If you’re a Twitter user in the US, you may have seen little notes attached to popular or notable tweets. These notes are from company notes Birdwatch programa new approach to fact-checking that provides more context for tweets – all done by users.
The first question that comes to mind knowing that this system is peer-reviewed is who decides (and how) that a context note is correct and should be attached to the tweet. Wouldn’t such a system simply reinforce the division often seen online? After all, social media platforms are plagued by misinformation and thrive on divisive, polarizing and extreme content. Whistleblowers from Facebook and many other sites have highlighted how this is how these places build loyalty.
Birdwatch does something different. It uses a system called bridging ranking. This method is generally concerned with the estimation of positive interactions between different audiences. In the specific case of box notes, the ones attached to tweets are the ones that people from different walks of life and with different opinions all find useful.
“In many online spaces, especially those that use engagement rankings, divisive content may be more likely to go viral. Bridging-based ranking systems aim to overcome this ‘split bias’. “Birdwatch’s use of bridging to upgrade the framework found useful by people who tend to disagree is an exciting step towards a better internet – one that supports those building common ground,” said Aviv Ovadya, of the Belfer Center at the Harvard Kennedy School. Blog post on Twitter.
Ovadya explored this approach in a detailed report. It’s important to know that it’s not just about showcasing opposing views (which can exacerbate the polarization issue), but instead about rewarding and ranking higher content that is less divisive. For Birdwatch, much of the content has been viewed as informative regardless of political affiliation.
Birdwatch is now available to anyone in the US and users can volunteer to participate. The program has been piloting for over a year, and the social media company has conducted several surveys and data analyzes to see its impact. For misleading tweets, there was an average 20 to 40 percent reduction in agreement with them when the context note was attached. Also, once a note is attached, users are on average between 15 and 35 percent less likely to choose to Like or Retweet a Tweet than someone who sees the Tweet alone.
Twitter seems to be going all-in with transparency for Birdwatch. The algorithm powering this approach is publicly available on GitHub.