Meta’s Oversight board criticise Meta’s deepfake rule as incoherent

Meta’s deepfake rules: Meta's Oversight Board has declared that a Facebook video implying US President Joe Biden's involvement in inappropriate conduct does not violate the current company rules. However, the board criticised these rules as "incoherent" and overly focused on AI-generated content.

The case, initiated in October in response to a user complaint, addressed a manipulated seven-second video of President Biden on Meta's flagship social network. This ruling represents the Oversight Board's first commentary on Meta's "manipulated media" policy, designed to restrict certain types of doctored videos amid concerns about the potential misuse of AI technologies in influencing elections.

The board contends that the existing policy lacks clear justification, is confusing to users, and fails to explicitly specify the harms it aims to prevent. It suggests Meta update the rule to encompass both audio and video content, irrespective of AI use, and implement labels identifying manipulated content.

While stopping short of extending the policy to photographs, the board cautioned that doing so might pose challenges in enforcement at Meta's scale. Meta, encompassing Instagram and WhatsApp, informed the board of its intention to update the policy to address the evolution of AI technologies.

Meta is currently reviewing the ruling and plans to respond publicly within 60 days. The contested video on Facebook manipulated authentic footage of President Biden during the 2022 US midterm elections, depicting him exchanging "I Voted" stickers with his granddaughter. The Oversight Board affirmed Meta's decision to keep the video under its current policy, which focuses on misleadingly altered videos produced by AI or those making individuals appear to say words they never uttered.

Highlighting the prevalence of non-AI-altered content, the board emphasized that it can be just as misleading. It recommended extending the policy to audio-only content and videos showing actions individuals never actually performed. The suggested enforcement approach involves applying labels to content, differing from Meta's current practice of removing posts from its platforms.

(With Reuters inputs)