X (formerly Twitter) has started testing a new functionality by which X Pilots AI chatbots also known as AI Note Writers can compose context-enhancing Community Notes to posts of the users. The program, which was unveiled by product chief Keith Coleman, is in testing to allow scaling the fact-checking but with human control. The initial batch of the AI bots will become operational at the end of the month, in a limited test scenario.
Another feature Twitter shares with Community Notes allows contributors to add notes to posts, clarifications, corrections, or provide more context. These notes will only be present when readers of different opinion find them useful to get balanced opinions. The new pilot enables third-party developers to create AI agents that can automatically draft a note or draft a note when they have requests or misinformation enhancement. All AI-written notes should be reviewed, qualified, and judged as either useful or non-useful, by human writers before they get to be available publicly.
During the first stage of pilot running, the chosen AI Note Writers will run on a test basis and only on those posts where customers will want to see them. Experts view their entries by Community Ratings and only the notes of quality standards are published. Bots will place notes with clear labels as opposed to the ones written by human beings. Coleman highlighted the aspect that, on the one hand, AI will be able to speed up the process of production, but, on the other hand, the ultimate decision will always be made by a person to achieve a compromise between effectiveness and correctness.
This hybrid model will focus on using the scalability of AI and maintain the reliability of fact-checking. The instant notes will not only be vetted by human feedback but also the feedback will aid in reinforcement learning that will be able to improve the AI models with the passage of time. Hundreds of Community Notes are printed each day, which means, according to Coleman, that with the help of AIs, much more can be produced, especially on those posts that would otherwise remain unread.
Integrating chatbots in the fact-checking scenario is risky, however. It has been known that AI models tend to create something that appears believable that ends up being not true at all also known as hallucinating. In order to protect against this, AI-generated notes will be reviewed in terms of quality in the same way that human-generated notes will. Nonetheless, it is feared that the abundance of moderators can overload a volunteer and to diffuse the confidence in the system. X is relying on openness (marked clearly) and human duty to maintain the credibility.
The decision follows a larger shift toward lighter-weight moderation based on community policing. Community Notes have been imitated by Meta, TikTok, and YouTube having abandoned standard third-party fact-checking programs. X locates the position of its pilot as an intermediate step-increasing scale but attaching reliability to community consensus.
In collaboration with an official pilot, X released a research paper which was co-authored by his MIT, Stanford, Harvard, and University of Washington colleagues. The article describes a scenario of a virtuous cycle in which both AI and people cooperate: backing up the AI with a possibility to suggest notes to review, the community ranks those notes to evaluate and approve of them, and the feedback is used to enhance AI models. The authors of the study established that hybrid AI and human judgment can be used to find deeper explanations of the context than employing AI or human judgment solely.
On a broader perspective, success indicators will go beyond the pilot: volume of notes, user activity, and accuracy based on quality scores. X will too keep track of whether the labelling of AI labels influences the feelings of neutrality and trust in the minds of readers. In the event that the pilot is successful, it will be expanded by X and make it possible to have other developers to develop Note Writers and potentially cover more posts, and more likely to be stormed by misinformation.For X, there is the transformative potential: AI Note Writers can be used to assist in real-time to combat misinformation, mark deep fake, and provide explanatory text to the fast-paced news. The use of generative AI should become more commonplace, and an extension of generative AI into the moderation tech is the scalable, community-grounded solution to the contesting of misinformation spread, the ultimate test to the beneficial direction of generative AI and its contribution to stronger democratic discourse.
