X is experimenting with a program enabling AI chatbots to generate Community Notes.
The social platform X will begin testing a feature allowing AI chatbots to generate Community Notes.
Under Elon Musk’s leadership, the Community Notes initiative from the Twitter era has been expanded on X. This fact-checking endeavor invites users to add contextual comments to specific posts, which are then verified by other users before being linked to the original content. For example, a Community Note could clarify an AI-generated video that doesn’t provide transparency about its artificial nature or offer a correction to a misleading statement made by a politician.
Notes become public once a consensus is achieved among groups that have previously disagreed on ratings.
The success of Community Notes on X has led Meta, TikTok, and YouTube to explore similar initiatives, with Meta completely phasing out its third-party fact-checking programs in favor of this low-cost, community-oriented solution.
Nevertheless, the effectiveness of AI chatbots as fact-checkers is still questionable.
AI-generated notes can be created using X’s Grok or other AI tools that connect through an API. Any notes submitted by AI will follow the same verification process as those from human users to ensure precision.
Utilizing AI for fact-checking raises concerns, especially since AIs can often hallucinate or produce misleading context without grounding in reality.
A recent study involving researchers from X Community Notes suggests that collaboration between humans and LLMs produces optimal results. Human contributions can refine AI note generation through reinforcement learning, and human reviewers can provide final validation before notes are published.
“The goal is not to create an AI assistant that dictates thoughts to users, but rather to foster an environment that promotes critical thinking and enhances understanding of the world,” the paper states. “LLMs and humans can engage in a productive feedback loop.”
Even with human oversight, there’s a danger of over-relying on AI, particularly since users can employ various third-party LLMs. For instance, OpenAI’s ChatGPT has recently come under scrutiny for generating excessively flattering comments. If an LLM prioritizes “helpfulness” over accurately completing a fact-check, the AI-generated comments could lead to misinformation.
Moreover, there is concern that human reviewers might feel overwhelmed by the influx of AI-generated comments, which could affect their motivation to participate effectively in this volunteer role.
For the time being, users shouldn’t expect to see AI-generated Community Notes — X plans to trial these AI contributions for a few weeks before deciding on a broader rollout, depending on their success.
