As moderation: X wants AI to write community notes

Soon, X will have “AI Note Writer”, i.e., AI-written notes on posts. Community notes are used on the platform to correct or categorize posts. They are placed under a post if the corresponding writers think a note is needed. This is intended to bring about a kind of self-correction on X and replace moderation. However, the AI notices are not displayed automatically. Humans operate and control them. A Community Notes member is responsible for an AI writer.

Community Notes are a more complex system than they initially sound. Not just anyone can write them, and publishing them requires the consent of other writers – They have to mark the content of the notes as helpful often enough, only then do they become visible. This will also be the case for the AI-generated notes. In other words, these are suggestions from an AI to add notes to posts. The AI notes will also be clearly marked as such.

AI and humans complement each other in moderation

Initially, according to an X post on the Community Notes account, the AI Note Writer will not simply be able to start everywhere, but will only be able to comment on posts where these comments are requested. If the initial tests are successful, this should change.

X hopes that this will result in faster and more comprehensive community notes. The system is supposed to improve itself, so to speak, by using feedback from the community to improve the AI agents and better. To confirm this, X has published a study available on arxiv together with MIT and the University of Washington, in which scientists investigated the extent to which AI can support humans in the evaluation of community notes. They call it Reinforcement Learning from Community Feedback (RLCF) – Humans get feedback from AI – vice versa.

Meta has also announced community notes based on the X model for its platforms. Initial field tests are already underway. Mark Zuckerberg hopes that the community notes will be less biased than fact-checkers. He has terminated contracts with these, at least in the USA. At the same time, the moderation rules on Facebook, Instagram, and threads have been significantly softened. Here, too, the community is supposed to be self-regulating and less dictated from above.

(emw)