Fake news, ranking algorithms and information cascades
https://www.ft.com/content/6973e6d6-d047-11e7-9dbb-291a884dd8c6
This article is about fake news circulating on social media websites. Specifically, the article refers to criticisms of Facebook, Google and Twitter in the wake of the October Las Vegas shooting for circulating misinformation, such as the idea that the shooter was very left-wing or connected to ISIS (both unsubstantiated at the time and proven false later). The article commented that Facebook in particular was already under heavy fire due to reports of its abuse in election meddling by Russian state operatives, and added the opinion of an NYU professor that they hadn’t seriously improved their countermeasures. What Facebook did say was that the reason this happened was that someone made a post, which took a few minutes for the system to take down, and in that time people took screenshots of the post and circulated it themselves making it very difficult to actually stop it from spreading. Google news searches also came up with bad stories about the man who had been accused (falsely) of perpetrating the shooting but was in fact innocent, until it was replaced by accurate sources after a while.
In these statements we can see some flaws of the ranking algorithm covered in class, especially in Google’s. While it’s fine to compute n iterations of such an algorithm, it assumes a self-symmetric network, so that looking at it at early times will be the same as late times. When we start introducing new nodes into the system, there is no guarantee that the page rank algorithm will behave the way we want it to – it will give good information taking all data into account but that means previous results may have been bad, as was the case here. When perfect credibility is required of these websites, we need some a priori method of assessing the reliability of a source instead of relying on algorithmic measures which take time. In Facebook’s case there is an element of information cascading. We see that the implied veracity of the story (because of the reposts) is at odds with reality since these people all operated under the initial assumption that the story was true, which exposed them to only part of the information. They then propagated this and caused a chain reaction which we call an information cascade. So we see that standard network effects like information cascades are prominent in the spread of fake news online, and further that its spread is facilitated by network algorithms rather than direct evaluations of content (which we see is what Facebook tried to do to combat the problem). One possible solution to this issue might be instituting a time-delay on the sharing and screenshotting of controversial posts (or those from unreliable sources) to allow enough time for a person to manually check them, which helps prevents an information cascade from starting if the news is confirmed fake when someone would want to repost it. People are of course impulsive and could act before seeing this, but I think this would be a good solution to the network problem because you eliminate the assumption of self-symmetry.