nick Clegg, President of Global Affairs, “It's critical that we assist people in recognizing when photorealistic content they're viewing has been produced using AI.”
In addition to invisible watermarks and metadata encoded into image files, images created with Meta’s AI image generator bear the title “Imagined with AI.” However, according to Clegg, Meta plans to use AI labeling for all content made with Google, OpenAI, Adobe, Microsoft, Midjourney, and Shutterstock.
“In order to establish universal technological standards that identify content produced with artificial intelligence, we have been collaborating with industry partners. We will be able to tag AI-generated photos that people upload to Facebook, Instagram, and Threads after we identify these signals,” adds Clegg.
Clegg claims that although AI businesses are adding information to their picture generators, they haven't done the same for AI technologies that produce audio & video on the same scale. As a result, Meta is unable to identify this content as AI-generated.
But, the business is introducing a feature that would allow users to reveal when they share audio or video produced by Artificial Intelligence, allowing a label to be applied while more research is done. When there is a substantial chance that digitally altered photos, videos, or audio files would mislead the public.
It’s a more obvious label that will be used to give consumers more details & context. While acknowledging that not all AI-generated content can currently be identified, Clegg says the company is working on solutions that take advantage of current technological capabilities.
"We are putting a lot of effort into creating classifiers that will enable us to recognize AI-generated content automatically, even in the absence of invisible indicators. Simultaneously, we're searching for methods to make undetectable watermark removal or modification more challenging," he stated.