At Meta, AI is the new moderator of Facebook and Instagram

Meta has embarked on a deep transformation of its content moderation system. Early 2025, the company was ending its fact-checking program to better align with the alternative reality of the new leader of the White House." While implementing a system of « community notes » to separate the wheat from the chaff, like on Twitter.

AI is policing

In parallel, Meta was testing « advanced AI systems » for content moderation. Systems that « can detect more violations with greater precision, while blocking more scams and responding more quickly to real-world events, with fewer over-moderation errors », the company claims.

In a haphazard manner, this AI has allowed detecting « about 5,000 additional scam attempts each day », identifying « twice as much illicit content (…) with fewer errors », and improving the detection of account hacking and fake websites. In addition, it covers « 98% of online languages », while adapting to local jargon and cultural contexts.

Clearly endowed with all virtues, these systems will expand to automate more content moderation while reducing reliance on external contractors. A way to make serious cost savings… Meta still insists on maintaining human supervision for sensitive decisions, which will remain made by experts. These experts will also be responsible for designing, training, supervising, and measuring the performance of AI systems.

AI will take on common tasks, « such as the repetitive examination of sensitive content or areas where malicious actors constantly change their methods, such as the sale of illegal drugs or scams. » Meta is moving forward on a tightrope by thus pushing AI to the heart of its moderation infrastructure, at the risk of quickly revealing its limitations.

👉🏻 Follow real-time tech news: add 01net to your Google sources, and subscribe to our WhatsApp channel.

Source : Meta