There is no easy fix for Facebook’s reliability problem
Rating news sources is a complicated task. But rating Facebook is especially hard, because of the way it’s designed. Here’s a look at several critical issues the company is facing.
Issue #1: The limits of flagging and fact-checking
Facebook’s most comprehensive response to the misinformation plague was to announce the hiring of 10,000 moderators. Spectacular indeed. But let’s consider the following: Facebook deals with 100 million links in its newsfeed every day. Even if we assume that only 10 percent are worth checking, each moderator will have about 25 seconds to judge a feed’s accuracy or trustworthiness. And I’m not counting the 300,000 videos uploaded every day, or the 400 hours uploaded every minute to YouTube that find their way to Facebook.
Last year, Facebook also made a big fuss about using professional fact-checking outlets to clean up its newsfeed. Research (and common sense) show that fact-checking is insufficient to deal with the problem. In the research paper Crowdsourcing judgements of news source quality, published this month, Yale scholars Gordon Pennycook and David G. Rand, measure fact-checking at its fair value (emphasis mine):
Despite the apparent promise of relying on professional fact-checking, however, there are reasons to believe that this approach is not an ideal solution for fighting misinformation on social media. First, fact-checking necessarily takes substantial time and effort, whereas false content can be produced easily and quickly. As a result, there is no way that professional fact-checkers will be able to keep up with the constant stream of new pieces of false content, and so most fake news stories will never get tagged. Furthermore, even for stories that do get fact-checked, there is necessarily a lag between when the story becomes popular online and when it gets flagged as false — for example, it typically took Facebook over three days to implement a “Disputed by 3rd party fact-checkers” warning.
Another study, The Effects of News Source Ratings on Fake News in Social Media by three scholars at the University of Indiana, leans in the same direction:
[F]act checking individual articles is simply too slow to be an effective solution against the spread of fake news. Disinformation and fake news spread through social media, “take root quickly and die hard”. By the time an article has been fact-checked, many users have already read the article and much of the damage from fake news has already been done. Hence, fact checking is often “too little, too late,” and the scientific community has been voicing the need to take a more proactive measure to prevent the spread and consumption of disinformation. […] News articles tend to have a short “shelf life,” and by the time fact checking has taken place, the articles would most likely have gone through their life cycle.
Exit, then, professional fact-checking of fake news. It might be commendable and comforting to journalists and publishers, but its real effectiveness is zero. In the context of social media, it’s like purifyng a glass of filthy water from the Ganges river while ignoring the rest of the stream.
Issue #2: The “Implied Truth” effect
Even worse, another effect comes into play: it is called “implied truth.” In short, flagging suspicious content suggests that the rest of the news stream is fine. Abstract from the Pennycook & Rand study:
[F]ailing to tag fake stories may actually make outcomes worse in some cases due a potential “implied truth effect”: evidence suggests that attaching warnings to some stories can increase belief in stories that are untagged because some people take the absence of a warning to imply that the story in question has been verified.
Issue#3: The excessive reliance on popularity
Another serious flaw in Facebook’s approach is the pre-eminence of popularity (expressed by clicks and “Likes”), which is at the core of the newsfeed system. This is actually the best way to spread misinformation. According to Pew Research (PDF here), 23% of social media users report that they have spread fake news, 16% by accident and 14% intentionally (some do both). This propagation is also helped by Facebook’s diabolically clever ad machine remarkably exposed by a former insider in this Wired piece :
Rather than simply reward that ad position to the highest bidder, though, Facebook uses a complex model that considers both the dollar value of each bid as well as how good a piece of clickbait (or view-bait, or comment-bait) the corresponding ad is. If Facebook’s model thinks your ad is 10 times more likely to engage a user than another company’s ad, then your effective bid at auction is considered 10 times higher than a company willing to pay the same dollar amount. (…) The above auction analysis is even more true for News Feed, which is only based on engagement, with every user mired in a self-reinforcing loop of engagement, followed by optimized content, followed by more revealing engagement, then more content, ad infinitum.
Issue #4: The Bandwagon effect
This issue is somewhat connected to the previous one. Every e-commerce site uses this trick: at the top of the list of items searched, it will always show you the most popular choice, which is an implicit validation of the crowd. It works exactly the same, as suggested by this research The effects of source cues on online news perception (Pennsylvania State University, 2014):
The effects of the bandwagon cue on users’ perceptions of online news content have been widely supported by previous studies […]. Specifically, the bandwagon heuristic suggests a mental shortcut that favors collective sources over individual sources (i.e., ‘‘if others think that something is good, then I should, too’’). […] The bandwagon effect [is] a possible type of heuristic processing in which people are more likely to consider an opinion as credible if the opinion is supported collectively.
[B]andwagon effects have also been examined in online news consumption. In a study, participants were told that a given news article was selected by either news editors, computers, other users, or the participants themselves (participants could then select their own articles). Surprisingly, participants who believed that the given news article was selected by other users rated the article more highly in terms of news quality than those participants who believed the article had been selected by news editors or those who had selected the article themselves. These results are clearly consistent with bandwagon effects.
Issue #6: The hybrid composition of fake news.
We touched on this issue in a previous Monday Note “When fake news will be made by pros” : as misinformation gains in sophistication (which is likely, given the profuse funding supporting it), it will increasingly entwine valid and false information. We saw this last week with a conspiracy video claiming that a young survivor of the Parkland school shooting, was in fact an actor. For hours, it was featured in Trending Topics on You Tube and picked up by Facebook users. A story in Wired exposed YouTube’s pitiful defense:
The Parkland video topping YouTube’s trending page seems especially galling because it appears to have gotten there not by accident but as the result of an attempt on YouTube’s part to fix fake news. YouTube says its system “misclassified” the conspiracy video “because the video contained footage from an authoritative news source.” Whatever minimal nuance was needed to block the Hogg conspiracy, algorithms lack it.
As expected, rumors reverberated on Facebook:
For both YouTube (the world’s main provider of videos) and Facebook (the dominant vector of fake news), solving this problem would actually be easy: kill Trending Topics, which has a terrible track record. But neither tech giant will do that, because that’s where the advertising money is.
Issue #7: Facebook’s gated communities and “confirmation bias”
Here is the analysis in the The Effects of News Source Ratings on Fake News in Social Media (Indiana U.) paper:
Facebook tracks and learns users’ preferences by tracking what they read and the actions that they take (e.g., like, comment and share). As a commercial entity, Facebook aims to maximize user satisfaction, and thus, it deliberately displays more stories matching the users’ preferences, so the users see more stories that match their existing beliefs. Such a process causes a decrease in the range of information that the users encounter, and, as a result, Facebook users often exist in small information bubbles — often referred to as echo chambers — that reinforce their beliefs and make them believe that others around the world are more like them. Thus, many articles that users encounter on social media are related to topics that users have previously viewed, topics on which the user often has already formed an opinion.
This tendency to favor information that confirms one’s pre-existing opinions and ignore information that challenges them is called confirmation bias. Thus, people are more likely to believe information that matches their pre-existing opinions.
Facebook’s composition of thousands of affinity and cognitive bubbles, also considerably weakens the effectiveness of any rating system. Again, based on Yale research, here is how the familiarity of a source influences its believability:
[Caption from the study] : [The figure on the right] shows the results when excluding trust ratings where the participant indicated being unfamiliar with the website. As can be seen, this exclusion dramatically increases the trust ratings of hyper-partisan websites (trust is roughly 1.5 times higher when excluding unfamiliar) and fake websites (trust nearly doubles when excluding unfamiliar), but produced a smaller increase in trust ratings for mainstream media outlets (trust is roughly 1.15 times higher when excluding unfamiliar).
None if these issues will be resolved anytime soon. That is not because social networks (Facebook, YouTube, Twitter) don’t take them seriously, but it’s because these flaws are intertwined with their internal structures — and their business models.
