Who Decides What’s True Online?
Key Takeaways from Our Webinar on Democracy, Regulation, and Global Power: Recap of Aishik Saha’s Presentation
Last week, INACH hosted its Webinar “Who Sets the Rules Online? Democracy, Regulation and Global Power.” Aishik Saha, a PhD candidate at the National and Kapodistrian University of Athens, took the time to explain how X’s Community Notes system really works, and to ask how algorithms like those underlying community notes can be improved.
X writes that Community Notes “remains a leading mitigation for the risk of misinformation, relating to both public security and civic integrity.” While X argues that Community Notes “leverages the collective intelligence of the public” in order to reach more posts more quickly than traditional fact-checkers, Saha asks if this solution is entirely beneficial by looking at its publicly available algorithms.
Saha describes how Community Notes are ranked based on rater’s “helpfulness scores,” which are used to ensure a diversity of viewpoints are present in notes and the system isn’t dominated by a single point of view. First, raters’ must rank a number of existing notes and achieve a high enough “helpfulness score” to make sure you have made enough valid rating (i.e., your ratings align with their system). Users’ ratings of notes are compared to their past ratings and to other people’s ratings, so that they can be given a score which represents their alignment on a linear scale, from left to right. After this process, when users rate a note as helpful, this note is only shown under the post if raters who had disagreed on previous notes now agree on the note in question. This forces the note to reflect a perspective in the middle-ground, but Saha says it has flaws.
Firstly, the content of notes is not formally fact-checked, and even notes voted as helpful can still contain misinformation. Additionally, the nature of the notes algorithm limits how many notes can be seen overall. The requirement of raters that have previously disagreed on posts to now agree on the post in question prevents notes from becoming too widespread. Together, this makes Community Notes susceptible to the sheer volume of large scale disinformation attacks.
Saha says these vulnerabilities arise from flawed assumptions underlying the Community Notes algorithm, such as the assumption people who disagree with each other on a post’s veracity are also ideologically distinct, and therefore need to be balanced out in order to see “objective” notes. However, this isn’t always the case, as the middle ground is not always factually accurate, and a greater volume of voters saying so doesn’t make this the case. This leaves Community Notes’ algorithm susceptible to wrongly amplifying false information as added context, representing a systemic risk in its design. Saha closes by asking people to consider how assumptions underlying algorithms may reflect dominant political frameworks: in this case, X assumes users exist in a two-party context in which partisan groups disagree on what is true. When algorithms are used to spread information globally, even for seemingly good causes, it is crucial to ask why they select certain outcomes, and what implications this can have.
Written By: Fox Oliver
Curious about what the other presenters at the Webinar had to say? Scroll through our blog archives to read our recaps of Aidan O’Brien’s and Zenith Khan’s presentations.