Facebook Algorithms and Ranking
In this article, a former Facebook data scientist talks about their own experience in how Facebook designs and regulates their algorithms. Written around the time when whistleblower Frances Haugen came forward, there was the question about whether this time would be when Facebook was held accountable for its handling of the information that appears and is spread on its platform. The article references how in the early days of the internet, there was no way for fringe content to reach wider audiences. The only way to access content was through a user intentionally searching with specific keywords or browsing a specific website. However, social media now boasts personalization and algorithmic amplification that rely on web cookies, big data systems, and powerful AI that chooses the content that is shown.
The problem with Facebook’s ranking algorithm is that it is engagement based. This is typically marketed to be a way that the platform promotes “relevant content” but ranking this way often perpetuates bias and promotes the most inflammatory content due to how and how much users respond. Having an algorithm that works this way can affect society in deeper ways that these algorithms are not able to detect. Currently, this is a law called Section 230 that allows social media companies that host user-generated content to not fear lawsuits from illegal content posted by users. There have been calls to change it so that companies face consequences for their intentional ranking decisions. Due to the increased potential liability along with the fact that no AI is powerful enough to keep up with every possible instance of illegal content, companies will probably want to get rid of algorithmic feeds entirely which would free social media feeds from unavoidable consequences.
This relates to how we’ve been learning about PageRank in class. I’d specifically also connect it to the concept of authorities and hubs and how different nodes have different voting power based on how often they are referenced to as well as how many times they themselves reference and whether they “vote” for nodes that many other people vote for. We see that by ranking information as more important based on how inflammatory it is versus the content’s actual validity, there are questions about the accuracy of ranking as a mechanism to spread the correct information. There’s also the question of the accuracy of the weight of certain node’s/individual’s votes and how they are able to spread information to a wide number of people. Is there a way to check the accuracy of the flow of information and the ranking? Can we gauge that those with a lot of authority are putting their vote towards accurate information?
https://www.nytimes.com/2021/10/06/opinion/facebook-whistleblower-section-230.html