(Washington) Meta (Facebook, Instagram) must imperatively, and quickly, update its rules on manipulated content, with or without artificial intelligence, as elections approach in many countries, the supervisory board said on Monday. placed by the Californian group.
The board made the binding recommendation as part of a decision on a crude video montage, featuring US President Joe Biden.
The social media giant must reformulate its regulations, and ensure that modified images and sounds are clearly identifiable as such by users, says the council that the company created to have an independent look at content moderation , after several scandals.
Meta must “begin labeling manipulated content, such as videos modified by artificial intelligence (AI) or other means, when such content is likely to cause harm,” the council orders in a statement.
Many major elections will take place around the world in 2024, notably the American presidential election, and the authorities fear a surge of “deepfakes”, more or more sophisticated visual and/or sound montages using AI.
The video of Joe Biden – showing the head of state touching the chest of his adult granddaughter, with a caption that describes him as a pedophile – was widely shared online last year.
The board confirmed that it did not violate Meta’s current rules, but called for them to be updated.
“As it stands, this policy makes little sense,” council co-chair Michael McConnell was quoted as saying in the statement.
“It bans edited videos that show people saying things they don’t say, but doesn’t ban posts showing a person doing something they didn’t do. It only applies to videos created using artificial intelligence, but lets other falsified content get away with it,” he summarizes.
“Perhaps most concerning,” he adds, “is that the regulation does not cover audio falsifications, which are one of the most powerful forms of election disinformation we see in the world. whole world. Meta must work to address these gaps as a matter of urgency.”
The board therefore orders Meta to attach “labels” to all content “identified as ‘manipulated'”, regardless of the context in which it is published.
He points out that “not all misleading videos or sounds are objectionable in themselves, in the absence of a direct risk of harm in real life, particularly when it involves humor.
Meta plans to respond within 60 days maximum.